Back

Strategy6 min read

18 Nov 2026

What an AI startup pitch deck needs to show investors

What an AI startup pitch deck should show: one real customer job, defensibility beyond the model, cost per task and margin. What investors are tired of.

Alma Johnson

Alma J.

Content Lead at Dtail Studio

Cover with the headline “Pitch the job, not the model.” and a deck outline of six slides led by one customer and one task, with the benchmark table moved to the appendix.

Short answer

An AI startup pitch deck should show the product doing one real customer job, end to end, with the model in the background. Investors want the workflow you replace, the data you can reach that others can’t, the cost per task and who signs the invoice. Model names, benchmark tables and architecture diagrams come last, if at all.

The money is all pointing one way. In Carta’s figures for the first quarter of 2026, over 60 cents of every venture dollar invested on its platform went to AI companies, and inside SaaS the figure reached 83%. The earliest stage looks much the same: AI took 49% of all pre-seed dollars in the first half of 2026.

So the partner reading your deck has read thirty like it this month. Nobody needs convincing that AI matters. They need convincing that you do something specific with it, for someone specific, and that it survives the next model release.


Why do AI startup pitch decks blur together? The benchmark table: Your scores against models you do not control, on tests your buyer has never heard of. It ; The architecture diagram: Boxes and arrows standing in for a product. It says what you built, not what it does for a; “Our model is our moat.”: The one claim a partner can check in a week, by asking their own portfolio founders.

Why do AI startup pitch decks blur together? A summary of this section.

Why do AI startup pitch decks blur together?

Because they lead with the model and keep the customer waiting until slide nine. The customer shows up late, as a persona with a stock photo.

What investors have seen plenty of:

  • The benchmark table.
    Your scores against models you don’t control, on tests your buyer has never heard of. It dates in a month.

  • The architecture diagram.
    Boxes and arrows standing in for a product. It shows what you built, not what it does for anyone.

  • “Our model is our moat.”
    The one claim a partner can check in a week, by asking their own portfolio founders.

None of it is wrong on its own. It’s just the part of the story everyone else is telling too.

What should the product slide show?

One customer job, from the moment it starts to the moment someone approves the result. Name the role, show the screen they’d see on a Tuesday, and keep today’s spreadsheet or queue of PDFs in the picture.

An investor is working out whether a real person would adopt this without a founder on the call. One honest screen answers that better than six bullets about capability.

Having something working is close to table stakes. DocSend’s research on pre-seed decks found that over 89% of startups that raised had some version of their product in alpha or beta. Your deck has to make that version readable in seconds, with real data and the human decision point marked. Our post on what investors want in a deck at each stage covers the rest of the structure.


Comparison of five pitch deck slides built two ways. Opening: the state of AI, versus one customer doing one task. Product: an architecture diagram, versus a real screen with real data. Defensibility: our model is our moat, versus data access, workflow depth, improvement loop and switching cost. Market: trillions then a 1% capture line, versus named buyers and what they pay today. Economics: left to the data room, versus cost per task and gross margin shown in the deck.

What changes when an AI deck is built around the customer job. Framework by Dtail Studio.

How do you show defensibility without claiming a moat?

Show what a well-funded competitor starting on Monday still wouldn’t have in a year. That’s rarely the model. It’s the things around it.

Investors read this slide closely. DocSend measured time spent on the competition slide rising 51%, from about 35 seconds to about 53 seconds across the decks it analysed.

Four kinds of defensibility that survive a model upgrade:

  1. Data nobody else can ask for.
    A signed agreement, a device in a clinic, a system of record you sit inside. Say who gave you permission.

  2. Depth in the workflow.
    You finish a task and hand back something a person can check, rather than answering a question and leaving.

  3. A loop that improves with use.
    Customer corrections make the next output better, and you can show the curve.

  4. Switching cost that’s earned.
    The record of the work lives with you and would be painful to rebuild elsewhere.

Write each one as a sentence a partner could repeat to their investment committee. If you can’t, it isn’t defensibility yet. The same test runs through our post on whether your pitch deck is killing your raise.

What do investors ask about data, margins and cost?

Cost per task now, gross margin now, and where you expect both to go. Put the numbers in before anyone asks.

AI margins read differently from SaaS margins, and investors know it. In Bessemer’s State of AI from August 2025, the fastest-scaling AI companies it surveyed averaged gross margins of about 25%, sometimes negative, while the steadier group averaged about 60%, a little below their SaaS peers. A 60% margin isn’t a red flag at seed. A margin you can’t explain is.

