The session was hosted by the Retail AI Council. They gathered senior practitioners who have run the pilots, defended the budgets, and lived with the results.

Here's what I came away thinking.

The ROI framework hasn't changed as much as you'd expect

The first thing that struck me was how consistent experienced retail leaders are in their evaluation of AI investments.

Despite all the changes in the technology itself, the best frameworks still separate hard returns (measurable revenue lift, cost reduction, trackable KPIs) from softer returns (operational efficiency, brand readiness, fraud protection) and longer-term value (how well an investment integrates with your technology roadmap and scales across the business).

What has changed is the discipline around applying that framework.

There's less patience now for vague pilots. Business owners want a clear north star metric before a project starts.

They want to agree upfront on who owns the measurement, what "good" looks like, and how you'll know if it isn't working. Investing time in that clarity at the beginning is the best way to get started.

Good data is still the foundation. And still underestimated

If there was one point of genuine consensus across the panel, it was this: the quality of your data determines the quality of your AI.

We've known this for years. It's still underestimated.

One panelist described building a demand forecast where a black t-shirt appeared in the system with "LG" in one SKU and "LARGE" in another. The algorithm had no way of knowing they were the same product. Years of investment, undermined by a data entry convention that nobody had standardized.

But what stayed with me was a different idea: solving data quality at the design level rather than the cleanup level. And it’s something we are focused on helping Scandit’s customers with through our intuitive interfaces and high-performance data capture.

Build your interfaces so that bad data can't get in.

For example, a dropdown that forces users to select "Large" rather than typing it free-form eliminates that class of error before it starts.

Preventing problems upstream is always cheaper than fixing them downstream.

Simulations are becoming more reliable

One of the persistent challenges in retail AI is that measurement is inherently messy. You can't run a clean, controlled experiment when a competitor opens a store next door mid-pilot, or when an unseasonable heatwave shifts demand across your test window.

For years, simulations served as a theoretical workaround, but they were often unreliable.

That's changing. Agentic AI is making simulations considerably more capable by enabling the addition of more objectives, variables, and real-world complexity.

It doesn't eliminate the need for real-world testing, but it raises the quality of the signal you get before you commit at scale.

The feedback loop is now a conversation

Perhaps the moment that stayed with me most was around how feedback loops have evolved. In early machine learning, the model gave you a number, and you acted on it. Humans were fairly passive.

Now, with agentic workflows, a store associate can tell the system it made the wrong call, explain why, and improve the algorithm. In natural language. Without a data scientist in the loop.

This matters more than it might sound. Inventory management, to take one example raised in the session, is not a pure data science problem.

Experienced buyers know things the algorithm doesn't:

• a particular vendor's reliability

• an upcoming local event

• a category quirk that's unique to their market

That knowledge used to be locked inside people's heads. Now it can flow back into the system.

At Scandit, we see this dynamic in our ShelfView deployments that are powered by vision AI.

When shelf monitoring insights are embedded into daily store workflows rather than presented in a weekly report, the quality of feedback improves significantly.

Associates notice when something doesn't look right. They flag it in the moment. The system gets smarter. The value compounds over time rather than plateauing after launch.

For vision AI, the real ROI is on the revenue side

I want to be direct about this because it reflects what we consistently observe across our retail deployments.

When vision AI is used to improve on-shelf availability or planogram compliance, the primary business case is sales, not labor reduction.

When a product is on the shelf when a customer wants it, the retailer makes money. When it isn't, they lose the sale and often the customer.

The gap between what inventory systems report and what actually sits on a physical shelf is larger than most retailers realize, and vision AI continuously closes it rather than through periodic manual audits.

Labor efficiency tends to be a side effect. The store associate who used to walk every aisle is now guided directly to the item that needs attention. They fix it faster. The time saved gets redirected to something more valuable.

It is a real productivity gain, not a reduction in headcount.

Change management is where most pilots fail

Every panelist touched on this. Repeated POCs without clear wins create fatigue. Fatigue breeds skepticism.

The adoption model that resonated most was the bottom-up approach rather than the top-down approach.

You need people embedded in each functional area who genuinely understand the tool, can show small wins to their colleagues, and build the credibility that no steering committee can manufacture.

Executive sponsorship matters for direction and budget. It doesn't create believers. Believers come from seeing something work with their own hands.

One practical point worth flagging: if you're approaching this from the technology side of the organization, be careful how you frame productivity improvements.

The conversation about what to do with the time saved is better led by the business owner whose team it affects. Technology's job is to prove the capability works. Business leadership's job is to decide what changes as a result.

Main takeaways

Retail AI is maturing, but the fundamentals haven't shifted. Measure what matters. Fix your data first. Embed AI insights into the actual work, not into a separate dashboard that gets checked when someone remembers to look at it.

The new tools, particularly around simulation quality, natural language feedback, and agentic workflows, are genuinely moving the needle on what's achievable.

But the retailers who will extract real value aren't necessarily those with the most technology.

They're the ones who've built the organizational habits to act on what the AI tells them.

That's still the hard part. And it's still what separates a pilot from a P&L impact.

Your brain picks out the expiry dates on the packaging labels instantly. Without even thinking about it. You ignore the packing date printed next to it, the barcode below, the weight, the price, and the nutritional information panel.

Now ask a barcode scanner to do the same thing. It’s impossible.

Traditional barcode scanning follows a simple contract: one barcode in, one data string out. Point, scan, done. That works beautifully when all the data you want is contained in one barcode on a simple label.

But walk through any real-world store and you'll find labels that laugh in the face of that simplicity. Electronics packaging with multiple barcodes (product code, batch number, serial). Food labelled with printed weights and expiry dates. Batch and lot numbers. And did I mention price labels?

Check out this montage and you’ll see what I mean. Labels aren’t even standardized. Layouts shift depending on retailer, product category, even the season.

For frontline workers scanning hundreds of items a day, this complexity adds up fast. They scan one barcode, manually type in the weight, visually check the expiry date, scan another barcode for the serial number. Repeat. All day.

That's the problem we built Smart Label Capture to solve. Powered by the Scandit AI Engine, it lets users scan an entire label with one press, and returns a complete, structured data set of exactly the fields their workflow needs.

Not everything on the label. Just what matters for the task at hand.

I've written before about using AI to scan multiple barcodes simultaneously. Smart Label Capture required something different. We taught machine learning models to interpret labels the way humans do, by understanding context, not just reading characters.

The hardest part isn't even reading the text. It's knowing which text to read.

That expiry date your brain effortlessly picks out and reads? Like many other things to do with vision, what’s easy for you is a hard engineering challenge.

Even using optical character recognition (OCR) to read an expiry date is difficult. OCR that works just fine on a typed document struggles when faced with the myriad of unstandardized formats, abbreviations, printing methods and fonts on everyday labels.

