FormalInsight resolves CRM, ERP, finance, support, and email into one live, queryable ontology — every customer, invoice, ticket, and promise on a single timeline. Critical operational risks are invisible to isolated systems. On the FormalInsight ontology, they become a few queries.
The account that churns next quarter looks healthy in every system it touches. The risk exists only in the joins — a recurring defect, then an unresponsive contact, then a declining usage trend. No isolated system can see it, because no isolated system holds it.
Renewal forecast: Commit. Account health 82. ✓
Receivables current. No disputes on record. ✓
All tickets resolved within SLA. CSAT 4.6. ✓
All threads answered. No open escalations. ✓
The resolved tickets trace to a single recurring defect · product usage −38% since the last workaround · invoicing covers two-thirds of provisioned seats · the primary contact unresponsive for 32 days · a renewal at risk — knowable today.
Every system, in isolation, reports green. The dashboards are not wrong. They are blind by construction.
What no isolated system can see, the ontology makes queryable.
Not another dashboard over your data — a formal model of it. Records from every system resolve into typed entities, documents, and events with defined relations. The schema is a first-class API object: a caller (human or AI agent) can ask what exists before asking what happened.
org · person · invoice · order · delivery · ticket · deduction · subscription — one node per real-world thing, however many systems it appears in.
crm:account_id ≡ erp:kunnr ≡ domain ≡ legal name — entity resolution across keys, so “Bluegrass Wholesale Distributors” is one node, not five.
invoice_issued · payment_received · ticket_opened · order_booked · dunning_final_notice · opportunity_won … every event links to the entities it touches and to the raw source record behind it.
every fact carries valid_ts (when it happened) and known_ts (when it became knowable). The ontology can answer as of any date — including “what was knowable the day before the loss.”
Query the company, not the systems. The same calls power a one-line lookup, a portfolio-wide audit, and an autonomous investigation — from whatever your team already uses.
Plain HTTP. Scripts, notebooks, BI tools, spreadsheets, schedulers — anything that speaks HTTP can query the ontology.
The same surface ships as MCP tools. AI agents read the schema, learn the shape of your business, and compose their own multi-step questions.
Entities, timelines, aggregates, search — small orthogonal calls that compose into arbitrarily deep questions, not fixed reports.
as_of on every call. Ask as of today, last quarter's close, or the day before a decision — the same API, replayable.
Complex decisions are multi-step questions. An agent walks the graph the way an analyst would — across every system at once — and returns an answer your team can work on.
Reads the schema and learns the shape of your business — entities, relations, event vocabulary — before asking a single question.
Chains queries into multi-step reasoning — entity to timeline to source record — following the joins no isolated system holds.
Assists complex, critical decisions: renewals, credit, escalations, supplier changes. Your team decides.
From two synthetic company datasets, generated to mirror real enterprise systems and run through the ontology. Every number below is quoted verbatim from our experimental operational risk detector runs on the synthetic demo dataset — measured on the records, not estimated.
SYNTHETIC DEMO DATASETOne order line in ten carries an identical 8.0% discount that appears in no pricing condition record — uniform across 856 accounts, every month, every channel.
Seats provisioned and in active use that were never invoiced, across 52 accounts — including one contract reduced to 113 seats on paper while 133 remained live.
Accounts whose listed champion lost their product seat and sent no email since — while the CRM still records the relationship as intact. One renewal was five days out.
None of this required new data. The records already existed in the source systems — unjoined, and therefore unknown. On the ontology, each finding is a few queries.
A limited set of read-only exports, shared under NDA, is sufficient to construct your ontology. We handle the resolution end to end; your team evaluates what it returns. No integration effort, no ongoing obligation.
Operational data from the systems you choose, shared under NDA.
We resolve entities across systems and place every event on one timeline — no integration work, no effort from your engineering team.
Your ontology behind the API — with findings from AI agents specialized in operational soundness. Grade them with your team, then decide what's next.