Where to Put AI First

The mistake every CEO makes in the first six months

Walk into any small-cap or mid-cap company in 2026 and ask the operating leader where they’re putting AI first. Nine times out of ten you get the same answer.

Sales. Marketing. Customer service. The places where vendors send the prettiest pitches.

It is almost always the wrong answer. Not because AI doesn’t work in those places — it does — but because an operating leader of a small-cap or mid-cap business is not running a SaaS company. The leverage in your business is not in the funnel. The leverage in your business is in the one operational moment where, if it happens once, you lose the customer.

Find that moment. Put AI there first. Everything else can wait six months.

This playbook is the framework for finding it.

The customer-losing moment

Every business has a customer-losing moment. It is the specific operational failure that, when it occurs, ends the relationship. Not slows it. Ends it.

For Lindsay Corporation, the $620M sprinkler maker in Omaha, the customer-losing moment is a gearbox that fails at 4pm on a Tuesday in July, while corn is silking and the wind is forty miles an hour. The farmer loses the crop. The farmer does not buy his next pivot from Lindsay. Randy Wood, the CEO, built Lindsay’s entire AI program around that moment. He named the product Smart Pivot, partnered with Taranis for leaf-level imagery, took a 49.9% stake in an Austrian sensor company, and pointed at an $80M MENA subscription deal as proof. The whole thing exists to predict the gearbox failure before it happens. That is a customer-losing moment turned into a customer-keeping moment.

For a third-generation distribution business doing $300M in HVAC parts, the customer-losing moment is a contractor who needs an obscure SKU on Friday at 4:55pm and gets told it’ll be Tuesday. He doesn’t call you next time. He calls SupplyHouse.com. The moment looks like a fulfillment problem. It is actually a relationship problem masquerading as a logistics problem.

For a $180M staffing firm, the customer-losing moment is a candidate who walks the first day of an assignment because nobody from the firm called to check on him. The client cancels the order. The competitor places the next ten. The moment looks like a candidate experience problem. It is actually a churn-prediction problem.

You probably know what yours is. If you don’t, ask the three people in your company who answer angry customer phone calls. They’ll tell you in under five minutes.

Why most AI deployments don’t go there

Three reasons.

First, the customer-losing moment is unsexy. It is operational. It does not produce a press release. There is no dashboard a vendor can demo. The CEO who builds AI around the customer-losing moment is, for the first eighteen months, the CEO who got nothing impressive done. Most CEOs cannot stomach that.

Second, the customer-losing moment is hard to model. The data is messy. The ground truth lives in the heads of the operators who answer the angry calls, and they have never been asked to write it down. The vendors who pitch you cannot solve this without you giving them eight months of cooperation, and most vendors cannot wait eight months for a contract.

Third, the customer-losing moment is political. It usually sits inside a function — operations, service, fulfillment — that has been quietly under-resourced for a decade. The general manager who runs that function does not want a spotlight on the failure mode. He has been managing it through heroics and scotch tape. An AI deployment that names the failure mode out loud is a deployment that puts him under the lamp. That makes him your enemy, even though he should be your closest ally.

The framework — three questions

Here is the test. For every AI use case on your shortlist, ask three questions.

Question one: Does this attack a moment that costs you the customer when it goes wrong?

If the answer is no — if the worst case is “we lose a deal we wouldn’t have won anyway” or “the email response is a little slower” — deprioritize it. The use case might be valuable. It is not the first thing you should ship. The first thing you ship sets the narrative for what AI is in your company. Make the narrative retention, not productivity. Productivity is what the IT department buys. Retention is what the operating leader defends.

Question two: Do you have data nobody else does?

The moat in small-cap and mid-cap operator AI is not the algorithm. The algorithm is a commodity. The moat is proprietary operational data — sensor readings, route logs, service call notes, dispatch records, time-stamped customer behavior — that exists in your business and nobody else’s. If your AI use case is something a vendor could deploy at your competitor with publicly-available data, you do not have a moat. Find the use case where the data only exists because you operated the business for thirty years.

Lindsay’s data moat is the sensor stream coming off forty thousand pivots in soil types and weather conditions Valmont doesn’t have. Rehrig Pacific’s data moat is the wear pattern on returnable pallets across grocery supply chains nobody else owns. Your data moat is hiding in a system somebody on your team is convinced is too messy to be useful. They are wrong. Mess is the moat.

Question three: Will it survive contact with your operators?

This is the question that kills more AI projects than the other two combined. AI deployed over the heads of the operators who run the business will be sandbagged. Quietly, persistently, indefinitely. They will work around it. They will keep the spreadsheet. They will tell the new hire to ignore the system. By month nine, the deployment is a $400K shelfware liability and the CEO’s first instinct is to blame the vendor.

Klarna learned this in public. In 2024 Sebastian Siemiatkowski announced that Klarna’s AI customer service agent had replaced the work of 700 humans, saving $40M a year. By 2025 Klarna was hiring humans back. The system survived contact with the easy 70% of inquiries. It did not survive contact with the edge case where retention was actually decided. The operators on the floor knew that. The CEO did not. The CEO had to learn it the way CEOs learn things — in public.

If you cannot answer “yes, my operators will use this” with high confidence, you do not have a deployable AI project. You have a vendor pitch deck.

The order of operations

Once you’ve found the moment, the order matters.

1. Name the moment in plain language. Not “improve customer experience.” The actual sentence. “We lose 7% of pivots a year because gearboxes fail in July.” “We lose 12% of contractor accounts a year because we miss the Friday 4pm SKU.” Operators will rally around a sentence. Nobody rallies around a strategy deck.

2. Find the SMEs who already know what’s happening. Three to five operators, not consultants. The dispatcher who has been there twelve years. The senior service tech. The customer success lead who can name the last six accounts you lost and why. Their knowledge is the moat. Get it on paper.

3. Build the smallest possible AI thing that addresses the moment. Not the platform. The thing. Lindsay didn’t build “an irrigation AI strategy.” It built Smart Pivot, one product, one job: predict the gearbox failure. Scope creep at this stage is what kills mid-market AI projects.

4. Wire the AI to the operator’s existing workflow. The single biggest predictor of adoption is whether the operator can use the new system inside the system they already use. If the dispatcher has to log into a new dashboard, the dispatcher will not log into the new dashboard. Make it appear in the dispatch screen.

5. Point at the win publicly until the market prices it in. This is a Hopkins move. When the first customer-losing moment becomes a customer-keeping moment, name it. On the earnings call, in the trade press, in the next sales meeting. Specific number, specific operator, specific dollar amount. This is the part most operators skip, and it is the part that makes the next deployment ten times easier to fund.

The closing test

If your current AI shortlist looks like a list of features — chatbots, content generation, “AI-powered” anything — you have a vendor’s roadmap, not yours.

If your shortlist looks like a list of moments — specific operational failures that, when they happen, cost you the customer — you have the start of a real strategy.

Reread the shortlist. Pick the moment with the highest customer-loss cost and the largest proprietary data moat. Put AI there first. Ship it inside the existing operator’s workflow. Point at the result.

Everything else can wait six months.

How we do this for clients

The Ground-Up Workshop is the structured version of what’s in this playbook. Two days in person with a small group of your subject-matter experts, plus a synthesis week. We surface the customer-losing moments, score them against the data moat and operator-survival tests, and walk you out with a target operating model and a 90-day plan to ship the first one.

It’s the alignment step. The deeper work comes after. More on the Workshop here.