From spreadsheet to sequence: an AI lead engine for Inform Holdings

Inform Holdings, part of the CPMAS Group, advises UK businesses on business rates. A large slice of their pipeline comes from deadline leads: ratepayer cases tied to specific 2023 Rating List entries where the window to act is closing. The leads arrive as spreadsheet exports, hundreds of rows of company names, addresses and case types, with no contact details and no obvious owner inside the sales team.

The problem: good leads, slow hands

Before this build, every batch meant the same manual slog. Someone had to look up each company, find a decision maker, work out which account manager should own it (the team has specialists by sector and by incumbent rating agent), then write an email worth sending. With hundreds of rows per batch and a hard deadline attached to every case, the maths never worked. Leads aged, allocation was inconsistent, and outreach defaulted to generic templates because personalised emails at that volume were not feasible by hand.

What we built

A campaign builder and AI outreach engine: a React and TypeScript front end on Vercel, an Express API on Render, and Supabase as the datastore, integrated with Apollo for enrichment and sending and with an LLM layer for research and drafting.

The pipeline runs in stages:

• CSV ingestion and qualification. The tool parses each batch, normalises messy headers and lead-type values, and qualifies rows against the active deadline lead types so dead rows never reach enrichment.
• Apollo enrichment. Qualified companies are matched in Apollo and decision-maker contacts are pulled in, turning a name-and-address row into a reachable prospect.
• AI research per prospect. For each contact, the engine gathers context on the company and its situation, so the email that follows has something real to say rather than a mail-merged greeting.
• 1-1 email generation. Every email is generated individually from that research. No shared template, no token-swap personalisation.
• Routing per salesperson. A configurable rules layer maps each lead to the right account manager: sector specialists, rating-agent specialists, and caseload allocation, with each account manager tied to their own exact Apollo sequence so outreach sends from the right person's mailbox under their name.

The routing rules live in configuration, not code paths scattered through the app. When the team changed, account managers leaving, a sequence moving to a different sender, the fix was a config update, not a rebuild.