From Four-Day Quotes to Four-Minute Quotes: A Sample AI Engagement, End to End

Reading time: 12 minutes. What you'll see: how we take a business from "we think AI could help somewhere" to a deployed system with a measured payback — including the analysis that would have told them not to build.

The company

Kestrel Beverage Co. is a contract beverage manufacturer: other brands bring them recipes, Kestrel produces and packages the cans. Mid-size for the industry — about 30 employees, $12M revenue, one highly automated canning line, and a deliberately small roster of around ten active brand clients at a time to protect quality.

Their growth problem isn't demand. It's that every request for quote is priced by hand. An emerging brand emails asking for 15,000 cans of a new sparkling tea. Someone at Kestrel works out material costs, line time, changeover, liquid loss on a carbonated short run, and packaging — then checks the production schedule for a slot. It takes two to four days per quote. The people doing it are the same people scheduling the line. Some weeks they quote forty jobs to win six.

If you run a business on repeatable operations — a print shop, a machine shop, a caterer, a clinic — you know this shape: skilled people doing arithmetic a system should do, while the work they're actually paid for waits.

Phase 1 — Observe (week 1)

We don't start with AI. We start with a map. Two weeks inside the business produces a process inventory: every recurring workflow, who touches it, how long it takes, what systems it lives in.

The map surfaced three candidate workflows. We scored each on two axes: annual cost of the manual version and feasibility of automating it against their existing systems. Quoting won on both. (The other two — invoice matching and a customer portal — went on the roadmap with their numbers attached, not in the build.)

Phase 2 — Analyze & Quantify (week 2–3)

Here's the part most proposals skip, and the reason ours don't: every input gets a label.

Then the honest part — the model runs across the assumptions, including the row where the project loses money:

Total year-1 cost in the model: $61,400 ($35,000 build + $2,200/month operation). Payback at the confirmed baseline: about six months.

That negative row is why the audit exists. If discovery had shown one part-timer doing quotes in a slow market, the written verdict would have been don't build — and the $2,500 audit fee would have been the cheapest $50,000 Kestrel never spent.

Phase 3 — Build (weeks 4–10)

The system, at architecture level:

Two design decisions worth noticing, because they're how we build everything:

AI only where AI earns its place. The intake parser is AI — unstructured emails genuinely need it. The pricing engine is deterministic code — a quote must be right, not plausible. Firms that put AI in the arithmetic are optimizing for the demo, not the client.

The human approves; the system prepares. Kestrel's quote quality was never the problem — the four days were. The system compresses the four days to minutes and leaves the judgment where it belongs.

Phase 4 — Measure (from day one of operation)

Every build ships with its own scoreboard, because the ROI claim from Phase 2 becomes a dashboard, not a memory:

• Quote turnaround: target under 4 hours from RFQ to approved quote (baseline: 2–4 days)
• Hours displaced: the system counts every quote it drafts; hours saved computes from Kestrel's own baseline timings
• Win-rate delta on time-sensitive RFQs: the capacity-gain assumption from the model, now measured instead of assumed
• Exception rate: how often a human had to fix the parse — the honesty metric on the AI component

Ninety days in, the dashboard answers the only question that matters: is the model from Phase 2 coming true? If it isn't, the retainer includes fixing that, and the sensitivity table already told everyone which assumption to check first.

What this engagement costs, mapped to our published pricing

And the exit that costs nothing: if Phase 2's verdict is don't build, you keep the analysis and owe nothing more than the audit — which just saved you the build.

---

This is a sample engagement — the client is fictional; the method is not. To run Phases 1–2 on your actual business: book the free scan. The audit methodology behind Phase 2 is published in full. Two companion engagements run the same method on different problems: an Indianapolis HVAC company losing calls after 6pm and a freight brokerage drowning in paperwork.