When is an AI quoting system the wrong thing to build?

More often than anyone selling one will tell you.

This cluster has spent twelve chapters on how quoting breaks and what to do about it. This chapter is the other half of an honest argument: the situations where the answer is that you do not need any of it, and the cheaper things worth doing first.

For anyone about to spend money on this. 6 minute read.

The question that qualifies it, before anything about ERPs or budgets

One question predicts fit better than company size, industry or system:

What share of your quote requests arrive as free-text email, an attachment, or a photograph?

Everything in this category earns its keep on interpretation. A request that arrives already resolved — a portal submission, an EDI message, a web form with item numbers in fields — has nothing to interpret, and no amount of intelligence improves an answer that was already unambiguous.

A low share means a weak fit, whatever else is true about you. A distributor whose customers all order through a portal has a fast, accurate quoting process already, and the honest advice is to protect it.

Note the question is about share, not about a category. Almost nobody is at either extreme — the real distributor has some structured traffic, a lot of email, and one important customer who sends photographs of handwritten lists.

Five situations where the answer is no

Your inbound is overwhelmingly structured. See above. The engine has little to do, and the remaining benefit — one queue, one clock — is real but much cheaper to get.

You have little repeat business. This is the one people miss, and it is the most important. The value compounds through relationships: a confirmed match teaches the system what this customer means by those valves, and next month that line needs nobody.

A first-time enquiry has no history, no learned names, and often no agreed pricing — simultaneously the harder problem and the smaller prize. If most of your requests come from people you have not sold to before, the compounding never starts.

You sell services or configured-to-order engineering. If what you quote is designed rather than picked from a catalogue, there is nothing to match against. The hard part of your quoting is estimation, which is a different problem with a different shape.

You already run a mapped document-automation platform. If one is configured and working, the marginal gain is small and the switching cost is not. Improve what you have.

Quoting is not what you compete on. Some distributors win on stock depth, on technical advice, on credit terms, on being the only one who carries the part. If response speed is not what decides your orders, this is an efficiency project rather than a revenue one, and it should be sized as one.

Cheaper things worth doing first

Three of them, in ascending cost, and the first two are the ones nobody sells because there is no software in them.

Fix your item descriptions. Any matching system is limited by what your catalogue says about your products, and Microsoft states this plainly about its own agent:

"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."

It names the failure mode too: "using cryptic abbreviations versus friendly names can reduce output quality."

Put every channel in one place, with a clock. Most of the elapsed time on a quote is waiting rather than typing, and a single queue that shows how long each request has been sitting is a process change, not a purchase. Chapter 1.

Switch on what you already own. If you run Business Central, Sales Order Agent [GA] is in the product — "the agent is readily available in the product" — and activation is pointing it at a mailbox. Two weeks of real traffic will tell you what share of your requests it closes, which is a better basis for any further decision than any article, including this one. Chapter 12.

And one cost worth knowing if you quote over WhatsApp. From 1 October 2026 Meta charges per delivered message for service replies inside the twenty-four-hour customer-service window — replies that are free under the policy in force as this is written, with per-country rates due by 1 September 2026. It does not change whether to build, but it changes the running cost of a channel, and it is the kind of thing that should be in a plan rather than a surprise.

What nobody in this category can prove yet

Here is the part that should make you sceptical of everyone, us included.

The argument for quoting faster is that the first quote back wins a disproportionate share of orders. It is plausible, every distributor believes some version of it, and we have not seen it established with evidence we would be willing to publish.

So treat any win-rate improvement you are quoted as a claim, not a finding — and ask what it was measured against. Ours included: we do not have that number, and this article is not going to invent one.

What can be measured, and should be from the first week:

Time from request received to quote sent, per request.
Win rate on the fastest quartile of quotes against the slowest, per customer.
Share of lines needing a human, tracked monthly, per customer.

The second one is the business case. If faster responses do not improve your win rate, then on the elapsed-time reading there is no proven value here — and you will know that from your own data within a quarter, which is faster than any vendor will tell you.

What we will and will not claim

Plainly, because the rest of this cluster is worth less without it.

This is a system we build. It runs, it is demonstrable on a call, and it has zero delivered engagements. No distributor is running it today. There is no reference customer to introduce you to, and no number anywhere in these thirteen chapters describes anyone's business.

Every mechanism described across this cluster is a design position we hold and can defend. None of it is a measured result, and the chapters that could have carried one — the review rates, the hours, the margins — deliberately carry the mechanism instead.

If that makes us a harder sell than the alternative, that is the correct outcome.

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/

Not sure this is worth doing for your business? Book a 15-minute call — we will run the qualifying question against your actual inbound, and tell you if the answer is no. No deck.

When is an AI quoting system the wrong thing to build?