You've classified your spend. Now what?

Quick answer: A classified line isn't the finish line — it's what lets you finally ask real questions: category totals, supplier consolidation, price comparison, negotiation leverage. But charge lines like freight, fuel surcharges, and packaging don't fit a product taxonomy. Tag them by charge type instead of forcing a product code, and decide deliberately whether to allocate them into the goods or report them on their own.

On this page: The invoice that doesn't fit · The principle · How to actually treat them · What the analysis looks like once they're tagged · Where Pearstop fits · FAQ

Classification is usually sold as the finish line. You send over a few hundred thousand invoice or PO lines, they come back with a category on every row, and the project is declared done. Except that a classified dataset isn't an answer. It's the point at which you can finally start asking questions.

The questions are the whole reason anyone pays for classification in the first place:

• Category totals you can act on. Not "roughly what do we spend on cleaning", but a number a category manager is willing to defend in a meeting.
• Supplier consolidation. Eleven suppliers delivering the same category across five sites, and no single view that showed it before.
• Price comparison across sites. The same goods bought by two locations at two different prices, from the same supplier.
• Negotiation leverage. Knowing your real annual volume with a supplier before you sit down with them.

We've written before about how that analysis is structured — see the spend cube framework, executed with AI. This piece is about a small, dull-looking thing that quietly breaks all four of those questions if you get it wrong.

The invoice that doesn't fit

Take a single invoice. Nothing unusual about it — this is an illustrative one with round numbers, not a real customer's:

Lines 1 and 2 are products. They classify cleanly. Lines 3, 4 and 5 are the problem.

A fuel surcharge is not a thing you can buy. Neither is "packaging" in this context, and neither is the transport line — it isn't a freight service you went out and sourced, it's a cost of receiving the goods above it. A product taxonomy like UNSPSC is built to answer the question what is this item?, and for these three lines there is no honest answer. They have no home.

So what happens to them in practice? Usually one of three things, and all three are bad in a different way.

The third option is the most common, and it's the one that erodes trust in the dataset fastest. A spend cube with a large unclassified bucket doesn't get used.

The principle

You could not have bought the goods without paying those charges. That makes the charges part of the spend — even though they are not part of the taxonomy.

Those are two different statements, and the mistake is treating them as one. The charge belongs to the spend. It does not belong in a product category. Both things are true at once, and a data model that can't hold both will force you into one of the three bad options above.

So: recognise these lines, decide deliberately how you treat them, and count them. In that order.

How to actually treat them

Separate the taxonomy question from the accounting question. "What is this thing?" and "does this cost attach to a category?" are separate fields, and they should be stored separately. A fuel surcharge has no product identity but it does have a cost attachment. Once those are two columns instead of one, the conflict disappears.

Tag the line with a charge type, not a product code. Freight, fuel surcharge, packaging, handling, small-order fee, environmental levy — whatever your suppliers actually put on invoices. This is a small, closed list, and it wants to be your list, not a branch of a product taxonomy bent to fit. Keep it short — a list that grows past a dozen values usually means someone is classifying free text rather than recognising a pattern.

Worth saying: freight and logistics services do have a legitimate place in a product-and-services taxonomy. If you source transport from a carrier under its own contract, that is a service you bought and it classifies as one. The charge-type tag is for the other case — a charge riding along on an invoice for something else.

Then decide: allocate, or report on its own. This is the choice that changes what a category manager sees, so make it on purpose.

• Allocate the charge back onto the goods it accompanied when you want landed cost — the true cost of having that material available at that site. Pro-rate the 1,000 of charges across the 5,000 of goods and the cleaning-consumables category now carries 6,000. This is the right basis for comparing sites, comparing suppliers, and building a business case.
• Report the charge on its own when the charge itself is the thing you want to manage. If freight across all categories adds up to a meaningful number, it is a sourcing target in its own right, and burying it inside forty product categories hides it from the person who could renegotiate it.

In most cases you want both, which is exactly why the tag matters: one tagged dataset supports both views. An allocated-only dataset can't show you the freight question, and a standalone-only dataset can't show you landed cost.

Keep it visible either way. Whichever treatment you choose, the line stays in the dataset with its tag intact and its treatment recorded. That's the part that beats both dropping and burying: someone can always ask "what's in this total?" and get an answer.

What the analysis looks like once they're tagged

These are the questions that become answerable — and that simply aren't, on an untagged dataset:

• How much of this category is product and how much is freight? A category that's 15% transport is a logistics conversation, not a price conversation, and you'd never see the difference from a single blended total.
• Why is one site's landed cost higher than another's for the same goods? Same supplier, same items, different delivery pattern. Often the answer is order frequency: lots of small drops, each carrying its own charge.
• Does this supplier's headline price advantage survive their surcharges? The cheapest unit price on the tender isn't always the cheapest invoice. With charge types tagged you can compare like for like, after charges.
• Is freight worth consolidating or renegotiating on its own? Sum every freight and fuel-surcharge line across all categories and suppliers. That total is often the first time anyone in the organisation has seen it, and it's frequently large enough to source deliberately.
• Are surcharges growing faster than prices? Fuel surcharges move on an index, not on your contract. Tracking them separately means you notice.

None of these need new data. They need the data you already have to stop throwing away one part of the invoice.

Where Pearstop fits

This is one of the things our classification engine handles directly: charge lines get a charge-type tag alongside the product classification, so they stay trackable and it stays clear what's actually in a category total. They aren't dropped, they aren't pushed into a product code where they don't belong, and they don't disappear into an unclassified bucket.

The trigger for writing this up was a facilities-management pilot where the surcharges on supplier invoices didn't fit the standard taxonomy at all, and a custom charge-type flag was the clean way through. It's a pattern, not an exception — if your suppliers invoice freight, you have this problem whether or not it's been named yet.

It sits alongside the other unglamorous parts of getting a spend dataset to a usable state: supplier matching, which is usually the real bottleneck, and classification quality itself, where the standard that matters is trustworthy confidence levels rather than raw accuracy.

Stephanie Wiechers

CEO & Co-founder, Pearstop

Stephanie leads Pearstop's go-to-market and strategic direction. She works directly with procurement and FM leaders across Europe to understand how data quality affects margins, contracts, and AI readiness.

LinkedIn →