But that isn’t even the hardest part. Before you even start decoding text, you need to find the correct text to decode in a very busy camera frame.

That frame might also contain a packing date, a manufacturing date, a supplier name, a lot number that looks suspiciously like a date, and a QR code encoding a timestamp.

In the image below, you can see Smart Label Capture capturing barcode data, net weight, and unit price, while ignoring total price, packing date, sell by date, and supplier information.

Knowing what information not to return is as important as knowing what data to capture.

Then you need to map that result to a field the target application understands. Your backend doesn't want the string "2.17" floating in isolation. It needs weight_kg=2.17, combined with price=4.50 and the product barcode.

All at the same time, and in a structured format ready to feed into inventory systems or point-of-sale software.

In other words, scanning complex labels relies on meaning, not just data extraction.

How Smart Label Capture works

The Scandit AI Engine distinguishes between different label fields — and label types — by understanding visual cues. In human terms, we call these "keys".

When you look at a label, you're constantly using keys. You see a dollar sign, and your eyes jump to the number next to it. You see "EXP", and you know the date that follows matters for freshness.

We designed Smart Label Capture to apply the same reasoning, at machine speeds, using a combination of OCR, machine learning, and computer vision.

It searches for visual cues — text positioned near barcode candidates, formatting patterns, spatial relationships. Then it uses those cues to classify nearby values according to specific workflows, such as:

• Format checks: Does the captured value match an expected pattern? A valid weight format, a price with two decimal places, a date in DD.MM.YYYY or MM/DD/YYYY?

• Context checks: Does the value sit near the correct cue text? Is a number positioned beside "Net Weight" or "Gross Weight"? Does it align with the expected label structure for this product category?

• Workflow checks: Do multiple extracted elements agree with each other and with the requested workflow? For VIN scanning, that's a barcode with alphanumeric text positioned above it. For weighted grocery items, that's weight + price + product identifier, with expiry date optional depending on the category.

Smart Label Capture ships with several pre-built label types covering common labels, such as price (pictured below), expiry date, packing date, weight, serial number, IMEI, and VIN number.

It also supports fully custom labels. With a bit of work, it will handle any combination of fields, any position, any layout you throw at it.

What does AI-powered label scanning look like in practice?

Our goal was never to "return everything." We designed Smart Label Capture to return only the fields necessary for a given workflow.

Weighted items at a grocery checkout? You need weight and price, plus a product identifier. That's it.

Fresh goods in a warehouse receiving dock? Expiry date and lot number become critical. Weight might not matter at all.

Electronics inventory with serial tracking? You need UPC number, serial number, IMEI1, IMEI2. As you can see in the image below, these are all encoded as separate barcodes.

The Scandit AI Engine understands what's needed for each context, and compresses what used to be multi-stage workflows — scan, type, verify, scan again — into a single action.

That results in:

• Fewer scans, because Smart Label Capture captures multiple barcode and text elements at once, not sequentially.

• Faster per-item capture, because users don't need to decipher labels or manually enter information

• Higher accuracy and fewer downstream corrections by eliminating manual data entry.

One of our US customers used Smart Label Capture to prevent undercharges on weighted items. They protected $1.3M in annual revenue and gained 500 hours annually through 7× faster picking.

Not 7% faster. Seven times faster.

Why AI-powered label scanning belongs in your workflows

Do your scanning workflows still rely on capturing barcodes only — and pushing the rest of the cognitive load onto frontline workers? Then you're paying for it in lost time, downstream errors, and employee frustration.

Smart Label Capture works because it understands the semantics of label elements, not just their pixel patterns. It treats every label as the sum of its parts, not as separate, disconnected pieces of information.

That leads to faster capture times and higher downstream accuracy. And, perhaps most importantly, frontline workers who can focus on decisions that actually require human judgment, instead of fighting with their tools.

The real problem isn't the waste itself. It's visibility. And until that visibility problem is solved, the waste will keep happening, no matter how good the forecasting tools get.

The numbers are hard to ignore.

For grocers specifically, the hidden costs of managing unsold and surplus food amount to €90 billion a year. That's not just the cost of the food. That's the labor, the logistics, the compliance burden, and the lost margin baked into every product that ends up in a bin.

Perishables drive over 80% of food waste, yet the way most grocers manage expiry dates today looks almost identical to how it was done 20 years ago.

What happens when expiration checks are missed?

Timely grocery markdowns recover revenue.
Late markdowns lose margin.
Missed markdowns generate waste.

And any expired product that makes it onto the shelf is a customer experience problem, not just an operational one.

The logic is simple. If a grocer knows a product is nearing its expiry date early enough, they can markdown the price, attract price-sensitive shoppers, and still recoup some of the item's value.

That's better than throwing it out.

But the process that needs to happen before that markdown can even be applied, capturing the expiry date in the first place, is where everything tends to fall apart.

Our work with leading grocers puts store coverage of expiry date capture at around 30%. Some grocers don't capture expiry dates at all, relying instead on system logic and periodic shelf checks to execute grocery markdowns.

That means 70% of perishable items on the shelf are effectively invisible.

Why expiry date management is stuck in the past

Walk into any major grocery store, and it looks like things are running smoothly.

Associates are scanning items, applying stickers, and managing the floor. But beneath the surface, a systemic problem is quietly compounding.

Capturing expiry dates, where it happens at all, is still done manually.

Workers scan a barcode, then read and type the printed date separately. Across hundreds of items, every single shift. It's tedious, it's error-prone, and it's difficult to scale.

The answer to why this hasn't changed lies in the nature of the problem itself:

• Most products only carry a standard 1D barcode.

• The expiry date is printed on the label, not encoded in a way a scanner can read.

• 2D barcodes that could change this are on the way, but adoption is slow.

So until now, there hasn't been a clean, fast, scalable way to capture both pieces of information in a single step.

Neither manual nor semi-automated capture is good enough, given the sheer volume of tasks store associates are expected to complete daily.

And without broad store coverage, grocers simply can't see what's actually sitting on their shelves.

The value of closing the gap

Grocery margins are razor-thin. Even recouping a fraction of an item's cost through a timely markdown is better than a full write-off. But all of that depends on visibility.

When associates can quickly and accurately capture expiry information alongside product barcodes, grocers finally have a real picture of what's on their shelves.

From there, grocery markdown logic can be applied with confidence.

• Category managers define the rules.

• The system flags items approaching expiry.

• Associates know exactly which items to check, when to mark them down, and by how much.

The consequences of not having that visibility multiply fast: products expire and get thrown away, or worse, a customer buys them.

Early markdown opportunities are missed. Price-sensitive shoppers, who would have happily bought a reduced item, walk past it instead.

The margin that could have been recovered is gone.

