← All notes
Note 08Fundraising

Is it just a wrapper? The five questions behind the question

Build an AI startup, go out to raise, and you'll hear it – usually before you've even shared your screen. It sounds like one question. It's really five: moat, data, unit cost, model risk and proof.

● In short

  • "Just a wrapper?" is shorthand for five concerns: defensibility, data, unit economics, dependence on one model provider and evidence that the product works.
  • Prompts are not a moat. Unique data, deep integration into the customer's workflow and distribution are.
  • Being a wrapper is fine. What matters is what you've built around the model that a competitor can't replicate in a month.

"Is it just a wrapper?" Build an AI startup, go out to raise, and you'll hear this. Usually before you've even shared your screen.

It's a fair question, and a slightly lazy one. Most AI products call someone else's model through an API, so in the literal sense almost everything is a wrapper. What the investor actually wants to know is whether the business survives the model getting cheaper, better and available to everyone. Take a closer look at the product, and the one question splits into five.

Since 2022 our studio has delivered 97 projects, most of them built on top of third-party models. The same five questions are the ones we ask ourselves before a build starts – so here is what a good answer to each one looks like.

Why the question comes so fast

Two years ago a working LLM demo was impressive on its own. Today anyone can get one running over a weekend, and investors have seen hundreds of them. Plenty of the companies behind those demos disappeared the day a model provider shipped the same feature in a release note.

So "wrapper" is less an accusation than a filter. It means: show me the part that's yours.

Just a wrapper? Five questions behind the question: moat – OpenAI ships it tomorrow; data – who owns it; unit cost – at 10× traffic; model swap – in a week; proof – beyond "users love it"
Five questions behind the question

1. OpenAI ships it tomorrow. What now?

This is the moat question. If the core of the product is a clever prompt and a good interface, the honest answer is uncomfortable: a model provider or a well-funded competitor can reproduce it quickly, and the prompt itself will leak the first time someone asks the model the right question.

Prompts are not a moat, I'm afraid. What holds up:

A good answer names which of these you have today and which you're building. Not all three – but at least one that's real.

2. Whose data is it?

The follow-up to the moat question. Investors want to know whether the data you depend on is yours to use, whether customers can take it with them, and whether it becomes more valuable over time.

The best answer, in my view: every new user makes the product better, and this is difficult to replicate. User corrections flow back into evals and fine-tuning, edge cases become test sets, and the product handles the next customer's requests a little better than the last one's. A competitor starting today can copy the interface. It can't copy two years of that loop.

Weaker answers are "we use public data" or a vague reply about consent and storage. Sometimes the customer has to own everything and nothing can be reused – that's a legitimate setup, but then the moat has to come from integration or distribution, and it's better to say so up front.

3. What does one operation cost at 10× traffic?

Unit economics are where wrappers most often break. I have seen demos that look great for 100 users and burn money at 10,000. Long context windows, chains of several model calls, retries and the most expensive model used for every step add up quickly – and in a demo nobody counts tokens.

A convincing answer covers:

If the cost per operation falls as you grow, that's a strong signal. If it doesn't, the investor will do the math before you reach the next slide.

4. Can you replace the model in a week?

Your provider raises prices tomorrow, retires the model your prompts were tuned for, or changes its terms. If your margin disappears overnight, investors will notice – and they'd rather find out how exposed you are before they invest than after.

Two things help you get past that:

The second matters more. Most teams can technically call a different API in an afternoon. Few can say with confidence that the product still works afterwards.

5. How do you make sure that it works?

The answer "users love it" will not work here. Neither will a few testimonials or a great live demo: a demo shows the best case, and investors are asking about the typical one and the worst one.

What they want to see:

Being a wrapper is fine

Many successful products started as a layer on top of someone else's platform. Being built on another company's model isn't the problem – being nothing more than that is.

The model is the easy part. What you build around it is the company.

So the real question isn't whether you're a wrapper. It's this: what have you built around the model that your competitor cannot replicate in a month? The data loop, the integrations, the unit economics, the evals, the process that catches mistakes. That's what investors are trying to find out in those first five minutes.

Before the next pitch

Go through the five questions with your team and write down each answer in two sentences. Wherever the answer comes out vague, that's the slide that will draw the hardest questions – and usually the part of the product worth strengthening first.

Stress-testing these answers before an investor does is a big part of my work as a fractional CTO. If you're preparing to raise soon, get in touch

Case · Mobile · NLP

Keemoji

The model inside the product, not a layer on top: emotion recognition built into the iOS and Android keyboards. Company valuation grew from $20M to $200M+

● More notes

AI made code cheap. It didn't make judgment cheap

AI tools took one team from 5 to 45 pull requests a week while revenue stayed flat. When code is cheap, the bottleneck moves to judgment: what to build, what to kill and how you'll know it worked.

Product5 min read

Fractional CTO: senior tech leadership without a full-time hire

When a company needs technical judgment more than hours. What a fractional CTO does, when the format fits and when it doesn't, and what the first 90 days look like.

Leadership4 min read

Contact

Raising soon? Stress-test the answers

Tell us what you're building and when you plan to raise. We'll go through moat, data, unit cost, model risk and proof with you – before an investor does

Start a conversation →