Sales intelligence engine

A specialist advisory firm sells to UK companies whose published accounts point to a need its service meets. The evidence is public, but it is buried in tens of thousands of filings, many of them scanned paper.

The sales team was guessing who to call. And salespeople stop trusting a tool the first time it gives them a wrong figure, so every number had to show where it came from.

An engine that reads public company filings and nothing else. Electronic accounts are read by rules. Scanned filings are read too, and marked as lower confidence. Where AI helps find a figure, it can only take one that appears word for word in the source, and a second, independent read checks it.

Each company is placed in a segment, scored from 0 to 100 and routed to senior staff or the phone team, with year-on-year changes flagged and a short call note drafted.

The team works from a portal with named logins: a call-these-first view, exports that show the source of every figure, and a weekly digest. It refreshes monthly, and every two weeks in the busy filing seasons.

The sales team starts each week from a ranked list with the evidence attached, instead of a guess.

From first conversation to accepted in production took about eight weeks, with acceptance tests run on every release.

The engine never decides that a company is a fit. It shows the evidence, and the firm's own people make every judgement and every call.

Estimates are always labelled as estimates, and no figure appears without its source.