Insights ·

Spend analysis case study: ten rules, one queue

How a multi-site senior living operator’s finance team stopped reading purchasing lines one at a time — with review rules they own, one ranked queue, and figures reconciled to the cent on every refresh.

A lot of spend analysis starts with a dashboard. This one started with the rules a finance team already used to decide which purchases deserve a second look — and the problem that applying them by hand, across thousands of invoice lines, was slow. Here is how we built Spend Intelligence for a multi-site senior living operator, and what we would tell anyone planning the same.

The problem with reading lines one at a time

Finance had clear rules: invoices over a threshold, counter sales and returns, restricted items, the same item bought at different prices by different communities. Each rule is easy to state. Applied by hand to thousands of lines, they are slow — and one of them is close to impossible. A price gap between two communities is invisible when you read one line at a time, because the line that would reveal it is somewhere else in the pile.

Is your purchasing review a candidate?

  • You have written rules for what deserves a second look, but people apply them by reading
  • The same item is bought by several sites, and nobody compares what each one paid
  • Thresholds live in someone’s head or an email, with no record of who set them
  • The flagged list has no owner, no status and no notes, so nothing is ever closed
  • Nobody is sure the review matches the finance system to the cent

What we built, in order

1. Write the rules down where finance can read them

The platform applies ten review rules to every purchasing line — from invoice and line thresholds to same-month price differences between communities, unusual quantities and possible duplicates. Each rule has a named owner, and every threshold shows who set it and when.

2. Rank, don’t just flag

A list of every flagged line is just a shorter pile. The platform scores each invoice and ranks them into one prioritized queue — critical, high, medium — with a status, an owner and notes on every invoice, so the review has a start, an end and a record.

3. Reconcile on every refresh

The figures reconcile to the source to the cent every time the data refreshes. When something doesn’t tie out, the integrity problem is reported, never silently corrected. A review the finance team can’t trust to the cent is a review they will redo by hand.

4. Read-only, inside their own cloud

It runs inside the operator’s own Azure subscription and reads one database view. It never writes back to the finance system. Sign-in goes through their own identity provider and is limited to a named list of finance users, and every settings change is audited, before and after.

Why there is no AI in it

This one is rules and computed figures only, on purpose. Its built-in analyst answers only from computed figures, shows the calculation, and says “can’t answer that yet” when it can’t. A purchasing review needed rules the finance team could read, own and audit — not a model’s opinion about which invoice looked odd. Not every problem needs AI, and saying so is part of the job.

The rule we took from it. If a finance team already has rules, the platform’s job is to apply them to every line, rank what they find and prove the totals match — not to replace the rules with something nobody can audit. It’s one of the rules we don’t break.

If this sounds like your month-end

Tell us which review your team does by reading. In a free 30-minute call we’ll show you a system we’ve already built for a comparable problem and tell you honestly whether yours is worth automating. Read the full Spend Intelligence case study, or see Spend-Down, the budget tracker we built for the same finance team.

Related