Context
A mid-sized transport company: ~50 trucks, 80 employees, operating both domestically and on EU routes. The owner — 15 years in the business — sees that “everyone’s talking about AI” and doesn’t want to fall behind. He’s willing to spend $50–100K on a pilot, but wants to be sure it isn’t “just for the sake of it.” The team: an in-house back office on 1C, a dispatch system from an external vendor, and accounting kept separately.
The pain
The owner met with several AI companies. Some offer “AI for route optimization” — sounds good, but he isn’t sure his routes are unoptimized (the dispatcher has 18 years at the company and manages intuitively). Others — “AI for demand forecasting” — also sounds nice, but his demand is steady, with contracts booked a quarter ahead. A third group — “an AI chatbot for customers” — but he has 40 customers, all of whom know their managers by name. The owner realizes he’s being sold generic solutions, without anyone looking at where his actual “weak spot” is. And he can’t find that weak spot himself — he’s inside the processes, so everything looks normal to him.
Approach
Eight weeks. Weeks 1–2 — deep reconnaissance: 12 one-hour interviews (the dispatcher, the accountant, 3 drivers, managers, an HR officer, even the security guard at the lot), plus observing the dispatcher’s working days (2 full shifts alongside him). We were looking not for “where AI fits,” but for “where it hurts,” “where we wait,” “where we do by hand what shouldn’t be done by hand.” Weeks 3–4 — a map of all processes with a time analysis: where the company loses hours every day. Three unexpected pain points surfaced: 1) ~12 hours a week the back office spends hunting for documents from drivers (photos in the messenger, in email, scattered around), 2) ~6 hours a week accounting re-enters data from delivery notes into 1C by hand, 3) 2–3 trips a month are delayed by misunderstandings at customs (paperwork filled out incorrectly). Weeks 5–6 — comparing solutions: which of the 3 pains is genuinely cured by AI, which by ordinary automation, which by a simple process fix. It turned out: #1 — AI document recognition genuinely helps ($8K/year), #2 — plain OCR without AI ($2K/year), #3 — not a technical problem at all, but driver training (0 cost, a week of work with the HR officer). Weeks 7–8 — recommendations + draft requirements for problem #1.
Result
The owner didn’t spend $50–100K on “AI route optimization” — instead he spent $10K on specific, targeted fixes for three real pains. Six months after we finished, the company did the math: ~18 hours a week of administrative work freed up, ~3 trips a month no longer delayed at customs. The owner got a map of his own processes he can use, a year from now, to look for the next points to improve.
What it taught
“Adopt AI” isn’t a task. The task is to find where it genuinely hurts, and only then decide whether AI fits, or ordinary automation, or a solution that isn’t technical at all. AI pilots often fail not because the technology is bad, but because it was applied to a spot that didn’t hurt.
This is an illustrative composite pattern from 8 years of work in IT — not a specific client. AG is in a validation phase.