What should happen when a customer's words match nothing…

What should happen when a customer's words match nothing in your catalogue?

It should say so.

That sounds too obvious to write down, and it is the single most commonly broken rule in this whole category — because a system that returns nothing looks broken, and a system that returns something looks like it worked. The pressure to produce a plausible line is enormous, and every bit of it points the wrong way.

A line that matched nothing is cheap: a person looks, resolves it in a few seconds, and the quote goes out. A line that matched the wrong thing at high confidence is expensive, and the cost lands weeks later when the wrong item is on a lorry.

For anyone specifying what a quoting system does with a line it cannot resolve. 7 minute read.

The near-miss is the failure, not the miss

Worth separating three outcomes that get lumped together:

Matched, correct — Nothing
Matched nothing, and says so — Seconds of a reviewer's attention
Matched something adjacent, confidently — The order, or a return, and it is discovered late

The third row is not a worse version of the second. It is a different failure class, and the thing that makes it expensive is precisely the thing that makes it look successful: it produces a complete, priced, plausible row that nobody has any reason to question.

Microsoft documents its own near-misses, which is unusually candid

You do not have to take this from us. Business Central's Sales Order Agent [GA] has a documented section titled Scenarios impacting reliability in handling customer requests, and two of its entries are exactly this shape.

The first is a substitution:

"Searches for related consumables often return the main product instead of the requested accessory or consumable. For example, suppose the customer requests a quote for WonderWash Waschpulver in the DE version. The inventory includes WonderWash Waschmaschine but not WonderWash Waschpulver. However, the agent returns WonderWash Waschmaschine."

Read the middle sentence carefully: the requested item is not in the catalogue at all. The correct answer was we do not stock this. What came back was a washing machine instead of washing powder — same brand, wildly different item, wildly different price.

The second is a partial-word failure:

"When a search term appears as part of a longer word, the system fails to match it. For example, suppose a customer requests a quote for Züge von Wagonwheel in the DE version. The inventory includes the items WonderWagon Nachziehzug and WonderWagon Elektrischer Zug. However, the agent only returns WonderWagon Elektrischer Zug."

Here the failure is a silent omission — one of two valid candidates is simply not offered, and nothing signals that the list is incomplete.

Both are bounded to specific local versions in Microsoft's own wording, and both are published by Microsoft rather than discovered by us. Publishing them is the right call and more than most vendors do.

Why a fuzzy fallback produces near-misses by design

This is not carelessness. It falls out of how any sensible matching strategy is layered.

Microsoft documents three layers: an exact identifier match, then semantic attribute matching, then —

"Broader fuzzy search falls back to a global search across all items when the direct and attribute-based searches don't produce results."

That fallback exists for a good reason. Customers describe things loosely, and a system that gives up whenever an item number is absent would hand back most of a real inbox.

But notice what the fallback is asked to do. It runs precisely when the first two layers found nothing — which is to say, it runs on the hardest cases, with the weakest signal, and its job is to produce a candidate anyway. A global search across all items will almost always return something.

So the design question is not whether to have a fallback. It is what the fallback is allowed to do with a weak result — put it on the quote, or put it in front of a person.

Say nothing matched, and say it at zero

Our position, and it is a design choice rather than a documented fact.

An unresolved line stays on the sheet, with the customer's own words attached, and a confidence of zero. Not the nearest insulation product with a hopeful tick against it. Not forty per cent. Not a blank row. Zero, stated, next to what the customer actually wrote.

Three properties matter and each is deliberate:

The line does not disappear. A dropped line is the worst outcome available, because the quote arrives complete and short, and neither party notices until delivery.
Zero is not the same as low. A number in the middle invites a reviewer to accept it on a busy afternoon. Zero is a refusal to guess, and it reads as one.
The customer's words sit underneath. A reviewer resolving the line needs the raw text, not our interpretation of it — that is what makes the fix take seconds instead of a mail-client search.

Three ways to settle an unresolved line, and the reviewer picks

A row that says "needs you" has to give the reviewer something to press, on the row itself, always visible. A control that appears only on hover is a control somebody misses.

Ask the customer. The question goes out from the same workspace and into the same thread, attached to this line — "pipe insulation 22 mm, which thickness, and is it the foil-faced?" The answer lands back against the row it belongs to rather than in somebody's inbox. Without this, an incomplete line is a dead end.

Find it by hand. Search the catalogue and pick the item. The reviewer's choice beats every heuristic, and it should be remembered for this customer so the same words resolve without a person next time.

Mark it not stocked. Which brings us to the outcome people are most reluctant to build.

"We do not stock this" is an answer, not a failure

There is a strong instinct to treat not stocked as an error state, and to quietly drop the line rather than admit it.

That instinct loses business. A quote that comes back covering thirty-nine of forty lines, and says plainly that line eleven is not something you carry, is a useful, professional document. The customer sources that one item elsewhere and gives you the rest. A quote that silently returns thirty-nine lines invites the customer to notice the gap themselves — and to wonder what else was missed.

Say it on the quote. It costs one row and it buys the thing the whole exercise depends on: a customer who believes what the document says.

The upstream lever is catalogue quality, and Microsoft says so plainly

Worth ending here, because it is the cheapest improvement available and it needs no software at all.

"Although the agent can find products based on vague and incomplete descriptions, the quality of product information in Business Central affects its effectiveness. You can improve the agent's ability to find products by enhancing descriptions, attributes, categories, and extended text of your inventory items."

Microsoft adds a practical note: "It might take up to 15 minutes for newly entered data to become searchable as the system reindexes the tables in the background."

No matching strategy rescues a catalogue of cryptic abbreviations — Microsoft says as much, noting that "using cryptic abbreviations versus friendly names can reduce output quality." Before anyone buys software to interpret customer language, it is worth asking how much of the difficulty is the customer's language and how much is your own item descriptions.

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 know how many of your lines would come back unresolved, and why? Book a 15-minute call — bring a real request and your catalogue, and we will look at both. No deck.

What should happen when a customer's words match nothing in your catalogue?