{"slug": "startup-spotlight-lexius-makes-checkout-cameras-count-what-never-got-scanned", "title": "Startup Spotlight: Lexius makes checkout cameras count what never got scanned", "summary": "Lexius, a San Francisco startup founded by David Elskamp and Liam Webster and listed in Y Combinator's Winter 2026 batch, has narrowed its camera-AI pitch to checkout loss software that matches existing store cameras against point-of-sale register logs to flag items that cross the scan window without a matching scan. The system prompts the shopper or cashier to rescan the item, saves the clip alongside the transaction, and ranks missed scans by cashier against the store average. Elskamp, a two-time founder who wrote a University of Twente bachelor's thesis classifying 13 categories of crime in surveillance footage, and Webster, the CTO, are betting the retrofit approach wins on accuracy and independently verified deployment scale.", "body_md": "# Startup Spotlight: Lexius makes checkout cameras count what never got scanned\n\n**David Elskamp and Liam Webster are narrowing a broad camera-AI pitch into checkout loss software that pairs existing cameras with register logs.**\n\n        By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)\n        · Published \n\nPrimary source: [Y Combinator](https://www.ycombinator.com/companies/lexius)\n\n## Why it matters\n\nLexius is testing a capital-efficient route into physical AI: sell software against cameras and registers retailers already own, then prove value one missed scan at a time. Its checkout focus gives the founders a measurable wedge, but accuracy and independently verified deployment scale will decide whether the retrofit advantage holds.\n\n[Lexius](https://www.lexius.ai/?ref=runtimewire) co-founders [David Elskamp](https://www.linkedin.com/in/davidelskamp/?ref=runtimewire) and [Liam Webster](https://www.linkedin.com/in/liamwebsterxyz/?ref=runtimewire) are turning the security cameras retailers already own into checkout auditors, matching what passes through a lane against what appears in the register log.\n\n[Y Combinator lists Lexius in San Francisco](https://www.ycombinator.com/companies/lexius?ref=runtimewire) and included the company in its Winter 2026 batch. Lexius entered the accelerator with a broader promise: make old corporate cameras useful enough to detect shoplifting and falls, search months of footage, trace people between cameras and assemble incident files. The founders laid out that pitch in their [original YC launch post](https://www.ycombinator.com/launches/PYc-lexius-ai-for-any-security-camera?ref=runtimewire), which named warehouses, hotels, offices, campuses, casinos, law enforcement and retail among the target markets.\n\nLexius's current homepage is much more specific about its initial wedge: missed scans at staffed lanes and self-checkout kiosks. That narrowing is the most consequential choice Elskamp and Webster have made so far. Checkout loss gives Lexius a defined buyer, a measurable transaction and a moment when its software can intervene before the goods leave the store.\n\nLexius says its system can flag an item crossing the scan window without a matching register event, prompt the shopper or cashier to rescan it, and save the clip alongside the transaction. It also ranks missed scans by cashier against the store average. The product connects two records retailers already generate: the camera feed and the point-of-sale log.\n\n### A founder who started with cameras\n\n[Y Combinator describes Elskamp](https://www.ycombinator.com/companies/lexius?ref=runtimewire) as a two-time founder who started a media company at 14, grew it to six figures and ran it for eight years while working with global brands. That gave him an early education in cameras as a business tool, followed by formal training in how machines interpret their output.\n\nWhile studying Technical Computer Science at the University of Twente, Elskamp wrote a [bachelor's thesis on classifying 13 categories of crime in surveillance footage](https://essay.utwente.nl/98330/1/Elskamp_BA_EEMCS.pdf?ref=runtimewire). [Elskamp wrote in a 2022 LinkedIn post about the research](https://www.linkedin.com/posts/davidelskamp_technology-artificialintelligence-ai-activity-6964221328006492160-syge?ref=runtimewire) that the work could eventually support automatic crime detection from camera feeds. YC says he later completed bachelor's and master's degrees in computer science, with a focus on computer vision and AI systems for real-world environments.