The archive · AI & Models · Product decision · 2024–2026
Lexius bets AI on stores' existing cameras will win loss prevention; YC W26
Lexius turns cameras stores already own into AI guards that flag shoplifting and falls in real time, with no new hardware.
Lexius
What the business is
Lexius sells software that turns existing security-camera systems into AI guards: it detects and documents shoplifting, falls and other incidents in real time, makes months of footage searchable, and sends alerts to phones, with no new hardware.
How it started
Lexius was founded in 2024 by David Elskamp and Liam Webster in San Francisco. Elskamp, who studied computer science at the University of Twente and holds a Stanford master's, wrote his bachelor's thesis on recognizing crimes in cities; Webster did research at UC Berkeley's ICSI. Their premise: post-pandemic shoplifting had become one of retail's fastest-growing loss problems - Dutch chain Jumbo alone loses around €100M in revenue a year to it, per an interview with Elskamp - while stores already owned cameras whose footage nobody had time to review.
What happened
The team went through Y Combinator's Winter 2026 batch (Demo Day March 24, 2026), and TechCrunch named Lexius one of the 16 most interesting startups in the cohort on March 26. An interview with Elskamp described a waiting list that 'exploded' without marketing, covering retailers with more than 4,000 stores under management, and emphasized a behavior-only model that ignores skin color, race, gender and clothing. By August 2026 the company's YC launch post said it was trusted by 7-Eleven, Erewhon and Prada, with Erewhon using the product across its stores.
No ending yet — it is still running.
Background
Lexius is a San Francisco startup, founded in 2024 by David Elskamp and Liam Webster, that sells an AI layer for security cameras stores already own. Its product detects and documents incidents such as shoplifting and slip-and-falls in real time, makes months of footage searchable, tracks repeat visitors across cameras, and packages evidence into case files - with no new hardware and setup measured in minutes rather than weeks.
The bet is that the installed base, not the technology, defines the market. Millions of locations have cameras with no intelligence; replacing a system costs $50,000 to $100,000 per site, and legacy CCTV is messy enough that most AI breaks on it. Lexius claims the hard part is not the model but making unreliable, fragmented camera systems work at scale with accuracy high enough for real decisions, and that its software layer wins precisely because it skips the rip-and-replace.
The company also bet on a privacy-safe design. In a Next Icons interview, Elskamp said the model observes movement and behavior only and deliberately ignores skin color, race, gender and clothing, arguing that older rivals' models trigger false alarms and carry profiling baggage. That positioning, plus an international team with ties to UC Berkeley, MIT and Stanford, helped it fill a waiting list - retailers with more than 4,000 stores under management, per the interview - before a broad launch.
After YC's Winter 2026 Demo Day on March 24, 2026, TechCrunch named Lexius among the 16 most interesting startups in the batch. By August 2026 the company's launch post said 7-Eleven, Erewhon and Prada were customers. The open questions are accuracy at scale on low-quality CCTV and whether real-world deployments sustain the privacy claims.
What has to be true
- Post-pandemic shoplifting is a measured, urgent problem - Dutch chain Jumbo alone loses about €100M in revenue a year - and stores already own the cameras; Lexius sells the missing intelligence layer.
- Rip-and-replace is the incumbent's cost problem: $50K–$100K per site makes retrofitting the only affordable path for chains.
- Behavior-only detection turns the industry's biggest liability - bias and privacy scrutiny - into a differentiator rather than a defense.
- A waitlist covering 4,000+ stores before a marketing push is evidence the pain, not the pitch, is selling the product.
- Landing named brands like 7-Eleven, Erewhon and Prada (company-reported by August 2026) gives it reference customers in a trust-heavy market.
What can be applied
When the hardware is already installed everywhere, the value moves to the software layer on top; and in surveillance, baking the privacy and bias answer into the product is part of the go-to-market.
Aftermath
As of September 4, 2026, Lexius is active with a three-person team in San Francisco: it finished Y Combinator's Winter 2026 batch, features in TechCrunch's March 2026 list of the cohort's most interesting startups, and its YC launch post says 7-Eleven, Erewhon and Prada use the product, with Erewhon deploying it across stores. It still faces a crowded field of loss-prevention AI vendors and must prove its accuracy and privacy claims hold on real, low-quality CCTV footage at chain scale.
Sources
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card