Organisations often extract less value than they spend on AI.
AI today fails on three dimensions at once.
It hallucinates.
Existing AI tools fabricate citations. Every answer has to be manually verified - which destroys the time savings.
It's slow.
Existing AI tools take hours-to-days at scale - too slow to use during the matter itself.
It's expensive.
Variable token costs swamp the subscription. Hard to plan for, harder to justify the return.
"We don't even believe we would make our money back from the time savings and efficiency that these systems would give us."
Partner at a top European law firm.
Trust, speed, cost, sovereignty.
All from one architectural choice.
AI without hallucinations
your associates can now stop verifying
Sub-90-second retrieval
matter-speed, not coffee-break speed
Lower cost by architecture
structurally cheaper, not as a discount
Your data stays yours
deployable on-prem, never used to train the model
Where Formic is already live.
Four independent classes of buyer have moved from interest to commitment. Each one validates a different part of the architecture.
Federal regulators + defence bodies
Live across federal regulation and government modernisation programmes.
"Hallucination-proof citations with provenance back to source. This is the AI we needed."
~1/100
The energy footprint of generic AI.
Deterministic retrieval is significantly less compute-intensive than generative inference, providing ESG advantages to organisations whilst also saving the planet.
Frequently asked questions.
Formic is enterprise AI for regulated industries. The architecture is built on deterministic retrieval rather than generative inference — every answer is retrieved from your own documents, with citations that trace to the exact source page.
Other AI tools rely on generative inference and fabricate citations. Formic uses deterministic retrieval — answers are pulled from your actual documents, never generated. The architectural difference produces three durable advantages: no hallucinations by design, sub-90-second response at scale, and around 140× lower cost per query than legacy AI.
No — by architecture and by contract. Formic uses agnostic models that aren't fine-tuned per client; your documents power answers without being absorbed into model weights. This is also what makes hallucination prevention possible — where competitors bake training data into model weights (creating the conditions for fabrication), Formic simply retrieves from your records. Formic is also deployable on-prem when total data sovereignty is required.
Around 140× cheaper per query than legacy AI — approximately $0.08 per query against around $11 for the closest generative-LLM equivalent. Pricing is a flat licence per user with no token meter and no quarterly variability. The cost gap is structural — deterministic retrieval is significantly less compute-intensive than generative inference — not a discount that can be withdrawn or matched by incumbents.