\n\nThe thesis established a straight line from Elskamp's research to Lexius. His earlier work asked whether computer vision could identify an event in surveillance footage. Lexius applies that research to live retail transactions, where a detection can prompt a rescan or become part of an incident record.\n\nWebster, Lexius's CTO, brings a different part of the technical stack. [YC identifies him](https://www.ycombinator.com/companies/lexius?ref=runtimewire) as a UC Berkeley electrical engineering and computer sciences graduate and an AI researcher at the International Computer Science Institute, where his work included machine learning and privacy. His public project history includes software for analyzing mobile-app traffic and comparing it with privacy policies. That background bears directly on software intended to inspect shoppers and employees in real time, where data handling can matter as much as detection performance.\n\n### The checkout is a narrower and harder test\n\nLexius says its [checkout product works with existing IP cameras](https://www.lexius.ai/checkout-loss-prevention?ref=runtimewire) and a read-only connection to the transaction log, without new hardware or register downtime.\n\nLexius lists six forms of checkout loss: skip-scans, sweethearting by cashiers, expensive items entered as cheaper produce, products left at the bottom of a basket, switched barcodes and void abuse. According to Lexius, its software compares visual events with scanned items, prices and void activity, then attaches the relevant video to the transaction.\n\nThat design gives the model more context than video alone. A camera may see a steak move across a self-checkout area while the transaction log records bananas. The mismatch gives store operators a concrete action: prompt a rescan, review a clip or examine repeated discrepancies at one lane.\n\n[Lexius says stores can go live in about an hour](https://www.lexius.ai/checkout-loss-prevention?ref=runtimewire). Its [homepage displays connections](https://www.lexius.ai/?ref=runtimewire) to NCR Voyix, Verifone, Toshiba, Square and Toast point-of-sale systems. Each additional register system still creates integration and support work, and camera placement will determine whether the software can see the scan window and bagging area clearly enough to make a decision.\n\nA false positive at checkout is not an abstract benchmark error. It can interrupt a shopper or place an incident on an employee's scorecard. Lexius's ability to sell beyond pilots will depend on the precision of those interventions, how quickly store workers can dismiss mistakes and whether the system performs consistently across locations. Lexius has not published an independently tested accuracy rate for the checkout product.\n\n### Customer logos are the proof point Lexius still has to earn\n\nLexius says its [current customer references](https://www.lexius.ai/?ref=runtimewire) include Pak'nSave, Piggly Wiggly, Erewhon, OnCue and Prada, though the supplied material does not independently establish the scope of each deployment. Its YC company page and launch post previously said Lexius was trusted by 7-Eleven, Erewhon and Prada, with Erewhon using the product across its stores.\n\nThose are Lexius and YC assertions rather than independently documented deployment records. Lexius also publishes comments attributed to store managers and loss-prevention workers. On its homepage, Lexius says that during the first week of a deployment across five stores, its software flagged previously unseen self-checkout theft, including one case that became a fraud investigation exceeding $5,000. Another statement, labeled Erewhon Los Angeles, says a free trial in three stores led to a chain-wide rollout. Lexius does not provide named executives, measurement methods or full deployment counts alongside those claims.\n\nThe distinction matters because a customer logo can represent anything from a limited trial to a paid deployment across an entire estate. The scale and economics of Lexius's cited deployments remain company-reported.\n\nLexius [does not publish a price list](https://www.lexius.ai/checkout-loss-prevention?ref=runtimewire), leaving the product's economics difficult to compare from the outside. The available primary materials do not disclose revenue, retention, total customer count or a confirmed funding total. Y Combinator is the only named backer supported by the supplied research; no other investors or valuation have been confirmed.