Spend analysis for normal companies: start with one question

Spend analysis has a reputation as an enterprise exercise: a data warehouse, a taxonomy, a consultant. That version exists. But the version that finds most of the money for a normal company is four questions against twelve months of invoices, answerable in an afternoon, and the discipline to act on what they surface.

Here is that version.

Get the data: the rough cut is enough

Export twelve months of supplier payments from your accounting system: supplier, amount, date. Add a category column and fill it roughly: ten to fifteen categories, "packaging," "freight," "IT," "maintenance." Do not build a taxonomy; a rough cut used this week beats a classification project that dies at 60% complete. Precision has diminishing returns that arrive almost immediately.

Question 1: Where is the 80/20 line?

Sort categories by annual spend. Almost always, three to five categories carry most of the money, and those are the only ones the first pass should care about. The long tail is real but it is next quarter's problem.

For each top category, one follow-up: when did this last see competition? A category that has not been competitively sourced in two years is running on prices nobody currently defends. That is not an accusation of the supplier; it is entropy. Prices drift when nobody is watching, and the three-quote habit is the watching.

Question 2: Where is the same thing bought twice?

Look for the same category split across several suppliers: five stationery vendors, three freight forwarders, four printers. Some fragmentation is deliberate (backup suppliers, geography). Most of it is accidental: different people solving the same problem without seeing each other, each purchase too small to negotiate.

Fragmentation is the cheapest finding to act on: consolidate the category into one structured request, let the incumbents and one discovered supplier compete for the combined volume, and the volume discount that nobody could claim individually appears. This is usually the fastest payback in the whole exercise.

Question 3: What was bought off-process?

Scan for purchases that ducked whatever process exists: bought on a card, expensed, ordered ad hoc at list price while a negotiated supplier existed. This is maverick spend, and its cause is usually not defiance but friction: the official way was slower than the workaround.

That diagnosis matters because it dictates the fix. You do not fix maverick spend with a memo; you fix it by making the official path the easy one, which is the entire argument for structured requests that take minutes instead of email rounds that take days.

Question 4: Which suppliers concentrate risk?

Flip the sort: which suppliers carry the most spend? For the top handful, two checks: is there a live alternative if they fail, and does their performance record justify the concentration? A dominant supplier who earns it is fine. A dominant supplier nobody has evaluated since onboarding is a quiet single point of failure with an invoice history.

From findings to money

The analysis is worth nothing until a category gets re-sourced. The conversion is mechanical:

1. Pick the finding with the best effort-to-money ratio, usually a fragmented category or a long-uncompeted incumbent.
2. Write one structured request for the consolidated, specified need.
3. Invite the incumbents and at least one new bidder; competition without a newcomer is a rehearsal.
4. Compare on total cost of ownership, award, and diarize the category for re-competition.

Then put one hour per quarter in the calendar: top ten suppliers, new fragmentation, categories drifting past their re-compete date. The habit outperforms the depth.

The version where the analysis builds itself

Notice what made the afternoon necessary: the purchase history had to be reconstructed from accounting data, because the sourcing itself ran through inboxes and left no structured trace. That is the part that changes when sourcing runs on a platform. Every request on VEXORS carries its category, suppliers, bids, award value, and date by construction, so the spend record accumulates as a side effect of buying. The AI spend summary then turns that record into plain-language narratives: where the money went, what changed, what is drifting. The afternoon of spreadsheet archaeology becomes a report you read.

Run the afternoon version once; it nearly always pays for itself in the first consolidated category. Then let the structured version take over: start free, route the next purchases through it, and watch the analysis assemble itself while you buy.