How to automate expiry date capture at scale

Scandit's Expiry Management and Markdowns Solution addresses the root cause directly. It enables automatic capture of product barcodes and expiry dates.

High-accuracy OCR handles the wide variety of date formats found across different labels. The workflow is fast enough that adoption actually sticks, without significant additional training.

The result:

• 60% reduction in the time needed to capture expiry dates per item.

• Store coverage that scales to include every product the grocer wants to track.

• And with more timely markdowns in place, food is sold, helping reduce waste by up to 40%.

What are the benefits of better expiry date management?

The grocers who solve this first won't just reduce waste. They'll recapture margin, improve compliance, enhance customer trust, and turn a persistent cost center into a genuine competitive advantage.

The waste problem was never really about waste. It was always about visibility.

Now, for the first time, grocers can actually see what's on their shelves before it's too late.

FAQs

Still got a question? Contact us.

Creating a brilliant ID scanning experience on the web results in higher completion rates. These directly translate to improved business metrics such as the number of customers arriving at the airport ready to fly.

But achieving this requires a customizable interface that supports whitelabelling, internationalization, and web accessibility standards. This is where the ID Bolt Studio helps. The easy-to-use dashboard allows you to experiment with brand styling and workflows before deploying an auto-generated code snippet directly to your website.

In this blog and video, I’ll walk through configuring and deploying web-based ID and passport scanning with the ID Bolt Studio, using an example airline site.

What is ID Bolt?

Scandit’s ID Bolt is a pre-built, secure ID scanning solution for any website, enabling instant scanning of passports, driver’s licenses, ID cards, and other forms of identification. It’s designed to reduce your deployment time, improve customer experience, and meet WCAG accessibility standards, without compromising data collection and regulatory requirements.

How does the ID Bolt Studio help?

With the ID Bolt Studio, you can quickly and easily apply your desired scanning workflows and brand to ID Bolt. Test different configurations and whitelabeling options directly in your Scandit dashboard, then copy working code to your website.

Here are the features you can configure using the ID Bolt Studio:

Ready to try the ID Bolt Studio? Let’s walk through an airline example.

ID Bolt Studio walk-through

Let’s say I work for an airline that accepts any passport valid for at least six months, but no other identity documents.

My website looks like the screenshot below, and I have four branding guidelines that the ID Bolt pop-up and phone experiences must match:

• Primary color: #007887

• Primary font color: #111827

• Secondary font color: #6B7280

• Font family: Montserrat

Pre-requisites

• A valid ID Bolt license key. (If you don’t have one, go to Scandit dashboard > Configuration > License Key, and click Generate a license key.)

• ID Bolt has been installed as a dependency to my website. (For instructions on how to do this, read How to Add a JavaScript Passport or ID Scanner to Web Apps.)

• Brand font files in WOFF2, WOFF, TTF, or OTF format. The maximum file size is 100 KB.

• Brand logo file (if required) in PNG, JPEG, SVG, or WebP format. The maximum image size is 50 KB.

After logging into my Scandit dashboard and selecting Configuration from the left-hand navigation, I see something like this:

In addition to the settings for ID Bolt’s features, I get:

• Preview: To see how your ID Bolt customizations will look before deploying them to production.

• Launch Code: JavaScript code you can copy and paste into your website.

• Launch ID Bolt: Open a live instance of ID Bolt to test your customizations with real documents, before deploying them to production.

1. Define allowed ID document types

1. Under Identity Document Configuration, I’m going to deselect all document types. Then, using the drop-down, I’ll select Canada - Passport, United States of America - Passport, and Any - Driver’s License. You can use the search feature for quicker access.

2. Under Validation Rules, I’ll select Must be valid for at least and enter 6 in the Months field.

2. Customize user interface

1. Under Fonts, I’ll click Change Font to upload my font files, and repeat this step for all font weights.

2. Under Customize User Interface > Colors, I’m going to set the following fields to the given values:
   a. Primary: #007887
   b. Image: #007887
   c. Text Primary: #111827
   d. Text Secondary: #6B7280

3. To change the link color, I’ll update the CSS by navigating to CSS Styles, enabling Link Style, and entering the following code:

3. Test new configuration

1. To see an updated preview, I can click on the Desktop or Mobile buttons under Preview. The instruction text and colors are updated to match my settings.

2. To try the new configuration in a live ID Bolt session, complete with a modal dialog to scan a real ID, I can click Launch ID Bolt. If you select Scan by phone, your phone experience will also match the customizations defined above.

   

3. If you click Show all data on the dashboard, you will get a dialog containing a copyable data structure of the scanned ID card’s data and images of the scan.

4. Add the new configuration to your website

Once I’m satisfied with the new configuration, I’ll copy the code under Launch Code and add it to the startIDBolt() method in my website application code.

And that’s it! Custom-branded ID scanning on your website in just a few minutes.

AI-powered configuration coming soon

Things are about to get even easier. With the AI theme generator we’re working on, you’ll be able just to provide the ID Bolt Studio with a URL. The AI theme generator analyzes your website and applies brand colors, fonts, and style immediately to ID Bolt’s configuration code.

You get the correct branded experience instantly, with no manual configuration needed.

That’s all it takes to create a branded and accessible ID scanning experience for everyone. Ready to try it yourself? Get a free 30-day ID Bolt trial and you’ll be scanning IDs in no time.

Driver’s licenses, passports, and visas vary widely. Many also remain valid long after design changes, which means older formats never truly disappear.

At the same time, fraudsters are getting better at mimicking legitimate documents. Today’s fake IDs can often only be detected by subtle formatting and encoding errors that humans and traditional ID scanners — constrained by rigid pattern-matching rules — struggle to detect.

Many industries use AI to interpret and validate dynamic datasets like this, but it usually comes at the cost of increased compute resources. This makes scanning and verifying IDs efficiently on frontline workers’ devices without blowing up processor cycles or battery life a hard engineering problem — one that Scandit has solved.

This blog explains how the Scandit AI Engine understands real-world identity documents while keeping data processing on-device by default to improve security and reduce latency.

Why do rules-based ID scanners fail on real documents?

Traditional, rules-based ID verification scanners are built on known formats, stable document layouts, and predictable encoding standards… and hoping they won’t change too often. They focus on extraction, rather than understanding, by mapping expected data to expected outputs.

In today’s world, this static approach to ID verification breaks down in these ways:

• Many jurisdictions implement ID structures and formats differently, even under the same industry standard, making it difficult for traditional algorithms to scale to all use cases.

• Authorities update document layouts and data formats frequently, making traditional ID scanning algorithms obsolete quickly.

• Additional forms of verification are growing in use, such as certificates, visas, and other document types that traditional ID algorithms are not designed to handle.

