What Your Records Don’t Say

An equipment dealer’s rep has a customer on the phone asking for a quote.

The rep needs to understand the request, choose the right equipment, set a price, and keep the conversation moving. The business needs something else from the same call: a complete, structured quote it can follow up on, report on, and learn from.

Those two needs pull against each other. Every field the rep stops to fill in is a moment they aren’t listening to the customer. So reps serve the customer first and fill in what they can afterward, and the company’s records come out thin.

Map the decisions that carry your own business from first contact to a customer who comes back. For a landscaping company, part of the map might look like this:

Pick a consequential one. The quote that takes three days and loses the sale. The schedule that creates idle time every Monday. The pricing decision that quietly erodes margin.

Write down how it performs today: turnaround time, rework, margin, missed commitments, or another measure close to the outcome. You will need that baseline to know whether anything improves. BCG’s 2026 survey of more than 1,300 senior leaders found that companies tracking AI’s value directly in their profit and loss realize three times the value of companies that don’t formally measure it.

An AI support product may be exactly the right purchase. So may a faster month-end close. The question is whether improving that work changes a result the company needs.

Time saved inside a task isn’t a business result until it has somewhere to go.

Start with the person who makes the decision. Ask them to explain it out loud, the way they would train a new hire.

Use a recent case so the explanation stays close to actual work. For the dealer, that means sitting with a rep and walking through last week’s quotes: what they asked first, what they already knew about the customer, which combinations of equipment they rule out without thinking, and when a price needs someone else’s sign-off.

Record the conversation, and give the transcript to an AI tool your company has approved for that information. Ask it to describe:

• What starts the decision.
• What information the person uses, and which sources they trust.
• Which rules and exceptions apply.
• When they ask someone else.
• How they know the result was good.

Ask the AI to flag gaps and uncertainty rather than fill them in. Then have the decision-maker read what it wrote and work through another case.

The useful part is what it misses or gets wrong. Not every mistake means missing knowledge; sometimes the AI simply misread what it was given. The rest are the sticky notes: facts and exceptions people have learned to carry because the systems don’t. Which customer has a special agreement. Which supplier is late every spring. Which field holds the current price and which one nobody uses anymore. Everybody knows them. Nobody wrote them down.

An hour can produce a useful first draft. A consequential decision may take several conversations, especially when exceptions are the hard part.

With the decision described, sort its steps by the authority you are prepared to give a system. For the dealer’s quote, it might look like this:

The middle pile is where the earlier tension resolves. The rep talks to the customer the way they always have. The system turns what they said into the record the company needs, and the rules catch what doesn’t fit.

Some work that feels like judgment will turn out to be a rule nobody had made explicit. Other work will reveal exceptions that an apparently simple rule cannot handle. Neither finding is a reason to rush into automation. Both help you choose the right design.

A clear rule may need ordinary software. A recommendation may benefit from AI. A difficult customer conversation may benefit most from a person who has the right context in front of them.

A convincing description is a starting point. To find out whether it is dependable, test it against recent real decisions, including awkward ones.

Give the proposed process the information available at the time. Check what it recommends, what evidence it uses, and whether it recognizes when to ask for help. Have the people who do the work examine the results.

For the quote, that means replaying recent calls. Give the system what the rep said, compare its draft with the quote the rep actually saved, and check whether it flags what it can’t resolve instead of guessing.

The measure that matters isn’t how much the system typed. It’s whether reps stay with the customer and the record comes out complete.

Then try it on live work with a person reviewing the output before it affects a customer or commitment. Compare the result with the baseline you recorded: did quotes arrive sooner, did rework fall, did the schedule hold?

Before granting more authority, define what the system may do, what it must refer to a person, and who can stop it. Keep checking the business outcome as the work changes.

The aim is to improve the decision without creating more checking, correction, or customer trouble elsewhere.

You now know what a product has to support: the decision, the required information, the exceptions, the limits on its authority, and the outcome you expect.

If a vendor can meet those requirements, buy the product. There is no prize for rebuilding something that already works. If the missing piece is company knowledge or the way work moves between teams, include that work in the plan.

The case for waiting is reasonable. The tools improve every few months, prices fall, and vendors are packaging AI for every industry. But the work in this brief doesn’t depend on which product you choose. A tested account of how your business makes a decision stays useful when the product changes, and it’s what you’ll judge the next product against. It does go stale as policies, customers, and practices change, so give it an owner.

Wait on the tools if you like. Don’t wait on the knowledge.

For the decision that matters most, pick the highest statement you can support with evidence.

You may be at different stages in different parts of the company.