\n\nThe product will ultimately be judged against recovered merchandise, reduced review time and the labor required to manage alerts. Company-reported savings and return-on-investment claims may help open sales conversations, but retailers evaluating a rollout will need results measured across their own stores.\n\n### Existing-camera AI already has established vendors\n\nOn its website, [Coram AI](https://www.coram.ai/?ref=runtimewire), a physical-security software company, says its platform works with existing IP cameras and combines video search, alerts, access control and emergency-response functions. [Lumana](https://www.lumana.ai/?ref=runtimewire), another AI video-security vendor, says its software works with existing cameras and is used across more than 50,000 cameras.\n\n[Everseen](https://everseen.com/solutions/evercheck?ref=runtimewire) is a closer comparison: its Evercheck product is marketed for checkout loss prevention. [Veesion](https://veesion.io/en/?ref=runtimewire) markets AI-powered theft prevention based on camera footage.\n\nThose companies show that retailers already understand the category. Lexius does not need to establish that cameras can do more than record video. It has to win on deployment speed, accuracy, workflow design and cost against vendors with larger reported footprints.\n\nThe current checkout focus is a credible response to that pressure. Lexius's earlier YC pitch covered warehouses, hotels, campuses, casinos, offices, retail locations and law enforcement. Each environment has different incidents, buyers and thresholds for mistakes. Checkout lanes produce a structured event stream and a direct financial outcome. A missed scan has a receipt, an item and a value.\n\nThat wedge also leaves room for expansion. Once Lexius is connected to cameras and transaction logs, the same infrastructure could support cashier coaching, operational audits and broader store analytics. Elskamp's original ambition was larger, with the [YC profile](https://www.ycombinator.com/companies/lexius?ref=runtimewire) describing Lexius as an eventual operating system for physical spaces. The checkout product gives customers a starting point where they can calculate the result.\n\n### The retrofit bet\n\n[Winter 2026 Demo Day took place on March 24th, 2026](https://www.ycombinator.com/blog/2026-demo-days/?ref=runtimewire). Lexius is an accelerator-stage company refining its commercial wedge months after presenting with its YC batch.\n\nElskamp and Webster are betting that physical AI will spread through hardware already hanging from store ceilings. YC says replacing camera systems can cost $50,000 to $100,000 per site, giving retailers a clear reason to consider software that works with installed equipment. The retrofit approach also makes compatibility work central to the product. Old cameras, inconsistent networks and varied register systems are the business.\n\nLexius's [homepage describes the product as \"built for the lane, not the lab\"](https://www.lexius.ai/?ref=runtimewire). The lab portion is already the founders' strongest credential: Elskamp studied crime detection from surveillance video, while Webster worked on machine learning and privacy. Their next test is operational. The founders have to turn those models into software store managers trust while a basket, a customer and a cashier are waiting for an answer.", "url": "https://wpnews.pro/news/startup-spotlight-lexius-makes-checkout-cameras-count-what-never-got-scanned", "canonical_source": "https://runtimewire.com/article/startup-spotlight-lexius-checkout-camera-ai-david-elskamp", "published_at": "2026-09-15 16:18:14+00:00", "updated_at": "2026-09-15 16:51:18.202997+00:00", "lang": "en", "topics": ["computer-vision", "ai-startups", "ai-products", "artificial-intelligence"], "entities": ["Lexius", "David Elskamp", "Liam Webster", "Y Combinator", "University of Twente", "San Francisco"], "alternates": {"html": "https://wpnews.pro/news/startup-spotlight-lexius-makes-checkout-cameras-count-what-never-got-scanned", "markdown": "https://wpnews.pro/news/startup-spotlight-lexius-makes-checkout-cameras-count-what-never-got-scanned.md", "text": "https://wpnews.pro/news/startup-spotlight-lexius-makes-checkout-cameras-count-what-never-got-scanned.txt", "jsonld": "https://wpnews.pro/news/startup-spotlight-lexius-makes-checkout-cameras-count-what-never-got-scanned.jsonld"}}