A service they couldn't offer at all, now running end to end.

Every consultancy that sells analysis has the same ceiling: the analysis takes analyst hours, and analyst hours are finite. For 80/20, the prospect deck was the sales asset — it showed a company what they were leaving on the table, and it required real work to produce. Data collection, analysis, and building the presentation all ran through their analytics team.

That math works for large prospects. It doesn't work for small and mid-sized manufacturers and distributors, where the deal size doesn't justify the analyst time. So that segment simply wasn't served — not because the opportunity wasn't there, but because there was no way to reach it that made economic sense.

An end-to-end pipeline. A prospect's profit and loss statements and transaction data go in; a completed PowerPoint arrives in the client's email, with the analysis done and the upside scoped.

The system does two different kinds of work. The financial analysis is deterministic — it had to reproduce 80/20's existing methodology exactly, because the numbers in the deck are the argument. The commentary and insight layer is generative, and it had to reflect 80/20's own philosophy about what matters in a company's cost and pricing structure, not a generic read of the numbers.

The system runs on Google Cloud — Cloud Storage for the uploaded financials and transactions, Cloud Run functions for the processing, and Firestore for state and results. A prospect's P&L and transaction data goes in, gets validated, gets analyzed against 80/20's methodology, and comes back as a finished PowerPoint in the client's inbox. The whole path takes five to ten minutes.

Serverless was the right call for the shape of this workload. Prospect analyses don't arrive on a schedule — they arrive in bursts when someone's working a list, and then nothing for a week. Cloud Run means 80/20 pays for the analyses they run rather than for capacity sitting idle between them, which matters when the entire point of the system is that a segment becomes economical to serve.

Two things, and they pulled in opposite directions.

The math had to align deterministically with the process 80/20's analysts already ran. Not approximately — exactly. The validation step is extensive because the inputs are other people's accounting. P&Ls and transaction exports arrive in whatever shape a prospect's system produces them, and an analysis built on data that silently failed to parse correctly is worse than no analysis — it puts a wrong number in front of a prospect. Getting the system to catch bad input and say so, rather than confidently producing a deck from it, was a meaningful share of the work.

Then the commentary had to sound like them. 80/20 has a specific point of view about where profitability actually comes from in industrial businesses, and generic AI-generated observations about a P&L would have undermined the deck rather than supporting it. Calibrating the generative layer against their underlying philosophy took more iteration than the analysis itself.