OnDeck AI

What do they actually do

OnDeck AI runs a hosted video search and analysis service for recorded footage. Users type plain‑English questions (for example, “find every time vessel X approaches buoy Y” or “show when someone climbs the fence”) and the system returns matching clips or detections. The core value is zero‑shot usage: customers do not label data or wait for custom model training, and OnDeck says it does not train on customer data. Their site also emphasizes search at large scale across big archives (“petabytes”) and fine‑grained event outputs (OnDeck site).

Today, OnDeck is working with early customers and pilots in defense, robotics research, security/port monitoring, offshore oil & gas, and fisheries/conservation. Public references include a robotics research customer (Parametric) and pilots reported in defense contexts; the company reports analyzing thousands of hours of video in early deployments and reaching six‑figure ARR during/after a pivot from fisheries work (OnDeck site, YC page, CDL profile).

A typical workflow: customers point OnDeck at stored video archives, submit a natural‑language query, review returned clips/timestamps or export them for alerts/reports, and integrate findings into their operations (for example, threat reviews for ports/fleets, behavior labels for robotics, or fisheries monitoring outputs). Engagements are pilot‑led and directly onboarded via a demo/contact flow (OnDeck site, EM4Fish demo).

Under the hood, OnDeck applies modern vision‑language methods with retrieval/knowledge augmentation to interpret scenes and match natural‑language queries to video content without per‑customer training. The founders presented related work at a NeurIPS workshop focused on ocean monitoring, which underpins the product’s zero‑shot approach and roadmap (NeurIPS workshop listing, PDF).

Who are their target customer(s)

• Defense and port-security teams: Need to find and review specific events across massive video archives quickly for threat assessment, but cannot wait for custom model training and often cannot allow their data to be used for external training (OnDeck site, YC page).
• Robotics researchers and autonomy labs: Need fast, repeatable labels and behavior examples to iterate on controllers and experiments; manual annotation or building bespoke detectors slows research (OnDeck site).
• Offshore fleets and industrial operators (oil & gas, vessels): Operate continuous camera streams across heterogeneous platforms and need to detect incidents or anomalous operations without maintaining separate models per site (CDL profile, OnDeck site).
• Fisheries and conservation programs: Must monitor vast ocean footage to find rare behaviors or vessel interactions but lack resources to label long recordings or train specialized models (EM4Fish write‑up, NeurIPS workshop).
• Enterprise security and compliance teams with large CCTV archives: Require searchable, low‑latency access to past footage with enterprise controls for scale, reliability, and privacy before production rollout; they need infrastructure that handles petabyte‑scale indexing and search (OnDeck site, YC page).

How would they acquire their first 10, 50, and 100 customers

• First 10: Convert existing high‑trust pilots (e.g., defense, robotics, conservation) into paid, referenceable projects with 3–5 mission‑critical queries delivered quickly, strict privacy (“no training on customer data”), and a short proof‑of‑value tied to operational metrics (OnDeck site, CDL profile).
• First 50: Standardize a paid‑pilot package (scope, deliverables, pricing, data/security docs) to launch in weeks; use early case studies and “thousands of hours analyzed” to replicate wins in the same verticals; add vertical sellers and systems‑integrator/channel partners in defense, maritime, and robotics to scale outreach (OnDeck site, EM4Fish write‑up).
• First 100: Run a two‑track GTM: continue bespoke enterprise/defense pilots while offering packaged monitoring and a low‑friction trial for labs and smaller operators; back this with SLAs and security certifications, and expand integrations/partnerships with camera, VMS, fleet, and cloud vendors (OnDeck site, YC page).

What is the rough total addressable market

Top-down context:

The closest top‑down market for OnDeck’s current product is video analytics software & services, estimated at about USD 12.71B in 2024 by Grand View Research, with similar 2024 estimates (~USD 10B–13B) from other analysts (GVR video analytics, Verified Market Research). Broader adjacent infrastructure markets like video surveillance (cameras + VMS + analytics) are ~USD 73.75B in 2024 (GVR video surveillance).

Bottom-up calculation:

A practical bottom‑up view for the serviceable slice: assume ~10,000 organizations across defense/ports, maritime/offshore, industrial fleets, enterprise security, and research with large multi‑site archives could deploy language‑based video search at scale. At an average annual software/services spend of USD 250k–500k per org, this implies USD 2.5B–5.0B serviceable spend, consistent with a multi‑billion SAM within the broader USD ~12.7B video‑analytics TAM (GVR video analytics).

Assumptions:

• Focuses on enterprise/agency use cases with multi‑site or fleet‑scale archives (not consumer).
• Average annual spend includes software plus deployment/support for petabyte‑scale indexing and search.
• Overlap with VMS/analytics stacks is assumed; spend reflects add‑on analytics/search budgets rather than camera hardware.

Who are some of their notable competitors

• BriefCam (Canon): Established forensic video analytics and investigative search used by public safety and enterprises; overlaps on natural‑language/attribute search and rapid video review.
• Twelve Labs: Developer platform for semantic video understanding and text‑to‑video search; relevant for teams building custom video search pipelines rather than turnkey deployments.
• Ambient.ai: Enterprise physical security AI platform focused on real‑time threat detection and investigative search across CCTV, competing for security budgets and analytics seats.
• IronYun (Vaidio): Video analytics suite with 100+ AI functions for detection and search; widely pitched to security, government, and enterprises with existing VMS deployments.
• Verkada: Camera + VMS provider with people/vehicle search and analytics; competes where buyers prefer integrated hardware/software stacks rather than model‑agnostic analytics.