How do you show a reviewer why a quote line matched?

Two rows on a sheet, both marked as matched, both priced, both looking equally settled.

One of them was found by its own item number and there was nothing else it could have been. The other was a coin toss between two catalogue entries that differ only in which supplier they come from, and the system picked one on a margin so thin it could have gone either way.

On most screens those two rows are identical. That is the problem this article is about, and it is not solved by a confidence number — it is solved by showing the evidence.

For anyone deciding what a reviewer sees when they click a matched line. 7 minute read.

Certainty is not the same as evidence

A single score compresses everything a reviewer needs into one number, and the compression throws away exactly the part that decides whether to look closer.

A reviewer does not want to know how sure the system is. They want to know what it is sure on, and those are different questions. A match on an exact item number and a match on a fuzzy description can both come back certain. Only one of them should end the conversation.

So the panel behind a row has one job: let a person confirm or reject the match in a few seconds, without leaving the screen. Everything below follows from that.

How it was found changes how much you trust it

The first thing on the panel is the route, not the result.

Your own item number — Accept. There is nothing to interpret
A name this customer has used before — Accept, and it gets stronger every time it is confirmed
This customer's order history — Accept for a repeat line — it is the answer to "like last time"
Description and attributes — Look. This is where the judgement happened

Microsoft ranks its own routes the same way, and says so in one sentence:

"Exact identifier match by item number, vendor item number, GTIN, or cross-references. Exact identifier matches have the highest confidence."

*That is a statement about the match type, and it is the right instinct* — where a match came from is evidence about how much it can be trusted, independent of any score attached to it. A row that does not say which route it took has withheld the most useful fact it has.

How far ahead of second place did it finish?

This is the field that changes reviews most, and it is the one we most often find missing.

Call it the margin over next best: the gap between the winning candidate and the runner-up. It costs nothing to compute, because the ranking was computed anyway, and it separates two situations that look identical from outside:

The winner beat everything else outright. There was no real contest.
The winner edged the runner-up by a whisker — and the runner-up was the same physical item from a different supplier at a different price.

The second case is common in distribution specifically, because the same physical part legitimately exists several times in one catalogue under different codes. A system that shows only the winner is presenting the outcome of a coin toss as though it were a finding.

Show the gap, and a reviewer knows in one glance which rows deserve their attention. That is the entire mechanism — not a cleverer matcher, a more honest display.

The attribute that made it the right item — and Microsoft's own worked failure

An elbow is not an elbow. It is a material, a size and an angle, and the reason a particular catalogue row is correct usually lives in one of those fields rather than in the description.

So the panel carries the attributes that make this item the right one. Microsoft documents what happens when they go missing, and the example is worth reading in full because it is the clearest illustration of this point anywhere in the vendor's documentation:

"The agent searches for items using item attributes, but the attributes aren't shown to the agent on the item availability page. This condition decreases reliability for customer requests that depend specifically on the attributes."

Microsoft then walks it through. The catalogue holds two shirts, one linen and one cotton. The customer asks for a quote for "nice shirts". The agent's reply:

"We have X quantity of ITEM-1 Shirt and ITEM-2 Shirt"

Both items come back described identically, because the attribute that distinguishes them has dropped out. The customer then has to write back — "I want the ones with linen material" — which is a whole extra round trip caused by an attribute that was used to search and then not carried forward.

Two lessons, and the second is the one for anyone building this:

The attribute that resolved the match has to travel with the match. If it was good enough to decide, it is good enough to display.
The same applies to what the customer sees, not just the reviewer. A quote listing two rows that read identically is a quote that generates a question.

Show the alternatives, and say so when there are none

A reviewer who disagrees with a match needs somewhere to go. If the panel offers a winner and nothing else, the only route left is a catalogue search from scratch — which is the expensive path the system was supposed to remove.

So the panel lists the other catalogue items that could serve this line, ranked, with the fields that differ between them. Rejecting a match then becomes a click on the row underneath rather than a fresh investigation.

And when there are no alternatives, the panel says that too, rather than showing an empty space. Those are different facts: "we looked and there is genuinely nothing else this could be" is a strong signal, and it looks identical to a panel that failed to populate.

That distinction runs through this whole article. A blank is ambiguous; a stated nothing is information — the same argument chapter 5 makes about an unmatched line, and chapter 7 makes about a missing quantity.

Where the line came from, in the customer's own words

Underneath every row: the raw text it was derived from. Not our interpretation of it — the actual fragment, and where in the request it appeared.

For a spreadsheet, that means the row number. For a message, the sentence. A reviewer who cannot see where a line came from cannot check it, and will either accept it blind or re-open the original attachment, which is the four hours coming back one line at a time.

This is also the cheapest form of trust available. It requires no scoring, no model, no ranking — it requires keeping the original and never throwing away the link between what arrived and what was produced from it.

Microsoft shows its working too, and the comparison should be fair

It is worth being accurate about this, because a lazier version of this article would claim the vendor offers nothing here. It does.

"For each task performed by the agent, users get a detailed timeline that shows the key steps taken by the agent and human users, including the email conversation. Users can review this information and update the values and actions the agent suggested if needed. Business Central can also display the reasoning used by the agent and citations that led to a suggested value."

Reasoning and citations for a suggested value is real explanation, and more than many systems in this category provide.

The distinction this article is drawing is narrower. Explanation answers why did it choose that. What the panel described above adds is comparison — how close the second candidate was, and what separated them. Those are different questions, and the second is what tells a busy reviewer which of fifteen rows to open in the first place. Chapter 8 is entirely about that.

About Cognilium Cognilium builds AI systems that work in tandem with Microsoft Dynamics 365 — the decisions the ERP records but does not make. Business Central and Finance & Operations, on your own governed stack. https://cognilium.ai · https://www.linkedin.com/company/37180269/

Want to see what a match panel would show against your own catalogue, ties and all? Book a 15-minute call and we will walk it with your data if you bring it. No deck.

How do you show a reviewer why a quote line matched?