For example, U.S. driver’s licenses follow the published specification for PDF417 barcodes, but individual states encode data differently. The field order, padding, delimiters, and optional attributes vary widely, and fake IDs often exploit these inconsistencies, knowing that most scanners look for “good enough” encoding.

In the U.S., our fake ID research showed that 45% of US adults aged 18-25 know someone who used a fake ID successfully to access age-restricted products or venues.

These challenges require ID scanning and verification software that can capture and analyze unknown and unstructured data, where the layout, format, and data encoding are not necessarily defined or known by the vendor.

How does AI-powered identity verification go beyond standard documents?

The Scandit AI Engine combines machine learning (ML), computer vision (CV), vision language models (VLM), and other technologies like optical character recognition (OCR) to transform data captured from a device camera into structured, verifiable identity data that businesses can trust.

The engine understands how fields are stored, how barcodes are generated and printed, and how visual elements are laid out across the ID — without relying solely on limited static definitions.

This allows our solutions to address specific challenges in ID scanning and verification:

• Accurate data capture: Improving extraction accuracy, up to 100% for major document types, even when IDs are obscured, damaged, under low light, and a variety of other degraded conditions.

• Fake ID detection: Scaling up analysis to look at hundreds of document characteristics, in less than one second, to achieve 99.9% authentication accuracy in real-world scenarios.

• Broader document support: Expanding capture and analysis to non-standard documents required in identity verification processes, such as visas, certificates, and letters.

The Scandit AI Engine also does this fast, on frontline workers’ or consumers devices, and with no identity data stored on-device. Designed to reduce risk in customer-facing teams and workflows without compromising speed, security, or user experience, scanning and validation takes less than 1 second.

How AI-powered identity verification works

Here’s the end-to-end workflow of the Scandit AI Engine for ID scanning and validation:

1. Locate and track the document: Using CV and image processing techniques, the engine detects the document in a live video stream and tracks it across frames. This allows it to select the frames with optimal characteristics, such as lighting and resolution, rather than relying on the quality of a single captured image.

2. Extract the data: The engine decodes barcodes and recognizes printed characters using OCR and ML models. It also detects physical characteristics such as punched holes, clipped corners, or perforations that indicate voided IDs.

3. Parse the data: Raw extracted values are converted into structured fields that are human-readable and queryable by other systems.

4. Analyze and authenticate: Structural analysis and AI-based methods are used alongside more straightforward techniques such as automated age checks. Of course, businesses can also take the structured data and validate it themselves, for example against a passenger manifest.

Scandit supports all major ID formats — Visual Inspection Zones (VIZ), Machine Readable Zones (MRZ), and PDF417 barcodes — and uses AI to parse unstructured and “messy” formats.

How AI-powered fake ID detection works

AI improves fake ID detection by understanding variation and nuance, such as knowing that numeric strings positioned next to a “Date of Birth” field are different from similar strings elsewhere, and that barcodes can be encoded and printed differently across jurisdictions.

Scandit’s ID scanning AI models are trained and validated on massive real-world datasets comprising millions of scans. For fake ID detection, the engine learns subtle, jurisdiction-specific characteristics, such as how a driver’s license in one state differs from those issued elsewhere.

One of the largest food delivery companies in the US scans half a million identity documents a week using the Scandit AI Engine. The data from this rollout shows that it detects deviations with 99.9% reliability in real environments.

How AI enables broader ID document support

AI enables broader ID document support by understanding non-standard forms of verification. Examples include e-visas, invitation letters, vaccination certificates, and approval PDFs emailed to travelers. These documents vary widely in layout and language, making traditional scanning techniques incapable of handling them.

Using VLMs, the Scandit AI Engine can extract relevant fields from document formats that have never been seen before by either AI models or frontline workers. It then outputs structured data that your app can validate against scanned passports or itinerary requirements without manual user input.

This allows airlines and travel platforms to verify multiple required documents against country or route-specific rules, reducing costs and operational risk.

What about the risk of blowing up compute resources? All our customers get a final, trained model that’s optimized to interpret millions of pixels in milliseconds on consumer devices. In most instances, this model runs on-device, keeping latency low and sensitive data where it belongs.

Minimizing hallucinations to boost accuracy and trust

Unlike general-purpose AI systems, the Scandit AI Engine is designed to narrow possibilities, not expand them.

Scandit monitors and maintains high inference accuracy, which measures how reliably a model produces correct outputs on new, unseen documents. To reduce risks, such as AI hallucinations, Scandit trains models on domain-specific data and constrains them to narrow, purpose-built tasks. Pre- and post-processing steps are also applied throughout development to ensure outputs remain within expected bounds.

A practical path to accurate, reliable ID verification

Simply extracting ID data is no longer enough. ID scanning software must understand how identity data is structured, how real documents differ from fake ones, and account for unexpected differences in the field.

And AI or not, the software must be easy to integrate with your existing systems and easy to use, otherwise all its value will remain untapped.

Scandit brings together over 15 years of scanning expertise, modern AI-powered identity verification techniques, and a mobile-first architecture to solve these challenges end-to-end. As with other Scandit products, our goal is always to create a system that “just works” and that neither the developer nor the user has to worry about.

• ID Bolt allows any business to launch an AI-powered self scanning journey on their website to reduce customer queues, frustrations, and administrative costs.

• The ID Scanner SDK provides fully customizable data capture and fake ID detection capabilities across various native, web, and cross-platform frameworks.

• With the turnkey app Scandit Express, businesses can scan IDs into any app, with no need to make any software changes.

One of our customers, a large US food delivery company, used Scandit software to validate over 2 million IDs monthly and gain over $25M in incremental revenue by expanding alcohol sales to new states.

Businesses everywhere are replacing manual ID checks with automated ID verification. But the ID verification software landscape is complex, information security is critical, and choosing the right solution isn’t always easy. Understanding the fundamentals of how AI powers modern identity verification helps cut through the noise and find the right option for you.

They shift priority, head out back, and scan the crowded storage area with their device. An augmented reality (AR) overlay instantly identifies not only the right product, but the box with the soonest expiry date.

Shelf restocked in minutes with the optimal product. No guesswork, no wasted time, no sales lost.All this is powered by the Scandit Vision AI Engine — the intelligence that powers Scandit Smart Data Capture.

By automating complex tasks and providing real-time insights, the Scandit Vision AI Engine empowers workers and customers to make smarter, faster, more accurate decisions. Think of it as the brain behind every scan, guiding actions and driving efficiency across an entire business.

What is vision AI?

Vision AI is the branch of AI that understands visual information from the physical world. Whereas generative AI (GenAI) creates new content, vision AI analyzes what is already there. It tells you what actual products are on your actual shelves, whether an identity card is real or fake, or what information you need from a complex label printed with three barcodes and six text fields.

