How do you read a vendor who will not name a weakness?

Because you will meet the weakness either way. The only variable is whether you meet it before you sign.

A product is a stack of choices, and every choice buys one thing by giving up another. Software that reads any format takes longer to set up than software that only accepts a form. A tool tuned for ocean groupage will be clumsy on one air shipment of dangerous goods. None of that is a scandal. It is arithmetic.

So D2 is not really about the product. It is about the vendor, and there are three things they can do with it. Name a real weakness. Dress a strength up as one. Or say there is not one.

The first tells you they know their product and expect you to still be here in two years. The second tells you somebody has been trained. The third tells you nothing about the software and a great deal about the relationship.

Most buyers stop at the third step. The value is in the fourth. “Name one thing” invites a sentence, and a sentence can be prepared. “What kind of forwarder should not buy this” asks them to describe somebody real, in front of a person who knows those firms. Much harder to invent.

It costs the vendor something to say out loud. That is the whole test.

A real one has a shape. It names a situation rather than a quality, it comes with a rough size, and it ends with what people do about it today.

One note on that last row. Do not push for a delivery date on a weakness. You will get one, it will be sincere, and it will move. What you want is the version of your operation that works whether or not it lands.

Public buying guidance treats this as your job, not a favour. The UK government’s guidelines for AI procurement tell buying teams to “try to establish any limitations of the model” and to ask suppliers “to describe their strategies on how to address these shortcomings”. D2 is that instruction in one sentence.

Article 13 of the EU’s AI Act shows what a written limitation contains. Instructions supplied with a high-risk AI system must state its “characteristics, capabilities and limitations of performance”. A freight quoting tool is not a high-risk system, so this binds nobody you are meeting. Borrow the shape: the circumstance, the size, the effect.

Four answers sound like candour and carry none of it. You will hear at least one.

One. The strength in a costume

“We are probably too thorough.” “Our setup is more hands-on than people expect.” Each of those is a compliment wearing a coat. Send it back with the follow-up, unchanged: what kind of forwarder should not buy this?

Two. The weakness that belongs to you

“We do not always get clean data from customers.” That is your data and your job, and it is true of every vendor, so it separates nobody. The version worth accepting names what the product does when the data is bad, which is usually to ask rather than guess.

Three. The weakness that is already mended

“Reporting used to be thin, we rebuilt it in the spring.” A weakness with a resolution attached is a case study. Ask for a current one: what is weak today, on the version you would give me?

Four. The price answer

“Honestly, we are expensive.” It may be true, and it is not a product weakness. Price is a negotiation, not a limit. Ask again about the work: I mean an enquiry it handles badly.

None of the four ends a meeting. Each is a reason to ask again in different words. A vendor who answers properly on the second ask is fine. One who hands you a second costume has shown you how the next two years will go.

An unanswered D2 is not a rejection on its own. It is a reason to trust only what you have watched happen.

You cannot grade an answer to D2 without knowing where this job is genuinely hard.

None belongs to any product. They are the places the work resists software, and where honest answers cluster.

• Rates that live in a person, not a system. The agent who quotes on WhatsApp when you ask, and only when you ask. Nobody has solved this, and a vendor who says they have is answering a different question.
• The lane you run once. Software priced from your own history is confident on the trade you move weekly and thin on the one asked about this morning.
• The enquiry that is not one enquiry. “Sort out last month’s invoice, and quote me Nhava Sheva to Hamburg.” Two jobs, two departments, one subject line.
• Attachments that are pictures. A rate sheet photographed at an angle, a packing list scanned like a fax, a screenshot pasted under three forwarded signatures.
• Cargo that needs a person to look at it. Dangerous goods where the class changes the routing, out-of-gauge that needs a survey, temperature-controlled with one customer’s own tolerance.
• Anything with a customs decision inside it. Classification, duty, the filing. That work stays with your broker and your own filing system, and a vendor who implies otherwise is worth a second look.
• The rule your firm has that nobody else has. This customer never sees a surcharge line. This agent is quoted at cost. Real, written nowhere, and software meets it late.

Five of those seven are about the input, not the intelligence. That is the shape of this job, and a filter on the answer you got. A vendor who names a weakness from this territory has been inside a forwarder’s mailbox. One whose weakness is abstract may only have been in the demo.

Put it on paper before the day ends, in their wording, on one line. By next week you will remember the tone of the room and nothing else.

The last column is the reason to ask at all. A named weakness hands you a test you would never have designed. Take ten enquiries that live inside it and run those, not ten easy ones. The vendor gave you the test.

Then do the uncomfortable half. Answer D2 about the system you already run. Ask the two coordinators who touch it most what it is bad at, in the same four columns. A longer list than the vendor’s is worth knowing before you sign.

That happens, and it is not the same thing as hiding it.

A young product with a handful of live forwarders may not know where its edges are, because nobody has pushed it hard enough. The honest version is “we think it is this, and here is why we are not certain”, and it is worth accepting.

What is not worth accepting, at any size, is “nothing really”. A vendor with three live forwarders who names two weak spots is more useful than one with three hundred who names none.

It cuts the other way too. Be fair to a long-established product. When a mature tool names a weakness you can live with, it has told you something precise, learned in somebody else’s operation. That is experience, not decline, and it should not lose to a newer product that has less to confess.

D2 and D3 sit next to each other for a reason. D2 asks what the product cannot do. D3 asks whether the company will be around when it can. Read the two together, because a weakness only matters for as long as the vendor lasts.