The Data layer: connecting CRM, PMS and OMS to your AI OS.
The most expensive part of any AI project is the part nobody wants to do. Connecting the firm's own systems is where the OS earns its keep.
AI is not a data project. But an AI OS without a data layer is a demo. The Data layer is the plumbing that connects the firm's own systems — CRM, PMS, OMS, ticketing, comms, docs, filings, calls — to the layers above. Do it right and every skill and agent gets grounded in reality. Do it badly and the OS hallucinates its way through the firm.
What to connect, in order
- The system with the highest signal density — usually the CRM for a brokerage/wealth firm, the PMS or OMS for a fund.
- The document estate — research notes, deal memos, KYC files, credit committee minutes.
- The comms estate — email, chat, call transcripts, meeting recordings, where consent and jurisdiction allow.
- The external data — filings, transcripts, sanctions lists, market data — usually easiest and often already licensed.
Why we do not migrate
Every AI vendor with a data ambition pitches a migration. Firms should refuse. The Data layer reads from source systems in place through connectors and only mirrors what needs to be indexed. The system of record does not move. That keeps compliance simple, keeps the vendors of the source systems whole, and keeps the AI OS decoupled from any single source of truth.
Do not migrate. Connect. The firm's system of record is not the AI's problem to solve.
Where to start
Every KJ Capital engagement begins with a two-week £15k AI Diagnostic. We map your existing data estate, model your firm's OS across the six layers above, cost the build, name the first three skills to ship, and hand you a plan your board can act on. It is deliberately priced to be a rounding error against the value it unlocks.
FAQ
Which systems do you already have connectors for?
The obvious ones — Salesforce, HubSpot, Microsoft 365, Google Workspace, common PMS/OMS vendors, S3, SharePoint. New connectors are one to two weeks of build.
What about legacy systems with no API?
SFTP, scheduled exports, RPA readers. Every legacy stack can be connected; the question is only how quickly.
How do you handle personal data and consent?
Consent metadata comes through with the record. The OS respects it at retrieval time. Nothing is used for training without an explicit policy decision.
Start with the £15k AI Diagnostic
Two weeks, one costed OS blueprint, three named skills to ship first.