Vision AI is related to the much older term computer vision — people have been trying to teach computers to "see" for decades — but is now replacing it. This reflects the way in which AI has accelerated the field and unlocked use cases impossible to achieve with traditional algorithms.

Vision AI analyzes visual data and surfaces actions from insights in real-time, making it highly effective in real-world environments.

What makes the Scandit Vision AI Engine special?

Vision AI looks outwards to the physical world, not inwards to a recursive digital universe.

Retail, logistics, manufacturing, and healthcare are not social media platforms. They depend for their existence on physical objects and processes — such as making sure products are on shelf when consumers want to buy them, delivering the right goods to the right person at the right time, or giving a patient the correct medication.The Scandit Vision AI Engine is purpose-built to solve these specific, measurable real-world problems. It's been developed through more than 15 years of building solutions for the world’s largest retailers, manufacturers, couriers, logistics brands, and pharmaceutical companies.

What makes the Scandit Vision AI Engine special is as much the principles driving it as our technical expertise:

1. Accuracy is non-negotiable: Getting to 99.9% accuracy is key in our space. When we say that an ID is fake, that there are five product facings, or that a patient-medication combination is correct, there’s almost no tolerance for error.

2. Security and privacy come first: By default, our models are not trained on live customer data. Instead, customers get a final, trained model, with the ability to interpret millions of pixels in milliseconds. In most instances, the model runs locally on end-user devices, keeping sensitive data where it belongs.

3. Models must run efficiently on commercial mobile devices: From the outset, we’ve had to build to run on frontline workers’ devices, where speed is critical and battery life precious.

4. AI is used in a deliberate and targeted fashion: We focus on automating specific (and tedious) tasks that previously required substantial manual intervention. This delivers significant, quantifiable value to businesses.

5. Workflows are always user-centered: The Scandit Vision AI Engine is built to deliver fast, robust, and ergonomic data capture even in difficult conditions. The overarching goal is always to create a user experience that “just works”, and that neither the developer nor the user has to worry about.

Here’s an overview of how the Scandit Vision AI Engine powers our three main product lines. Our barcode scanning products share a unified codebase and common machine learning models. ID scanning and ShelfView are technically distinct — but they're built by the same team, draw on the same deep expertise, and follow the same principles.

Barcode scanning

Scandit has been using machine learning (ML) to decode barcodes from the camera feed of smart devices for over 15 years.

Back when we founded Scandit, that was a hard engineering problem. Modern AI was in its infancy. The cameras on early smartphones also lacked autofocus and were simply not high enough quality.

Early on, the CIO of one of Europe’s largest grocers told us that our technology would never be good enough. We now count seven of the top ten US grocers and five out of the top ten European retailers as customers.

If you don’t work in the field, it’s easy to underestimate just how hard vision AI is. Humans are highly visual creatures. Around half of your brain — the most powerful supercomputer we know of — is devoted to visual processing.

What’s easy for humans is hard for computers. Take a look at the image below. It's easy for the human eye to tell that the barcode the user wants to scan is the one on the product they're holding, not any of the barcodes visible on the shelf in the background.

Until 2024, when the Scandit SDK 7 released, there was no software in the world that could do that reliably.

Let's break it down. To successfully scan the barcode in this image, the Scandit Vision AI Engine has to:

• Identify which barcode the user wants to scan, by analyzing contextual data such as movement and barcode characteristics.

• Decode the identified barcode, which in this instance is tiny, printed on a curved surface, and with some glare. (Other factors we often contend with are damaged codes, low light, extreme angles, and blurry images.)

With the Scandit Vision AI Engine, edge cases like this don’t need to be hard-coded. Instead, our AI-powered barcode scanning does all of this automatically. It adapts to different environments to reliably scan codes without requiring explicit instructions from either the developer or the end user.

And that’s just scanning a single code. Our MatrixScan products do all of the above, but here the Scandit Vision AI Engine also scans multiple codes in parallel.

It tracks their position and adds augmented reality (AR) overlays to solve specific use cases such as counting, finding, or picking items.

Smart Label Capture, our newest barcode scanning product, goes beyond scanning multiple barcodes to capture text and understand label structure.

With one press, users can scan complex labels in their entirety: multiple barcode formats (e.g. 15-digit IMEI numbers), field positioning, and contextual relationships (e.g. "BEST BEFORE" adjacent to a date).

The result is that your application receives precisely the data it needs with a single press. Nothing more. Nothing less.

ID scanning and verification

Identity documents are messy: formats vary across jurisdictions, older designs stay in circulation, and fraudsters mimic subtle layout and encoding details. This makes accurate ID checks hard for frontline teams.

In ID scanning and verification, the Scandit Vision AI Engine combines ML, vision language models (VLM), and other technologies like optical character recognition (OCR). This transforms unstructured visual input into structured, verifiable identity data that users can trust.

Traditional ID scanners rely on rigid templates and mainly “extract” data, so they fail when documents deviate from specs or when fakes exploit inconsistencies. For example, US driver’s licenses all use PDF417 barcodes, but encode fields differently by state.

Scandit's AI-powered identity verification works differently. Instead, it mirrors how counterfeit art is detected. Not by looking at the subject itself, but by examining brush strokes, techniques, and materials to assess whether they align with an authentic work.

Similarly, the Scandit Vision AI Engine doesn’t simply capture and decode individual data values. Instead, it analyzes structural characteristics — how fields are stored, how barcodes are generated and printed, and how visual elements are laid out — to accurately capture data and detect fakes.

ShelfView

Our shelf analytics product, ShelfView, is a specialist solution that analyzes images of store shelves captured using smart devices, fixed-position cameras, and robots. Similarly to how LLMs are trained, we ingest vast datasets to understand retail shelves comprehensively and accurately.

It lets retailers see what is really on their shelves, not just what their inventory system says.

• Scene parsing identifies trays, shelves, products, and shelf labels.

• Products are identified down to the SKU level with image recognition.

• OCR and barcode scanning are used on shelf labels to extract product and price information.

Together, they create a precise digital representation of the shelf (often called a realogram). Associates then receive prioritized alerts for missing or misplaced products, pricing updates, and promotional errors, so issues can be resolved before customers notice them.

What's next for the Scandit Vision AI Engine?

The future of AI lies in domain-specific models tailored to unique industry needs. Ultimately, the Scandit Vision AI Engine is moving towards a situation where you don’t scan barcodes, scan IDs, scan objects, or scan text. You just… scan.

Holistic scene analysis will ingest multiple data sources and context, then return exactly the data and insights you need for your specific scenario — whether you’re a consumer, store associate, delivery driver, or operations manager.

You won't have to switch tools. You won't have to explicitly instruct it in what you want.

