DTEK.ai Wants You to Stop Scanning and Just Put Your Groceries Down

There is a specific moment in a convenience store that almost nobody has ever enjoyed. It is 6pm, the queue is five deep, and the person at the front is turning a yoghurt pot over in their hands, hunting for the barcode. The scanner disagrees with them twice. Somewhere behind, a customer does the arithmetic on how much they actually want that yoghurt, puts it back on a shelf it does not belong to, and leaves. Every retailer in the world knows this happens. Very few of them can tell you how often.

DTEK.ai, a company of roughly 62 people headquartered in Dubai, has built its second act around that moment. Its flagship product, SWIFT, is an automated checkout kiosk that does not ask the shopper to scan anything. You set your items down. A camera looks at them. The system identifies what it is seeing - the company says in under 200 milliseconds - and hands you a total. Up to fifteen items, according to DTEK, in under thirty seconds. The barcode never enters the transaction.

In May 2026, ADNOC Distribution, the largest mobility and convenience retailer in the United Arab Emirates, announced it would roll SWIFT out across 50 Oasis by ADNOC stores beginning in Q2, following a pilot. The stated target was a reduction in average checkout time of more than 60 percent. For a company that most of the retail-technology world had not heard of three years ago, it is a conspicuous piece of validation.

The company that sold barcode scanners

The most interesting fact about DTEK.ai is what it used to be. In January 2021, three founders - Sanad Yaghi, Shadi Joulani and Ali Alsayegh - started a company called Dukkantek. The name is a compound of the Arabic dukkan, meaning shop, and the possessive: your shop. It sold cloud point-of-sale software to the community grocers of the Gulf, the small independent stores that sit under apartment blocks and stay open past midnight.

The product was not glamorous. It was a box with a receipt printer, a cash drawer, a customer display, a payment terminal and a barcode scanner, wrapped in software that handled inventory, VAT and margin reporting. Yaghi's original observation, given to The National in 2022, was blunt: "It was unclear to me why these stores were yet to have any kind of digital presence... I realised quickly that they would need an operating system to facilitate both online ordering and digital payments."

It worked. Dukkantek bootstrapped its first year, closed a $5.2 million seed round led by Global Founders Capital in October 2021, and followed it with a $10 million pre-Series A led by BECO Capital in August 2022, with Rocketship and Colle Capital participating. By mid-2022 the company reported thousands of merchants and operations in seven markets - the UAE, Saudi Arabia, Oman, Qatar, Kuwait, Bahrain and Turkey. BECO's Abdulaziz Shikh Al Sagha noted at the time that "launching 7 countries in 18 short months since founding is no easy feat."

For too long, the owners of small and medium-sized businesses have been left on the margins of the technology shift; now we are serving their unmet needs. Sanad Yaghi, Co-Founder & CEO - Forbes, May 2022

And then, around late 2023, they changed the company. Dukkantek became DTEK.ai. The POS platform receded. Computer vision took its place. Yaghi has framed the shift as a graduation rather than an abandonment: "Dukkantek POS served as an invaluable learning platform, uncovering critical insights into the needs of the market and the operational hurdles retailers face."

Read that with the installer's-eye view it comes from. Two years of walking into small shops, mounting hardware, taking support calls, watching the till at peak hour. What that teaches you is not that grocers need better reporting. It is that the counter is where the store loses money, and that the bottleneck is a physical task performed by a stranger under time pressure. The scanner requires the shopper to find a small printed rectangle and present it at the right angle. That is a poor job to delegate to a customer, and DTEK's conclusion was that no amount of software polish fixes it.

What SWIFT actually is

SWIFT is a kiosk. It sits on a counter a retailer already has. The customer places items down - loose, unarranged, without hunting for anything - and vision models identify multiple products simultaneously. Payment happens by card or phone. The company markets it as "barcode-free, multi-item product recognition," and positions it as the fastest checkout system for high-traffic retail rather than the cheapest or the most feature-complete.

Around it sits a second product, SIGHT, which is arguably the more durable business. SIGHT takes the CCTV cameras a store has already installed and paid for and turns them into operational analytics: footfall, dwell time, heat maps, queue dynamics, conversion, zone occupancy, and flags for compliance and security gaps. Its defining constraint is what it does not do. DTEK states that SIGHT runs on 100 percent non-biometric data. No face matching. Movement and flow only.

That is a deliberate ceiling on capability, and in regulated procurement it functions as a feature. A large enterprise buyer evaluating an in-store vision system has to route it past legal. "We do not collect biometric data" is a sentence that shortens that conversation considerably. On top of both products, DTEK sells configurable AI agents for loss prevention, automated reporting and event-triggered alerts, plus a decision-intelligence layer that stitches store data together for predictive staffing and stocking.

