WellSense Health Plan

Thomas Gilbertson

Automation Strategy Advisor

Thomas Gilbertson has spent ten years driving AI investment programs from inception through postlaunch evaluation. This checklist reflects his 2026 thinking on building AI capabilities from the ground up.

The conventional wisdom in most AI strategy meetings is simple: find the highest-value use case, build it, and the returns will follow. After ten years of discovering, funding, building and delivering AI investment programs, I can tell you this misses much of what actually determines whether an AI investment succeeds.

The truth is less romantic. An AI investment’s payoff has far less to do with model sophistication than with the decisions made before a single line of code is written. Those decisions are often made by default rather than design, shaping the return profile long before anyone measures it.

Here are the seven things nobody tells you before you bet millions on AI:

Size the Ceiling before You Fall in Love 

Every AI effort has a hard cap on the value it can create, and that ceiling is set by raw volume—how much activity the AI actually touches, not by how elegant the technology is. Computer scientists have known this for sixty years as Amdahl's Law: optimizing a small piece of a system can only ever produce a small result, no matter how perfectly you optimize it.

I saw this firsthand. One team spent a year building a sophisticated tool that touched only a few thousand transactions annually. Another built a rougher tool that reached hundreds of thousands. The simpler tool ultimately delivered three times the business value.

Decide How Much Risk You Are Taking, and Take It on Purpose

Risk and return move together. That is not a special AI insight; it is the oldest rule in finance. AI risk comes in distinct layers: the sensitivity of the domain to change and whether you choose to make, buy or reuse.

Buying is typically faster and safer in the near term, especially in fast-changing domains like compliance or tax law complex billing. A capable vendor absorbs the cost of staying current and spreads that burden across its customers. Build internally, and you own that maintenance burden indefinitely.

Building is slower and riskier, but it creates internal capability that compounds. It develops engineering talent, pipeline infrastructure and institutional knowledge that can make the next ten use cases faster and less expensive.

Understand that the Calculus Shifts Over Time

The build-versus-buy decision isn't static. It shifts as non-financial goals outside the use case are achieved and fall away.

Early on, the priority is not just financial return. It is proving that AI can work in your environment, avoiding a highly visible first failure and building organizational readiness. While those goals remain active, buying is often the better choice, even when the use case economics favor building. Early on, you are buying proof and safety.

Once those goals are met, the decision returns to the economics of the use case, and building internally can start to win on its own merits. Staying on the buy path beyond that point can create permanent dependence on another company’s pricing, roadmap, and priorities. 

"An AI investment’s payoff has far less to do with model sophistication than with the decisions made before a single line of code is written."

The only exception that survives is change-sensitive domains. These are often better bought regardless of organizational maturity because the decision is driven by domain volatility.

Understand What You Are Really Paying for When You Pay to Go Fast

Slower, cheaper builds cause an invisible delay in seeing the firstdollar return, while your daily labor line item is staring you right in the face everyday. Higher upfront labor costs push out your investment's breakeven point. But they can pull forward the exact date you see your first dollar of real value. Those are two entirely different clocks. Paying more can mean losing on one while winning on the other.

But there is a second reason AI costs run high that nobody likes to say out loud in board meetings. Sometimes you are not paying for speed, but for credibility.

Bringing in a more seasoned, more expensive team than the technical problem strictly requires can be a deliberate, strategic play. It is what you do to convince skeptical executives that a new capability is safe to trust.

That is the price of belief that you are protecting when you work to avoid a visible failure on your very first swing. In corporate reality, belief is almost always the actual bottleneck.

Decide Whether to Charge the Business Unit and Plan for What That Decision Hides

When a business unit has never used AI before, they have absolutely no idea what a fair internal price for it even is. Choosing not to charge them initially is a strategic free trial. It is how you get skeptics to try something when they have zero reference point for its actual value.

If the business unit did not pay for the AI, the savings those tools generate will quietly disappear. They get sucked right into that unit’s general budget instead of getting counted as an AI win.

You must plan your measurement strategy long before you launch. If you don't, you will eventually find yourself standing before the board swearing your program was a financial success when it actually succeeded invisibly.

The People Who Built It are Rarely the People Who Should Run It

Early-stage AI work requires intense data science, deep process understanding and custom engineering. But mature, operational AI is a different beast entirely. It needs monitoring, support and steady iteration.

The classic mistake in software is hiring the right people for the wrong phase, and then never re-checking the fit. This applies to AI in spades. If you don't re-scope your build teams the moment your work matures, you will end up paying premium builder rates for basic maintenance. And you will keep paying them, for years, without anyone in leadership ever noticing.

Look at All of This Together, Not One Decision at a Time

This is the part almost nobody does. The volume you are working with sets your hard ceiling. Your risk threshold and make-or-buy choices determine how much of that ceiling you can actually reach, and what kind of lasting internal capability you walk away with. Speed decides how fast you get there and what it costs. Your adoption strategy decides whether the win is visible when it happens. And your talent strategy determines whether that win keeps growing or slowly erodes over time.

Over ten years of delivering AI programs from scratch, I have watched countless initiatives stall out. It almost never happens because leaders make a single wrong decision. It happens because they never realized all of these decisions are deeply connected. AI technology is never the hard part. The decisions before it are.

Thomas Gilbertson is the author of AI Bullseye Tactics for Non-Technical Business Leaders available wherever books are sold.