Our new store intelligence platform is an important step in that direction. Here, workflow data from specific retail solutions (shelf intelligence, expiry date management, planogram compliance, and in-store picking) is captured and consolidated into a shared intelligence platform.

It doesn't just provide inventory visibility, but real-time actions for store teams.

The evolution of vision AI makes it possible to add a comprehensive intelligence layer between business environment and user. That's something that retailers, healthcare providers, logistics companies, manufacturers, and all the other industries who create value in the physical world have never truly been able to do before.

Look up from your screen. The world's biggest problems are physical, not digital. The Scandit Vision AI Engine tackles this head on.

Scanning multiple barcodes at once requires more than running the same algorithm repeatedly. It needs contextual awareness, spatial understanding of the environment, and pinpoint positioning of visual feedback on users screens.

Ideally, it’s also customized to the exact workflow it’s intended to accelerate. Otherwise, the software won’t know whether to return 20 barcodes for cycle counting versus one barcode out of many for order picking.

Here’s how we built the Scandit AI Engine to make multiple barcode scanning smarter.

Why is multiple barcode scanning difficult?

It’s no surprise that our recent Frontline Retail Revealed report found that scanning one barcode at a time is the top scanning pain point for store associates.

They walk down aisles and scan barcodes one by one. Each time, they have to move between items, tilt their device at different angles, check that labels are in focus, and double-check that the captured data is correct.

Building technology to automate all of this requires a different approach than single barcode scanning.

The technology must quickly identify all visible barcodes, determine which ones matter, track them as the device moves around, and present clear, stable visual feedback to the user. And it has to do all of this without draining the battery — because it runs not on a laptop, but on frontline workers’ mobile devices, where usage is heavy and battery life precious.

The Scandit AI Engine integrates three strategies into one cohesive process to solve this:

• AI-powered scene analysis: Computer vision (CV) techniques continuously evaluate the scene — barcode positions, angles, distance, focus, and device motion sensor data — to determine which barcodes to capture and track over time.

• Aligning digital and physical worlds: The barcodes’ dataset is mapped onto augmented reality (AR) overlays that help users relate on-screen content to real-world objects, reducing guesswork and errors.

• User-centered workflows: The overall user experience (UX), from UI elements to task-specific workflows, is optimized for specific use cases such as counting, picking, and task tracking.

Our challenge was to design a system where these strategies cooperate across various mobile devices and operating systems without excessive computational overhead.

Making multiple barcode scanning smarter

To batch scan barcodes reliably in real-world enterprise environments, the Scandit AI Engine addresses five challenges:

1. Speed and resource balance: Every additional barcode in the camera’s field of view multiplies computational load. The Scandit AI Engine’s CV models use algorithmic optimization to interpret millions of pixels in milliseconds. This allows them to decode multiple barcodes in parallel, without increasing latency or draining battery life.

2. Avoiding duplicates and false positives: Meeting business goals requires extremely high accuracy. The engine’s CV models identify barcodes uniquely, and track each one across frames. It works even if a barcode is tiny, printed on a curved surface, or looks identical to others nearby. This is crucial for workflows where many items share the same SKU and each one must be counted, not dropped.

3. Tracking through motion: As the camera moves, the Scandit AI Engine remembers each barcode’s spatial position, even if it goes outside the field of view. This avoids forcing the user to recapture the same barcode and allows the software to count duplicates.

4. Handling imperfect conditions: Low-light environments, curved, damaged, or tiny labels, and low-resolution cameras require adaptive decoding. All Scandit barcode scanning products include CV models that dynamically compensate for blur, glare, camera quality, and other degradations.

5. Stabilizing AR overlays: The AR layer runs in sync with upstream algorithms, ensuring visual cues remain anchored to physical objects and do not flicker — a subtle yet critical factor in maintaining user confidence.

Each capability is designed for tight cooperation between the engine’s CV models, AR rendering logic, and UX design, and is optimized for real-time performance on commercial mobile devices. And in case you’re wondering, our models aren’t trained on live customer data. Instead, customers get the final, trained model, preserving security and privacy.

Our overarching goal is to create a system that neither the developer nor the user has to worry about. The engine provides clear, easy-to-use APIs and UX-optimized workflows that act as a targeted intelligence layer between environment and users — without anyone having to worry about them.

From AI to action: MatrixScan options

Scandit’s MatrixScan products package the Scandit AI engine’s capabilities into configurable workflows ready to deploy. Each one builds on the same foundation, providing developers and product managers with scalable options, ranging from fully packaged components to flexible APIs that allow fully customized integrations.

• MatrixScan Count: Lightning-fast, intuitive, and error-free scan and count solution for receiving goods, inventory, and cycle counting.

• MatrixScan Find: Scan multiple items and instantly highlight the right one using augmented reality (AR)

• MatrixScan Pick: Turn any scan into a visual, interactive checklist that makes following complex workflows consistent and efficient.

• MatrixScan Batch: Batch process barcodes and handle multi-code labels with unparalleled speed and accuracy.

• MatrixScan AR: Capture multiple barcodes and display information with customizable augmented reality (AR) overlays.

Multiple barcode scanning needs collaborative technologies

The Scandit AI Engine brings intelligent perception, adaptive decision-making, and responsive UI to users’ fingertips.

To learn more about Scandit AI Engine-enabled products, explore our MatrixScan family of multiple barcode scanning solutions.

Retail is a busy environment. Products come and go. Shelves are restocked and emptied throughout the day. Employees work hard to keep inventory moving, customers satisfied, and the store operating smoothly. At the heart of all this is one key need: having clear, accurate information exactly when it’s needed. In other words, visibility.

Enter retail computer vision - the technology that helps stores truly see what’s happening.

What is retail computer vision?

Retail computer vision uses cameras and AI to understand what is happening in a store by visually observing it. Single-image captures and video sequences of barcodes, products, and text from different cameras, such as smart devices and fixed-position cameras, are transformed into data that allows retailers to see shelves, products, and labels, not just monitor transactions.

Retail computer vision converts physical store activity into digital data to improve visibility and decision-making. It turns smart devices into powerful retail tools. Helping workers see more and act faster.

Not all retail computer vision works the same way

You can think of it like a set of tools, each designed for a specific job or challenge in retail. Scandit uses computer vision across its platform in this way.

Let’s look at the most common uses and how they benefit retailers.

Barcode capture

Retail runs on barcodes. But slow, inaccurate scanning can lead to inefficiencies and operational issues.

Using a smart device camera, Scandit applies machine learning to detect barcodes in images. We train our models to recognize codes in many different settings. Once a barcode is found, computer vision decodes it. Even if there are multiple codes, the barcode is damaged, very small, or has glare, our advanced algorithms can handle these challenges.

And capturing barcodes on a smart device this way enables immediate task visibility. Real-time feedback and AR overlays guide associates on their next-best actions.