Company-reported performance claims - SWIFT

Reduction in customer waiting time70%

Increase in traffic throughput60%

Cut in average checkout time (ADNOC target)60%+

The retrofit argument

Every hardware company in physical retail eventually faces the same fork: retrofit the store, or replace it. Amazon answered with Just Walk Out - ceiling sensor arrays, gated entry, a whole-store rebuild. It is technically remarkable and commercially heavy. Conventional self-checkout from NCR Voyix, Diebold Nixdorf and Toshiba took the other extreme, keeping the barcode and simply moving the labour from a cashier to the customer, which is why so many people quietly hate it.

DTEK sits in a third position, alongside players like Mashgin in the United States and Tiliter in Australia: keep the counter, remove the barcode. This is less impressive as a demo and considerably easier to buy. A retrofit kiosk can be justified by a regional operations director with an existing budget line. A sensor ceiling needs a capital committee. In a category where the technology has largely been proven and the constraint is deployment, that difference matters more than accuracy benchmarks.

Three approaches to removing the queue

Approach	What the shopper does	What the retailer installs
Whole-store sensing Amazon Just Walk Out, Trigo, AiFi	Walks in, takes items, walks out	Ceiling sensors, gated entry, store rebuild
Vision kiosk DTEK.ai SWIFT, Mashgin, Tiliter	Sets items down, pays. No scanning	A kiosk on the existing counter
Classic self-checkout NCR Voyix, Diebold Nixdorf, Toshiba	Finds and scans each barcode	Lane hardware plus supervising staff

Who buys it, and how the money works

DTEK.ai sells B2B, and only one customer is publicly named: ADNOC Distribution. That is a thin roster by any conventional measure, and worth stating plainly rather than dressing up. But it is a particular kind of thin. ADNOC Distribution operates the UAE's largest fuel and convenience network and says it runs more than twenty AI initiatives across its value chain, from fuel demand forecasting to supply chain optimisation. Its CEO, Eng. Bader Saeed Al Lamki, described the DTEK partnership as marking "a key moment as we accelerate our transformation into an AI-native mobility and convenience leader."

The commercial model has inverted from the Dukkantek era, and that inversion is the strategy. The old business charged small merchants a low flat annual fee - reported at roughly $400 - plus payment gateway commissions, and needed volume in the thousands to work. The new business sells a bundled hardware-and-software system with white-glove integration into a retailer's existing POS, expanding through additional store locations and through attaching SIGHT and custom agents to a deployment that already exists. Fewer customers, each worth vastly more, each taking far longer to close. Current pricing is not public.

There is also a policy tailwind here that would be easy to miss from outside the region. The ADNOC partnership was announced at Make it in the Emirates, an event organised around in-country value and local manufacturing. A company of 62 people got access to a national retail network partly because it was built where the network is. That is not a coincidence any founder in the Gulf would consider accidental.

Where the expertise sits

DTEK's technical claim is item recognition speed and multi-item handling. Its less obvious asset is domain knowledge purchased the slow way. The founding team did not arrive at retail from a research lab. They ran a hardware-and-software business inside thousands of small shops across seven markets before they wrote a line of vision code for a kiosk. They know what a store manager will actually tolerate at 6pm, which is a narrower set of things than a demo suggests.

Yaghi has described the internal culture in terms that fit that history. Speaking to Entrepreneur Middle East after the 2022 raise, he said what worked was "having high levels of accountability, demanding criticism, and acknowledging and owning our flaws," and that "bootstrapping our pre-seed period was critical to the overall success that we've had to date."

We saw a clear opportunity in the market where checkout technologies had not evolved to match the pace of today's shoppers. Sanad Yaghi - on the ADNOC Distribution rollout, May 2026

The open questions are the ones the public record cannot answer. DTEK has not raised a disclosed round since August 2022, which for a hardware-inflected AI company nearly four years on is notable either as discipline or as constraint. Headcount appears to have contracted substantially from the 140 reported in the Dukkantek era to around 62 today, consistent with a pivot that shed a support-heavy small-business operation. No revenue figure or valuation is public. And a rollout announced is not a rollout completed: the meaningful test of the ADNOC agreement will be whether 50 stores becomes the rest of the network.

The ten-year question

Barcodes are half a century old and extremely good at their job, which is why every attempt to remove them has failed on economics rather than on technology. What has changed is that vision models got cheap enough to run on a countertop box, and that the labour maths in convenience retail got worse. DTEK.ai is one of perhaps a dozen companies worldwide positioned for that specific window, and the only one with an anchor deployment inside a Gulf national retail network.

If the ADNOC rollout holds, the interesting outcome is not that a startup won a logo. It is that a category which has spent a decade in pilot purgatory finally produced a reference deployment at meaningful scale in a market that other operators watch closely. If it does not hold, DTEK will have learned - again, expensively - something about what retailers will tolerate. Given the founders' track record, that is not the worst position to be in. They have rebuilt this company once already.