Your AI Pilot Year Is Done. Now Comes the Hard Part: Winning the Year-Two Budget

Let me open with the ledger. In 2024, a 3PL ran an AI vision cycle-counting pilot in an 80,000 sq ft warehouse in Dallas. Twelve months, books wide open:

Net gain: $80,000. ROI: 44%. Numbers so clean they look like a textbook.

Then in January 2025, the CFO sent the year-two budget request back. One line of commentary: "A pilot is a pilot. A budget is a budget."

Here is the contrarian take I want on the table first: a successful pilot does not automatically win year-two funding. A pilot proves the technology works. A year-two budget proves the economics scale. Those are two different ledgers with two different formulas. Pilots run on vendor discounts, project-team adrenaline, and executive sponsorship. Scale runs on marginal cost, organizational capacity, and verifiable returns. If you submit a pilot-era formula for a scale-era budget, the CFO will see through it instantly — because your math is missing three ledgers.

Ledger one: the renewal price hike

The price in a pilot contract is almost always a "pilot price." The vendor is betting on your year-two renewal and year-three expansion, so year one comes discounted. At renewal, a hike of 8–15% is normal: pure SaaS subscriptions go up around 10%, hardware-plus-maintenance bundles 12–15%.

You have three levers to push back. First, trade a multi-year term for a discount: sign three years and you can usually hold the increase under 5% — the vendor wants revenue predictability. Second, bundle the expansion: negotiate two or more warehouses together and a further 10% unit-price cut is a reasonable expectation, since the vendor's customer-acquisition cost gets amortized. Third, tie price to an SLA: write recognition accuracy and system uptime into the contract with penalty clauses — the vendor will not dare raise prices recklessly when every extra dollar has a service commitment attached.

The Dallas client faced a 12% renewal hike, from $72,000 to $80,640. The project team went back with all three levers — a three-year term, a second warehouse bundled in, SLA penalties — and landed at $79,200, a 10% increase. That extra $7,200 goes into the year-two budget at full face value. No hiding numbers. Burying a cost is the cardinal sin of budget requests: once a CFO catches it, he will not trust your numbers for the next three years.

Ledger two: the expansion math — marginal cost declines

This is the highest-value table in a year-two budget. Copying from one warehouse to N warehouses changes the cost structure completely:

Read it carefully: warehouse 2 costs 72% of warehouse 1 at the margin; warehouse 3 drops to 64%. Implementation collapses because the integrations, workflows, and training materials already exist. Hardware falls on volume purchasing. Internal labor falls because the team has already survived the learning curve once.

But there is a trade-off you must state honestly: declining marginal cost assumes replication, not reinvention. If warehouse 2 looks nothing like the pilot site — say, an e-commerce small-parcel facility versus an auto-parts bulk warehouse — implementation will not be $18,000. It will be closer to $35,000–40,000, nearly a fresh build. When choosing expansion sites, pick the ones closest to the pilot in business profile, or the marginal-cost curve breaks. Show this table to the boss and he immediately understands why year two should be budgeted as three warehouses at once: renewing a single warehouse alone has the highest unit cost. It is the worst deal on the menu.

Ledger three: the sunk-cost ledger — killing it costs more than keeping it

This is the most underrated weapon at the negotiation table, built to answer the boss's favorite line: "Why don't we pause for a year and see?"

The cost of stopping: $38,000 in hardware written off to zero — cameras and edge boxes have no resale market. Process rollback: rehire the cycle-count team, retrain them; at Dallas-area wages, a four-person count team costs $96,000 a year in labor — money you already saved in year one, handed straight back. Data assets zeroed out: a year of trained recognition models and anomaly-pattern libraries, worthless if you switch vendors.

Line the three ledgers up: renewing costs an extra $7,200 (the hike). Pausing costs $38,000 in dead hardware, $96,000 in labor walking back through the door, plus the data. Which one is the wasteful spend? Not a close call.

And one more real lesson. In Q2 2024, an e-commerce warehouse in Ontario deployed 20 AMRs for pick assist. Six-month pilot: pick productivity up 32%. The team marched in with a year-two request, confident. The CFO asked exactly one question: "Did you hire fewer temps last peak season than the year before?" Silence. The pilot ledger counted "unit pick cost down" but never "total labor cost." Peak season headcount was unchanged — the productivity gain had been swallowed whole by order growth. Not a dollar actually saved. The request went back, the team rebuilt it on "all-in peak-season labor," and won approval three months later. That is the difference between feeling like you saved money and booking the savings. I am allergic to the word "feel" for exactly this reason: feelings do not appear on the P&L.

The year-two budget request: the boss only reads three numbers

With the three ledgers done, the request is one page. Nobody reads a 50-slide deck. Three numbers:

Number one: all-in year-two investment. For the Dallas client's three-warehouse plan: warehouse 1 renewal $91,200 (subscription $79,200 + maintenance $12,000), warehouse 2 buildout $128,800, warehouse 3 buildout $114,800. Total: $334,800. Hike and expansion, every dollar visible.

Number two: verifiable annualized net benefit. Warehouse 1: $168,800 (benefit $260,000, verified over a full year, minus renewal cost $91,200). Warehouse 2: $101,200 (benefit estimated at $230,000 — new sites get a 20% haircut — minus $128,800). Warehouse 3: $95,200 ($210,000 benefit minus $114,800). Total: $365,200. Note the discipline: new-site benefits must be haircut. Projecting pilot-site full benefits onto a new site is the most common number-padding in budget requests.

Number three: payback period. $334,800 ÷ $365,200 × 12 ≈ 11 months. Hold on — let me not do this in my head. Monthly net benefit: $365,200 ÷ 12 = $30,433/month; $334,800 ÷ $30,433 ≈ 11 months. Yes, 11 months. My first mental guess said 6 and it was wrong — which is exactly why every number gets written down and computed, never done in your head. An 11-month payback is healthy for capex (anything under 18 months generally clears the bar).

The computable conclusion: year-two approval = all-in investment $334,800 + verifiable annualized net benefit $365,200 + 11-month payback. Put those three numbers on page one of the request. They beat 50 slides. CFOs do not approve feelings; they approve arithmetic. Year one proved "this technology works." Year two must prove "this money gets cheaper as it scales" — show him the declining marginal-cost curve and the payback math, and the year-two budget becomes an arithmetic problem, not an essay question.

By Lao Zhou, Logistics Cost Analyst