This means fewer errors and faster workflows. Perfect for order fulfillment and managing inventory.

Text capture

Not everything is included in a barcode. Dates, weights, price tags, and serial numbers are all needed for operational visibility.

Scandit’s Smart Label Capture tool collects this information along with barcodes. It uses computer vision, optical character recognition (OCR), and AI to extract and analyze label data in many different formats.

It understands the structure of barcodes and text (e.g., an IMEI number is always a sequence of 15 decimal digits), the relative position of fields, and the context of elements close to each other (e.g., “BEST BEFORE” next to an expiry date).

This accelerates data entry and eliminates manual errors, like entering the wrong weight or expiration date, preventing revenue loss and customer dissatisfaction.

Object recognition

Arguably, the biggest blind spot in a store is what’s actually on shelves. Shelves are complex, products look alike, and stock gaps go unnoticed.

Our shelf intelligence solution, ShelfView, uses advanced computer vision and machine learning to capture images of store shelves and generate actionable insights. Simply put, it lets retailers see what is really on the shelf, not just what their inventory system says.

Images are captured using smart devices, fixed-position cameras, and robots and then processed:

• Scene parsing identifies trays, shelves, products, and shelf labels.

• Products are identified down to the SKU level with image recognition.

• OCR and barcode scanning are used on shelf labels to extract product and price information.

Together, they create a precise digital representation of the shelf (often called a realogram).

Associates then receive prioritized alerts for missing or misplaced products, pricing, and promotional errors, so issues can be resolved before customers notice them.

ID scanning and verification

Selling age-restricted goods carries some risk. Fines and license revocation may occur if age and authentication checks aren't carried out, and minors receive goods.

Eyeballing an ID to determine age takes time. Detecting a fake ID is nearly impossible.

Scandit uses computer vision to perform secure ID validation checks on the device for fast, accurate authentication. It works in 3 steps:

1. The camera captures a usable image from a video feed.

2. It understands the document. Computer vision identifies the ID type and key areas, such as photos, text fields, and barcodes, instantly and on-device, ensuring data protection.

3. It analyzes authenticity. ID Validate uses computer vision models to inspect hundreds of visual characteristics. These include fonts, spacing, alignment, print quality, and format patterns.

The system looks for inconsistencies that are invisible to the human eye but common in fake IDs. It also cross-checks data. The system compares the information printed on the ID with the information encoded digitally. Mismatches raise flags.

Finally, it delivers a simple result to the user: authentic or not.

This helps retailers reduce fraud, save time, and protect their business and reputation.

The power of combining retail computer vision tools

One tool is good. Many together are better. Order fulfillment is a great example of a workflow that relies on all the tools combined.

• Object recognition plugs shelf gaps.

• Barcode scanning tracks the item.

• Label capture enters the correct weight and price.

• ID verification ensures a compliant handover.

What might seem like a complex, experimental, or even futuristic technology with limited applications is, in reality, already commonplace.

Retail changes fast, and retailers need visibility. With Scandit, you see more. You know more. You act faster.

• Seeing what’s actually on the shelf

• Reducing fresh food waste

• Making in-store fulfillment sustainable

• Helping frontline teams succeed

Here’s a concise recap of the main themes and some concrete approaches that came up in conversations at the show.

Inventory blindness: Seeing the real shelf, not just the system

Many retailers reported the same issue: systems show stock, but shoppers still find empty shelves.

Common pain points:

• Manual, sporadic cycle counts.

• Receiving errors that propagate through replenishment and e‑commerce.

• Limited trust in on-shelf availability data.

Practical approaches discussed:

• Moving from occasional counts to more frequent, lighter-touch checks using mobile devices.

• Using computer vision to capture shelves and multiple barcodes in one view so teams can scan full bays, top stock, or back-of-house areas quickly.

• Turning shelf observations into ranked tasks (e.g. highest-impact gaps first) rather than static reports.

The underlying idea: make it easy for store teams to surface and fix issues as part of their normal routines, instead of treating accuracy as a separate project.

Fresh food waste: Treating expiry dates as data, not a chore

Fresh and grocery retailers pointed to expiry as a major driver of waste:

• Date checks vary by person, time and store.

• Products close to expiry are hard to spot at scale.

• Markdown timing is often based on habit rather than insight.

Practical approaches discussed:

• Capturing both barcodes and printed expiry dates automatically, so checks can cover more items in less time.

• Aggregating expiry data to see patterns, for example, which SKUs or stores consistently drive higher waste.

• Triggering markdowns or removals from actual expiry data, rather than rough rules of thumb.

The shift is from “walk the aisle and eyeball dates” to “capture expiry once, then use it for both daily decisions and longer-term planning.”

In-store fulfillment: Making picking repeatable and scalable

As more orders are fulfilled from stores, retailers are re-evaluating how picking is done.

Key concerns:

• Time lost walking inefficient routes.

• Dependence on a small group of experienced pickers.

• Picking errors and mis-sorted items in totes.

Practical approaches discussed:

• Guiding pickers visually along an optimized route, rather than leaving path decisions entirely to individuals.

• Using the device camera to confirm the correct item on the shelf, especially where the packaging is similar.

• Showing on-screen which tote or compartment each item should go into to reduce sorting errors.

The aim is to make performance less dependent on “knowing the store by heart” and more on clear, simple guidance that anyone can follow.

Luxury & specialty: Keeping the conversation on the floor

Luxury and specialty retailers talked about a familiar tension: high expectations for service vs. the practical need to check stock and product details.

Typical challenges:

• Associates frequently leaving the customer to go to the back room.

• Fragmented access to product and inventory information.

• Clienteling tools that feel disconnected from the live interaction.

Practical approaches discussed:

• Giving associates a single, discreet device for both product information and stock checks.

• Ensuring that basic questions (sizes, variants, availability in nearby stores) can be answered without leaving the customer.

• Automating back-of-house tasks like inventory counting, product locating, and order assembly, so more time can be spent with customers.

The goal isn’t new technology for its own sake, but removing the small frictions that interrupt a high-touch interaction.

Three big questions retailers were asking themselves

Across different segments, three broader questions kept surfacing:

How do we reduce inventory blindness without overburdening stores?

Ideas included lighter, more frequent counts, better receiving support, and using shelf intelligence systems to cover more shelf in less time.

How far are our real shelves from our planograms?

Retailers are increasingly interested in “realograms”: a view of what the shelf actually looks like, captured regularly enough to spot gaps and non-compliance.

How does all of this connect to our wider ecosystem?

Many teams are thinking about how shelf and product data can feed analytics, AI initiatives and even retail media planning, ideally using the devices and cameras already in place.

Broader NRF themes: AI, media, and “quiet tech”

The show’s larger narratives aligned with these operational concerns.

