Hypotheses for data and AI services

Every company says it's doing AI. Very few of them have it running in production, and the leaders on the hook know it.

That gap is where your hypothesis lives. Not in the promise of AI. In the distance between what the company announced and what it can actually ship.

The problem is rarely the model

Data and AI firms tend to lead with capability. Models, platforms, accelerators, a slide of logos. The buyer has seen fifty of those decks this year.

What the buyer hasn't seen is someone who understands why their last three pilots stalled. And the honest answer is almost never the model. It's the data underneath. It's ownership nobody agreed on. It's a security review that takes longer than the pilot did. It's a business sponsor who moved on.

So your misery hypothesis should usually point below the AI layer. That's where the pain is, and that's where a services firm earns its fee.

Where the research comes from

Run 3x3 research before you write a word. Three sources, up to three facts each.

Public record first. Read the last 10-K or the latest earnings release for how leadership talks about AI. A company that names AI as a strategic priority in its filing has made a promise to investors. Job posts tell you the rest. A wave of openings for data engineers and platform roles says the foundation isn't there yet. A posting for a head of AI governance says legal and risk are getting nervous.

Social context next. A new chief data officer is the strongest signal in this category. There's more on that in when a CDO arrives, data budgets follow. Watch what data leaders post. Someone who writes about data quality three times in a month is telling you what keeps them up.

Market last. If a direct competitor just announced an AI feature, the board is asking why they don't have one.

Four hypotheses that tend to land

The pilot-to-production gap. They've run pilots, the demos worked, and nothing is live. Your guess is that the data pipelines and governance weren't built for production load.

The data foundation. Leadership wants AI, but the data lives in a dozen systems with no shared definitions. Your guess is that every AI project starts with months of cleanup nobody budgeted.

The talent squeeze. They're hiring for roles they can't fill. Your guess is that the roadmap depends on people who haven't been hired yet.

The governance stall. Legal and security are slowing everything down. Your guess is that nobody owns the approval process, so every use case gets reviewed from scratch.

Pick one. A message with four hypotheses is a brochure.

An example

Say a mid-sized specialty insurer names AI-driven claims automation as a priority in its 10-K. In the same month it posts six data engineering roles and a new VP of data starts.

You started as VP of data last month, and the 10-K names claims automation with AI as a priority for this year. That puts a public commitment on your desk in your first quarter. My guess is the models aren't the hard part, and the claims data underneath isn't ready for production. Is that accurate, or is the bigger issue something else?

Research hook, personal trigger, misery hypothesis, binary exit. No capability pitch. No mention of your platform partnerships.

If she writes back that the data is fine and the real block is the compliance review, you've learned more than a discovery call would have taught you. That correction is the opening. The binary exit question post covers why.

Write different versions for the committee

The VP of data owns the outcome. The champion is often a data engineering lead who's drowning in requests. The technical lead might be the enterprise architect who'll have to approve whatever you build.

Each one gets a different trigger. The engineering lead's version is about backlog and rework. The architect's version is about the platform decision. Same research, three openings.

What to avoid

Skip the word transformation. Skip anything about the future of AI. Skip the claim that you've done this for a company just like theirs. The buyer can smell a template.

Name what you saw. Guess at the pain below it. Let them correct you.