Implementing Agentic AI Ep 5: Prove It, Then Scale It

At another company — not ours — eleven months after their first pilot, the CFO asked the board's question: "So — what have the agents actually earned us?" And the room went quiet. Because nobody had measured the before, so nobody could prove the after. Eleven months of genuinely good work, unprovable — which in a boardroom rounds down to zero. This is the episode that makes sure your silence never happens. It's also the finale — so the playbook goes home with you today. Prove it, then scale it. Part five of five — the finale. For everyone who will face that boardroom question — and everyone who'd rather fund year two with evidence than with enthusiasm. In five minutes: the baseline habit that makes value provable, the four numbers boards actually accept, the failure that finds most companies after launch, and the flywheel that turns one boring win into a programme. Then the full playbook — five rows, five departments — becomes the template you download and fill in for your own organisation. Our last floor: sales operations. The pipeline lives in three systems that disagree — the CRM says forty deals, the spreadsheet says thirty-six, the forecast deck is two Tuesdays old. Reps spend Friday afternoons reconciling instead of selling. Episode two's tests would score this in seconds: busy, rule-shaped, measurable, survivable. But this floor earns the finale for one reason — here, the metrics are money, and money is the language the top floor speaks. Here is the finale's one commandment: measure the before — while the before still exists. For sales operations this week, before the agent arrives: the hours spent reconciling, for every rep, every week. The error rate between systems. The days of forecast lag. Written down, dated, boring, priceless. And notice — every playbook row already did this. Support: the minutes each ticket takes. Finance: cost per thousand invoices. HR: days to first-day-ready. Procurement: renewals caught before lapse. The baseline column is the quiet hero of the whole document — it's what turns "we think it helps" into "here's the delta." Boards accept four kinds of number, and only four. Cost: hours recovered, cost per transaction down. Revenue: deals advanced, renewals saved, pipeline unblocked. Quality: error rates down, first-contact resolution up. Speed: cycle times, forecast lag, days-to-ready. Every agent metric you'll ever present is one of these four wearing work clothes. There is no fifth pillar for vibes. The industry's own survey data this year shows the shift: direct financial impact has overtaken vague productivity as the number one reported return — because "we're experimenting" has stopped being an acceptable answer to a board. And one honesty rule: present every delta net. Subtract what the agent costs to run — the tokens, the licences, the review minutes themselves. Those costs are small and knowable, and subtracting them out loud is what makes the rest of your numbers believable. One failure deserves its warning label, because it arrives after the applause. Industry retrospectives keep finding the same thing: most organisations discover their data infrastructure is lacking only after launching. The agent is fine; the pipes feeding it are not — the CRM field nobody fills, the spreadsheet column that means three things, the export that breaks monthly. You met it in miniature already: finance's duplicate invoice numbers, and this floor's spreadsheet column that means three things. Scale turns small pipe problems into programme problems. Budget for the pipes. So how does one boring win become a programme? A flywheel, four pushes. Prove: the pilot's delta, in the four numbers, against its baseline. Publish: one page — the playbook row — circulated where budget lives. Pick: the next process, scored exactly as episode two taught, often suggested by the last team's neighbours. Repeat: same brief, same gates, same ledger — the method is the flywheel; only the nouns change. Skip the first spoke and you're not scaling a result — you're multiplying an anecdote. Five turns of that wheel is how our logistics company went from one support workflow to five departments in a year — without a single keynote demo. Pocket rule number five: baseline first — scale what you can measure. And with that, the set is complete: five cards, five floors, one method. We'll read them together in a moment — laminate that card, and you've internalised the whole method. And the playbook itself: five rows, five departments, complete. Every row reads the same way: the process, its four-test score, its classification, its written brief, its owner, its gates, its baseline, its lifecycle. One page per process. An audit answer, a budget case and an operating manual in the same document. The empty template is yours — linked below this video. Print the pocket rules on the last page. Fill in one row for one boring, busy, reversible process this month. That's the entire ask. And day one is smaller still: pick the process and measure the before. Everything else in this method waits patiently behind that. One year, five floors. Support answers in minutes, with citations. Finance matches nine thousand invoices with ten minutes of human review a morning. New joiners are first-day-ready in two days. The cleaning contract died its long-awaited death — rejected at a gate, on the record. And sales reconciliation happens continuously, so Friday afternoons belong to selling again. And the people? Nobody was replaced. The invoice clerk became the reviewer at the finance gate. Support handles the strange cases the agent hands over — the work that was always the point. The boring processes weren't anyone's favourite work. That's part of why they went first. No moonshots. No keynote. Five boring, busy, reversible processes — proved, published, repeated. That's what implemented actually looks like. The whole course, in five lines. One: who chooses the next step? Two: first pick — boring, busy, and reversible. Three: if the brief isn't written, the agent is writing it for you. Four: owner, intent, scope, lifecycle — or it doesn't ship. Five: baseline first — scale what you can measure. One method. Five departments. Yours now. That's the series. If it earned it: the playbook template is linked below — and episode one is where to send the colleague who still says "agent" and means "chatbot." The Simple Thinker publishes one short analysis every week — complex ideas, simply explained, no noise. Subscribe for the next one. And when your first boring, busy, reversible process ships behind its first gate — that quiet Tuesday when the ledger just works — that's the keynote. See you there.