Agentic AI

“Agentic AI” – systems that take actions, not just answer questions – was a major theme.

For stores, that raises a simple prerequisite: any agent making replenishment, pricing, or routing decisions needs accurate, timely information about what’s actually in front of customers.

That’s where consistent capture of item-level data (from shelves, stockrooms, and labels) becomes a critical input, not just a nice operational extra.

The store as a media asset

Retail Media Networks are moving into the aisle via digital screens, smart shelves, and app-guided journeys.

One implication: store execution (availability and pricing) and media performance are now tightly linked. If the promoted product isn’t available or correctly priced on the shelf, media results and shopper trust suffer.

That’s another reason accurate, near-real-time shelf data is becoming more important: it underpins both operations and media credibility.

“Quiet tech” and frontline work

Several speakers talked about “quiet tech”: solutions that:

• Improve inventory accuracy and shrinkage.

• Fit into existing workflows and hardware.

• Require minimal behavioral change for staff.

Alongside this, there was a clear focus on frontline experience in a tight labor market:

• Making complex tasks more guided and visual.

• Reducing training time for new staff.

• Designing tools that feel closer to consumer apps than traditional enterprise systems.

Computer vision and mobile workflows were often positioned as ways to incrementally improve the workday. Less hunting, fewer repeated scans, clearer next steps, rather than as dramatic, one-off transformations.

Key questions retailers should be asking post-NRF

Whether or not you attended NRF, the discussions suggest a few useful prompts for internal conversations:

• How confident are we in our on-shelf availability numbers, store by store?

• Where are expiry dates and other key attributes still being handled manually?

• How big is the gap between planograms and what customers actually see?

• How easy is it for a new associate to pick accurately, count stock, and verify prices?

• Are our AI, analytics and retail media efforts grounded in reliable, real-world store data?

Answering these doesn’t require starting from scratch, but it often does mean looking at how existing devices, cameras, and workflows could be used differently.

(P.S. Want to see how accurate we were in predicting 2025’s trends? Look back at our 2025 data capture trends report.)

The rise of physical AI

At the start of 2025, we predicted a shift from talking about LLMs to talking about other forms of AI, particularly computer vision. One of the biggest names in tech – Nvidia – agreed with us in 2025.

Physical AI draws its knowledge from the physical world, by capturing data from sensors (camera feeds, microphones, radar, thermometers, and more). While it includes robotics, it goes beyond this, also encompassing areas like shelf intelligence and environmental monitoring.

A chatbot can’t reduce warehouse downtime. A generative image tool can create a picture of a fully stocked grocery shelf, but that doesn’t help if your real shelves are empty. Physical AI will solve real‑world problems at a scale purely digital AI could never reach, and transform the 75% of industries worldwide that rely on physical interaction.

Hybrid data capture becomes the dominant strategy

In retail, a new hybrid era was dawning at the start of 2025, where fixed cameras, drones, wearables, and robots work in harmony with smartphones and handheld computers. This allows retailers to scale shelf intelligence solutions without the need to invest in costly hardware.

By the end of the year, hybrid data capture was becoming the operational norm, according to research conducted by IHL Group. Retailers who have already adopted a hybrid data capture strategy are 136% more likely to maintain profitability leadership.

Over the next year, adoption will accelerate, with 36% planning to deploy hybrid data capture strategies. A further 21% are planning adoption within the next 24 months.

AR‑guided workflows will go mainstream in store operations, supply chain, and last mile delivery

In 2025, Walmart, Dior, and Amazon all revealed augmented reality (AR)-powered workflows that deliver measurable, repeatable impact in retail, logistics, and supply chain. Today, Walmart’s implementation is likely the largest commercial use of AR worldwide. Dior’s solution reduced shipping control time by 85%.

Image credit: Amazon

Meanwhile, Woop's 2026 Delivery Trends Report revealed how AR will redefine last‑mile logistics as real‑time guidance becomes standard, including cutting loading errors by 30% and saving over $500,000 per depot annually.

2026 is poised to be the year AR finally crosses from early adoption to ubiquitous operational use.

AI in barcode scanning won’t remove humans from the process

Barcodes remain the cornerstone of data capture around the globe, and that isn’t likely to change in 2026. Yet even this mature technology will enter a new era as AI reshapes the future of scanning.

At Scandit, we defined AI capabilities in barcode technology across three distinct levels in 2025. These range from barcode scanning which is AI in name only, through to context-aware scanning that understands not just barcode data but environment and user intent.

But Level 3 doesn't claim to bring full autonomy to barcode scanning, and that's the point. AI anticipates, filters, and assists, but humans remain the ultimate source of context and control.

In 2026, we’ll see another leap in algorithmic intelligence as Level 4 in AI barcode scanning becomes a reality. Systems won’t just decode barcodes efficiently, they’ll infer context and relationships. Is the code part of a product label or printed on a shipping box? Is the user rescanning the same code, or are there multiple identical codes in view?

This enhanced contextual understanding will unlock new use cases and opportunities we can barely imagine today. At Scandit, we’re already pushing this frontier with the Scandit SDK 8 and innovations like Smart Label Capture to bring that future within reach.

Automated ID verification will reach a tipping point

One in every 25 verification attempts involved fraud in 2025, and high-quality fake IDs are now available online for as little as $9.

Frontline workers are also verifying identity under increasing pressure. The 2026 World Cup is expected to create a surge in travel to the US, but experts doubt that U.S. travel infrastructure is prepared to handle the influx of travelers.

In 2025, manual identity verification showed clear signs of breaking under pressure. In 2026, we predict that powerful, AI-powered identity verification will become a strategic necessity that not only protects businesses but streamlines operations and opens up new revenue streams.

For example, Air France-KLM implemented document scanning using Scandit ID Bolt to allow passengers to pre-verify travel documents before they arrive at the airport.

This prevents passengers from being turned away at boarding gates, reducing delays, improving customer experience, and minimizing financial penalties. It also aligns with rising traveler expectations, with 78% now wanting to be able to use a single smart device to manage their journey.

AI in identity verification will also see the use of VLMs (vision-language models) to make this process even more streamlined in 2026.

Invitation letters, vaccination certificates, and approval PDFs emailed to travelers vary widely in layout and language. Using VLMs, relevant fields can be extracted and verified alongside standard identity documents, even if the document formats have never been seen before by either AI models or frontline workers.

Conclusion: Solving the world’s biggest challenges

In business and elsewhere, the biggest challenges ahead are not digital, but physical.

In 2026, the combination of physical and digital AI, hybrid device strategies, more robust AR, and the continued evolution of advanced data capture technology will accelerate rapidly. Together, these forces will finally make true digital transformation attainable for physical industries such as retail, logistics, manufacturing, and supply chain.