The AI Value-Creation Thesis
for the Investment Memo

Open most investment memos to the AI section and you find one sentence: the company will use AI to drive efficiency. That is not a thesis, it is a hope, and a diligence team cannot underwrite a hope. An AI value-creation thesis is specific enough that someone could hold you to it at exit. Here is what that takes.

The vague-thesis problem

A thesis that says "AI will improve margins" commits to nothing, so it delivers nothing. It names no function, no number, no owner and no proof point, which means at the first board review there is no way to tell whether it is on track. The vagueness is not caution, it is the reason the initiative later shows up in the 95 percent of AI efforts that returned nothing.

The four elements of an underwritable thesis

• The named functions: the two or three places in this specific company where AI moves margin first, usually high-volume rules-heavy back-office work like customer service, finance, procurement or HR operations, not a generic "across the business."
• The EBITDA range per play: for each named function, the expected margin or cost impact as a range, with the assumption behind it. A range you can defend beats a point estimate you cannot.
• The 100-day proof point: the one workflow that will ship in the first hundred days to prove the thesis on a single function before scaling, so the fund is buying evidence, not faith.
• The delivery owner: who is accountable for shipping it, typically a fractional AI operating partner in the delivery seat, so the thesis has a name attached and not just an intention.

Where the margin claim comes from

The number in the thesis is only credible if it traces to something real: a baseline of the current cost of the targeted process, a defensible assumption about how much of it AI removes, and a comparable from a company where it has been done. A thesis that cannot show its work is the kind that evaporates in the data room. Sourcing the number properly is the same discipline that lets you attribute the EBITDA at exit.

The sell-side version at exit

The thesis you underwrite at entry becomes the story you tell at exit, so write it to survive the round trip. If the memo named the functions, the ranges, the proof point and the owner, then at exit you can show realized margin against each one and a buyer can underwrite the continuation. If the entry thesis was "we will leverage AI," there is nothing to point to and the AI work adds no multiple. The specific thesis is worth more than the vague one twice: once because it gets built, and once because it sells.

Frequently Asked Questions

1. What should an AI value-creation thesis include?

Four elements: the two or three named functions where AI will move margin in this specific company, the expected EBITDA range per play with its assumption, the one workflow that ships in the first 100 days to prove it, and the named owner accountable for delivery. If any of the four is missing, it is a hope, not a thesis a diligence team can underwrite.

2. How do you write the AI section of an investment memo?

Replace "the company will leverage AI" with specifics: name the functions, give an EBITDA range per play tied to a cost baseline and a comparable, commit to a 100-day proof point on one function, and name the delivery owner. Write it so someone can hold you to it at exit, because the entry thesis becomes the sell-side story.

3. How do you underwrite AI value creation in a deal?

You underwrite it the way you underwrite any value-creation lever: trace the margin claim to a baseline cost of the targeted process, a defensible assumption about how much AI removes, and a comparable where it has been done. A claim that cannot show its work does not get underwritten, it gets discounted to zero.

4. How does AI value creation show up at exit?

It shows up as realized margin measured against the thesis you wrote at entry. If the memo named functions, ranges, a proof point and an owner, you can show the number you actually moved on each and a buyer can underwrite the continuation, adding to the multiple. A vague entry thesis leaves nothing to point to and adds nothing at exit.