Out of 100 real enquiries, how many quotes go out with nobody editing them, and how do you count that?

Because “nobody touched it” is not a fact about software. It is a sentence about what somebody decided to count.

Two vendors can run identical products on identical work and quote you 45 and 88. Neither is lying. One counted every enquiry that arrived in the mailbox. The other counted enquiries on lanes where a rate card was already loaded, and treated a pricer approving a draft as no touch at all. The gap between those two numbers is definition, not capability.

Industries that had to settle this argument wrote the definition down. The European Central Bank glossary defines straight-through processing as “the automated end-to-end processing of trades/payment transfers – including, where relevant, the automated completion of confirmation, matching, generation, clearing and settlement of orders”. Written for payments rather than for freight, so it does not transfer to your quoting desk unchanged. What transfers is the discipline. End to end is named. The stages inside it are named. There is nothing left to stretch.

The NIST AI Risk Management Framework, which is intended for voluntary use, asks for the same thing in a different register. Its measure function “employs quantitative, qualitative, or mixed-method tools, techniques, and methodologies to analyze, assess, benchmark, and monitor AI risk and related impacts”, and it expects that “test sets, metrics, and details about the tools used during TEVV are documented”. Documented. Not remembered by the person presenting the slide.

Before you can argue about the number, you need the map. One enquiry passes through six or seven places where a person can put a hand on it, and only two of them look like editing.

Ask a vendor to point at this line and say which steps their percentage covers. Most quoted figures cover the fourth and fifth boxes only. The first three are where your team actually loses its mornings, and the last one is where a quote quietly becomes two quotes.

Three, and none of them requires anybody to lie. They are choices made while building a slide, by somebody who wanted a good number and was never asked to define it.

The denominator is where most of it hides

Work out your own denominator before the call. If a quarter of your week is agent-priced lanes, and agent-priced lanes are outside the vendor’s count, then an 88 that is true is a 66 in your business. Nobody has misled you. You just measured a different company.

The definition of a touch is where the argument is

This is the follow-up the checklist gives you, word for word: does a person checking it before it goes out count as touching it? There is no right answer. There is only a stated one. A vendor who says “no, approval does not count, and here is why” has given you something you can hold. A vendor who has never considered the question is telling you the figure came from a slide, not from a system.

The window is the one nobody thinks to ask about

Ask for three consecutive months rather than one. A percentage that holds across a strong month, a slow month and a peak-season month is worth more than a higher figure from a single quarter. If only one month exists, that is the honest answer and it is fine, as long as it is said.

It has four parts, and it is usually a little uncomfortable to say out loud.

• 01The number, with its denominator. “Sixty-two of every hundred enquiries that reached the pricing mailbox.”
• 02The definition. “A touch is any change to any field after the draft was created. Approval counts as a touch. So does correcting a postcode.”
• 03What the rest needed. “Twenty-one needed a missing weight or dimension. Nine were on lanes we had no rate for. Eight were not enquiries.”
• 04The period and the volume. “Three months, about four thousand enquiries, one customer of your size and two larger ones.”

The third part is the one worth pushing on. What the rest needed tells you what your team will still be doing on a Tuesday, and it is the only part of the answer that describes your day rather than the vendor’s product.

Turn it on us as well. FreighAI is built so that “Your team approves. They do not compose, forward, file or chase. The reply goes out in their name”. Read the strict definition against that design and the honest consequence is plain. If approval counts as a touch, then a product with a person on every quote has a touchless share of zero, and any higher number it quotes has to be excluding the approval step. Ask us which definition we are using, and hold the answer to the same standard as everybody else’s.

The UK guidance on responsible AI in recruitment, written for a different kind of purchase, puts the burden where it belongs: “a supplier should be able to provide evidence of these claims to your team, regardless of your level of technical literacy”. You do not need to be technical to ask what was counted. You need to refuse to move on until somebody says.

Because a vendor counts quotes and an operations head feels effort, and the two do not move together.

Sixty easy quotes that go out untouched save your pricer an afternoon. The forty that came back save nothing, because each one is a chase, a missing weight, an agent who has not replied and a customer who wants it today. A high percentage on a low-effort majority can leave the working week almost exactly as it was.

So ask the second question underneath the first. Not only how many went untouched, but how many hours of your pricing desk the untouched ones represent. A vendor who has run at volume will have an answer. A vendor who has not will hear the question for the first time in your meeting, which is itself worth knowing.

Write four things down the same day, in the vendor’s own words, for each supplier on your shortlist.

• The figure and the exact denominator behind it.
• The definition of a touch, including whether approval is one.
• What the remainder needed, broken into reasons.
• The period, the volume and the kind of forwarder it came from.

Then do the thing almost nobody does. Take the definition the vendor gave you and put it in the pilot. Measure your own first month against their own words rather than against a number they chose. The UK guidance defines performance testing as “a process for assessing the performance of a system with respect to predetermined quantitative requirements or benchmarks”, and predetermined is the whole point. Agree the measure before the first enquiry runs, not after the first argument.

One last thing to carry back to the previous question. A vendor who has just run twenty of your own enquiries in front of you has handed you a denominator you chose yourself. Twenty enquiries, watched, counted by you, beats any percentage on any slide.