Loading page…
Turn existing CCTV feeds into pedestrian and vehicle analytics for property owners and advertisers, priced per camera.
roughly 50 paid cameras across 10 customers; most cameras remained in pilots
Sell a focused outcome on infrastructure customers already own: Fima sends existing CCTV streams to its computer vision service instead of asking buyers to install new hardware.
Fewer bars = easier, cheaper, or faster for an AI-assisted solo builder. Editorial judgments based on the case details.
Fima.ai packages computer vision as a no-hardware service for organizations that need to understand movement in physical spaces. The founders first saw the pain while delivering a city traffic measurement project through an AI consultancy: sourcing data science help, collecting and labeling data, training a model, deploying infrastructure, and visualizing results took six months. They say Fima can get a customer gathering data in about 20 seconds by connecting an existing CCTV feed. The initial focus became commercial real estate and out-of-home advertising, where operators can count pedestrians and vehicles around malls, streets, parking lots, or billboards. The founder described charging €200 per camera per month. At the interview, the company had about 10–15 customers and around 50 paid cameras, though many installations were still pilots; the founder estimated revenue at about $6,000 per month. It had raised a $250,000 pre-seed in 2019 and a €1.55 million seed round the prior December. The team numbered 12–13, with cofounder Karen Burns as CEO and the guest as CTO.
For Fima, the first customers are best found where a camera already exists and someone has a specific measurement decision to make. The initial wedge came from a consultancy project measuring cars on city streets; that exposed a six-month process the product could compress.
The quick-reference card — acquisition channels, replication playbook, risks & traps, and our verdict — is members-only.
Join the Unicorn Club — $5/moCancel anytime, in one clickPaste into Claude Code / Codex and get a working version of this product end-to-end.
Two production-grade prompts per idea — a full build spec and an SEO growth plan — plus 10 tailored build specs a week. Members only.
Join the Unicorn Club — $5/moCancel anytime, in one clickEvery claim traces back to a listed source — open any of them and check the number yourself.
Data credibility: Founder interview hosted by Nathan Latka; the guest reported about $6,000 monthly revenue, roughly 50 paid cameras across 10 customers, and 10–15 customers overall. The host’s arithmetic used the founder’s €200 per-camera monthly price to illustrate potential account value. The revenue estimate was not supported by a dashboard or third-party records in the transcript. The guest said most cameras remained in pilot phase. Customer counts, team size, fundraising amounts, and dataset size are founder-reported and unaudited.
The product’s first commercial dollar should follow a live camera test that answers a buyer’s operational question. Fima charges by camera, with the founder citing €200 per camera each month.
Fima’s compounding loop comes from persistent measurement and a product that improves as more camera footage is processed. Once a property team uses the data to compare traffic across seasons, layouts, or operating experiments, it has a reason to keep the cameras connected and build a longer history.
Enterprise computer vision has a long quiet stretch between a promising demo and dependable recurring revenue. Fima’s own story shows why: it had no paying customers about a year before the interview, then reached pilots and an estimated $6,000 monthly revenue, while many cameras were still not fully converted.
Fima takes a video stream from a customer’s existing CCTV cameras, processes it with computer vision, and returns structured movement data. The founder named pedestrian and vehicle detection as core examples and positioned the system for commercial real estate and out-of-home advertising. A mall operator can use traffic counts to evaluate changes around a parking lot or food-truck area; an advertiser can use camera views to understand movement around billboards. The product’s main promise is operational simplicity: no new camera hardware and, according to the founder, no integration fee, with setup described as a few button presses. That promise is only valuable if access, privacy, and camera compatibility are handled cleanly in the customer’s environment. The company started with open data such as COCO and built its own image dataset, which the founder estimated at 200,000–300,000 images. Its service is not a generic vision model sold to everyone; the stronger product shape is a repeatable analytics workflow for a specific type of physical site. Keep reporting definitions stable so customers can compare locations and time periods.
