Government Contracting & DistributionFederal supply contractor with two operating companies

Finding the Real Bottleneck at a Federal Supply Contractor

Discovery across five departments at two companies, 43 use cases cut to 29 priced workstreams, and one finding about lost quotes the client fixed for free.

43use cases cut to 29 priced workstreams

25%of order lines traced to a fixable cause

12month program signed

August 22, 2026

The ask

A holding company with two federal supply businesses hired us for strategy. Similar work, different scale. What should they do with AI, what shouldn't they, and in what order.

What I did

I ran discovery across supplier relations, sales and marketing, business development, procurement and IT at both companies, and joined their IT team's annual offsite.

Their language first. The industry runs on jargon none of us knew. I built a 101-term glossary, then turned it into a Claude skill so our team spoke the client's language in every meeting.
A data map. Every source, what it held, and how we could reach it.
One list, then a cut. 43 use cases in one master library with permanent IDs. Then a cut to 29 priced workstreams, sequenced with training first.
Cost the CFO could check. An operating-cost model with seats, tokens, vendors, a visible contingency and the seasonal peak. Every input was tagged by where it came from. Seats were about two-thirds of the monthly cost. Tokens were about 11%. Which platform they picked barely moved the total.
Value the CFO could check. A business case per use case: what the problem costs today, the leading and lagging indicators to watch, and what success might be worth. Every input was labeled client-provided or estimate. Their CEO and CFO relied on it.

A colleague checked integration feasibility, and another reviewed the business cases before they went out.

The finding

In 2025, about a quarter of their 84,738 order lines got no vendor response. Not because the team was too busy. Requests went to one contact per vendor, often an info@ inbox nobody watched. The request email looked like spam. The attached spreadsheet opened as "file corrupt."

I told them how to fix it. No contract needed. They fixed it themselves.

A hard call

Midway through, the federal government designated Anthropic a supply-chain risk for contractors. They'd planned to use Claude, and I'd have picked Claude.

It was their risk, not mine. I sent the legal question to their lawyers, then ran the trade-off: what they'd lose and gain with ChatGPT instead. My advice was to start with one tool the whole company could use safely now, and revisit Claude once things settled.

What happened

About two-thirds of their volume lands in August and September. So the question became: find more bids, or win more of the ones you already see? The emphasis went to win rate.

The 12-month program signed and kicked off in September. I'd left Cadre by then. I wrote the handover guide for the team taking over, and joined the kickoff anyway.

What I learned

The most valuable fix in the whole engagement wasn't AI. Discovery earns its keep when it finds the boring problem.

What I did

I ran discovery across supplier relations, sales and marketing, business development, procurement and IT at both companies, and joined their IT team's annual offsite.

Their language first. The industry runs on jargon none of us knew. I built a 101-term glossary, then turned it into a Claude skill so our team spoke the client's language in every meeting.

A data map. Every source, what it held, and how we could reach it.

One list, then a cut. 43 use cases in one master library with permanent IDs. Then a cut to 29 priced workstreams, sequenced with training first.

Cost the CFO could check. An operating-cost model with seats, tokens, vendors, a visible contingency and the seasonal peak. Every input was tagged by where it came from. Seats were about two-thirds of the monthly cost. Tokens were about 11%. Which platform they picked barely moved the total.

Value the CFO could check. A business case per use case: what the problem costs today, the leading and lagging indicators to watch, and what success might be worth. Every input was labeled client-provided or estimate. Their CEO and CFO relied on it.