Most procurement teams I have worked with can tell you exactly how much they spent last quarter. What they cannot tell you, without a week of spreadsheet archaeology, is which suppliers actually matter. Procurement analytics is supposed to close that gap, and most of the time it does not, because the work stops at a spend report instead of turning spend into a supplier strategy. A report tells you where the money went. A strategy tells you what to do about it. Those are not the same thing, and the distance between them is where most procurement programs quietly stall.
So let me walk through a project that crossed that distance, what it took, and why the boring part was the part that mattered.
What procurement analytics gets wrong before it starts
The usual version goes like this. Someone pulls a year of purchase orders, sums them by supplier, sorts descending, and ships a dashboard. The top ten suppliers light up. Everyone nods. Nothing changes.
The problem is that raw spend is a terrible ranking signal on its own. A supplier you spend a lot with might be completely replaceable. A supplier you spend almost nothing with might be the only source for a part that stops your line if it is late. Sorting by amount tells you who gets the biggest invoices, not who carries the most risk or the most bargaining power. Category managers know this in their gut, but the data they are handed never reflects it, so they keep negotiating on instinct.
Good spend analysis has to answer two questions at once. How much does this supplier cost us, and how exposed are we if something goes wrong. One axis is money. The other is risk and substitutability. You need both, and you need them per category, because a strategic supplier in one category is a transactional one in another.
The case: a large industrial buyer with no supplier map
Here is one our team ran. A large industrial buyer, the kind that buys everything from raw materials to office consumables across dozens of categories. The procurement team had spend data. What they did not have was a way to separate strategic suppliers from critical ones, operational ones from purely transactional ones. Every supplier looked the same in the system, so every negotiation got roughly the same effort, which meant the high-value relationships were under-managed and the trivial ones were over-managed.
The before state was not a lack of data. It was a lack of structure on top of the data. The spend existed. The supplier strategy did not.
What we built was a spend-analytics dashboard with a supplier segmentation matrix at its center, plus category and supplier filters so a manager could move from the whole organization down to a single category in a couple of clicks. The matrix placed every supplier into one of four cells.
- Strategic. High spend, hard to replace. These are partnership relationships, joint planning, long horizons.
- Critical. Lower spend, but a supply failure hurts badly. These need risk management and second sourcing, not price haggling.
- Operational. Meaningful spend, easy to substitute. These are where competitive tension and consolidation pay off.
- Transactional. Low spend, low risk. These should cost almost no management attention at all, ideally automated through to a catalog or a card.
Why supplier segmentation beats a spend ranking
The reason this works is that it tells a category manager where to spend their own time, which is the scarcest resource in any procurement function.
A flat spend ranking pushes everyone toward the same move: go squeeze the biggest suppliers on price. Sometimes that is right. Often it is the wrong tool. You do not squeeze a strategic partner on price and expect them to prioritize you when capacity gets tight. You do not run a sourcing event for a critical single-source component just because the spend is low, you go find a second source so a delay does not stop production. Supplier segmentation makes those distinctions explicit and puts them on a screen the whole team can argue over with the same picture in front of them.
The filters mattered as much as the matrix. Aggregate procurement analytics hides everything that makes a category specific. A supplier that is operational at the company level can be strategic inside one category where they are the main source. Letting managers filter by category and then by supplier meant the four-cell logic held up at the level where decisions actually get made, not just in the boardroom summary.
The unglamorous part: making spend data trustworthy
I will be honest about where the real effort went. It was not the matrix. The matrix is the easy part once the inputs are clean. The effort went into making the spend data say what it claimed to say.
Spend data in a large buyer is messy. The same supplier shows up under four spellings and three legal entities. Purchases get coded to the wrong category. One-off accruals look like recurring spend. Intercompany transactions inflate totals. If you build a segmentation matrix on top of that without cleaning it first, you get a confident-looking dashboard that is quietly wrong, and a wrong dashboard is worse than no dashboard, because people act on it.
So the work that nobody films was supplier deduplication, category normalization, and a set of rules deciding what counts as spend and what does not. That layer is what let the matrix mean something. It is the same lesson I keep relearning across enterprise data work: the model or the visualization is the last 10%, and the data preparation is the 90% that decides whether the last 10% is believable.
Reading the matrix as a set of decisions
A segmentation matrix is only useful if it ends in actions, so the dashboard was built to push toward decisions rather than admiration.
Strategic cell, the conversation became relationship and joint planning, not quarterly price pressure. Critical cell, the conversation became supply risk: who is the backup, what is the lead time, what happens if this one source goes dark. Operational cell, the conversation became consolidation and competitive sourcing, fewer suppliers doing more volume at better terms. Transactional cell, the conversation became removal of effort: automate the buying, stop spending manager hours on items that do not move the number.
That is the output category managers actually used. Not a static report, a working view of spend data that told them where their attention was worth the most. The procurement team stopped treating every supplier as the same kind of problem, which is the whole point of procurement analytics in the first place.
How to scope this for your own spend
If your procurement team is sitting on spend data it cannot turn into a supplier strategy, the mistake to avoid is commissioning a giant platform build. The way our team runs it is in provable steps.
- Map the data first. Get your spend extract, your supplier list, your category structure. Most of the early value is just seeing how dirty the inputs are and fixing the deduplication and categorization before anything else.
- Build the matrix on one or two categories. Prove the four-cell logic on a category the team knows cold, so they can sanity-check the placements against their own judgment.
- Then scale across categories with the filters in place. Once managers trust the matrix on familiar ground, the same engine extends to the long tail where their intuition runs out.
The team behind this has run 100+ enterprise AI and data projects across mining, energy, finance and manufacturing, with senior specialists only. The pattern is always the same: clean the data hard, structure it into decisions, keep the people who know the categories in the loop. You can see more of this work on our case studies page, and the broader procurement and supply chain angle is where spend analysis connects to the rest of the operation.
If you want to see what your own spend looks like under a supplier segmentation lens, book a call and we will walk through it with your categories in mind.
The dashboard is the easy part. Turning messy spend data into a supplier strategy a category manager can act on is the work that decides whether procurement analytics changes anything.