The core pricing unit is a camera per month. The founder quoted €200 per camera monthly and said a typical brand had around 20 cameras onboard, although deployments ranged from one to 50. A 20-camera account at that quoted rate represents €4,000 in monthly list-price value, but the interview’s reported revenue was about $6,000 per month because most cameras were still in pilot phase. Those figures should not be conflated: one is an illustrative contract calculation, the other is the founder’s estimate of current revenue. The flexibility to turn cameras on and off can help buyers fund experiments, such as measuring a temporary street redesign, but daily variable usage can make revenue less predictable. A durable commercial model can pair a site subscription or minimum monthly commitment with a defined allowance for camera feeds, then charge for additional views or locations. Use paid pilots to establish measurable value, but set an explicit conversion path so pilot pricing does not become permanent. Revenue quality depends on recurring camera use and customer renewal, not simply the number of feeds connected during a trial.
The service avoids installing proprietary camera hardware, but it still carries meaningful compute and data costs. The founder said a large share of fundraising went toward development and lowering the cost of running GPUs, machine learning models, and neural networks. That is the central unit-economics problem: a €200 monthly camera price must cover video ingestion, inference, storage, support, and the work required to deploy reliably across customer systems. Measure gross margin by active camera and by use case, since continuous feeds and occasional experiments can have very different compute profiles. The team also employed software developers, data scientists, and data labelers; the founder described roughly four software engineers and three or four data science staff within a 12–13-person company. That is substantial staffing for a product with around $6,000 estimated monthly revenue at the time discussed, supported by two funding rounds. A leaner entrant should prove a narrow camera workflow with sampled or event-triggered processing before promising continuous analytics. Track inference cost per camera-hour, onboarding time, labeling burden, and support hours alongside revenue.
Copy Fima’s move from custom consulting pain to a repeatable product in a constrained environment. The six-month city project exposed a concrete bottleneck: each bespoke vision deployment required data, model work, infrastructure, and visualization. Turning existing CCTV into a fast path to an answer is a sharper promise than selling “AI computer vision” in the abstract. Also copy the willingness to choose a few buyer groups rather than claiming the model works everywhere; the founder said the company explored different verticals before focusing on property and advertising. Change the sales motion by setting stronger pilot boundaries. A test should name the camera feeds, metric, delivery schedule, privacy approvals, and commercial conversion decision before work begins. Since Fima’s interview describes many cameras in pilot and about $6,000 monthly revenue, a new builder should optimize for paid conversion and repeat use, not inflated pilot counts. Start with one narrow metric—such as pedestrian counts around a defined site—and earn expansion through accuracy and a decision the customer can point to. Build for the customer’s existing camera estate, but be explicit about compatibility and data handling.
The first risk is enterprise adoption speed. The founder described long sales cycles, pilots, and a period with no paying customers, so a technically successful trial can still consume months without becoming a recurring contract. The second is unit economics: continuous video processing can make GPU costs and support expensive, and the founder specifically said funding was used to reduce model-running costs. The third is deployment reliability. Existing CCTV simplifies hardware procurement, but feeds vary in format, image quality, network access, and operating conditions; poor footage undermines detections and buyer trust. Privacy and security are also central because camera streams can capture identifiable people, so customers will expect careful access controls, retention limits, and documented data practices. Finally, the per-camera model can be volatile when customers switch cameras on for experiments and off afterward. Establish minimum contract terms and distinguish temporary measurement projects from ongoing analytics. The founder reported only planned short-term project churn at the time, but that is not evidence that every pilot becomes durable. The business works when accuracy, deployment, and customer decision value all survive outside a demo.
The four founder playbooks — first customers, first dollar, the flywheel, the silent stretch — are in the Unicorn Club, on this idea and on all 1,403 others. The story, the numbers, the sources and every later section stay open to everyone.
Join the Unicorn Club — $5/moCancel anytime, in one click