Carta’s Q1 2026 data puts a foundational model Series A at a USD 300m median pre-money valuation, against USD 55m for a non-AI startup at the same stage. Most founders are in the second group, which has to show margin.

Our slightly unpopular opinion: put a margin slide in a seed deck. Founders leave it out because the number isn’t flattering. Yet cost per task over the last six months, what you did to bend it and how pricing holds if inference costs stay flat all read as a team in control.

What does this look like on a real product?

A product that hides its own cleverness. Cinelytic is an AI platform for the entertainment industry, and we’ve worked on it since 2018. That means workflows, dashboards and data visualisation so teams can research talent, assess projects, compare scenarios and decide faster, all held together by a flexible design system. The case study’s heading is “Complexity, without the clutter.” The page says the platform is trusted by major entertainment companies including Warner Bros., Sony Pictures and Lionsgate. “Effective workflow, smooth communication, and remarkable talent across multiple disciplines,” says Dev Sen, Founder of Cinelytic Inc.

The deck was designed in the same scope as the platform. The screens an investor sees should be the screens a customer uses, not a special set made for slides. Our funding round page works the same way, and health products get a stricter version in our post on HealthTech pitch deck design.

How we’d approach it

We’d start from your last investor call, not from a template. For an AI deck that usually means:

  1. Read or watch the last pitch.
    Every question you answered twice is a missing slide.

  2. Pick the one job.
    One role, one task, one output. Everything else moves to the appendix.

  3. Design the product slides from the real product.
    Real data, the decision point marked, the output in its final form.

  4. Write the defensibility and margin slides with your numbers.
    Data access, workflow depth, the loop, switching cost and cost per task, in sentences rather than adjectives.

  5. Build it as a system.
    Type, charts and screen frames as components, so the deck survives three more versions.

Bring your last deck, a recording of a customer using the product, your cost per task and the investor questions you didn’t enjoy. More on how we work.

FAQ

Should we name the models we use?
Once, in small type, and say why you can swap them. Naming a provider as your advantage invites the obvious question about what happens when they ship your feature.

What if we have no revenue yet?
Show use of the real thing: how many tasks ran last month, how many needed a human to fix the output and how that ratio moved. Our post on getting a reply to a first deck covers the rest.

Is the deck different for European and US investors?
Same questions, different order. Keep the data agreements, customer logos and your reporting currency ready either way.

Try this today

Hand your deck to someone who doesn’t work in AI and ask them to describe your product back to you in one sentence. Whatever they say is roughly what a partner will repeat to the rest of the room. If it comes back as a category instead of a job, your first three slides are in the wrong order.

If you want a second pair of eyes on the deck before you send it, let’s talk.

About the author: Alma Johnson is Content Lead at Dtail Studio and writes about brand strategy, positioning and how founders explain what they do.

Sources: Carta, State of Private Markets: Q1 2026 (Ashley Neville, May 2026), Carta, State of Pre-Seed: Q2 2026 (Hamza Shad, August 2026), DocSend, How to nail the 3 most scrutinized sections in your pre-seed pitch deck (August 2021), Bessemer Venture Partners, The State of AI 2025 (August 2025)

Cover with the headline “Eight UX studios for AI startups.” and a brushed metal phone keypad with the 8 key lit.

Strategy6 min read

UX design agency for AI startups: 8 studios to call

Choosing a UX design agency for AI startups: 8 studios in Europe, the UK and the US with AI work you can open, plus what to ask them.

Cover with the headline “Show the real product.” and a browser window with a live product screen and a play button to watch the demo without a form.

SaaS6 min read

AI SaaS website design: how to show your product

AI SaaS website design: what buyers check after an AI recommends you, the four screens to show, the limits to state and why gating the demo costs you.

Cover with the headline “Fixed price or retainer?” and two stacked dark cards: a fixed price card and a monthly retainer card with a switch and months.

Strategy6 min read

Design agency pricing models: fixed price or retainer?

Design agency pricing models: fixed price, sprint or monthly retainer, when each fits a founder, what goes in the contract and the risk of each.

Pick a day

Let’s chat about your project.

A 30-minute video call, no strings attached. We’ll get to know your idea, give you our honest opinion and suggest the next steps.

Free 30-min video call

Thursday 19 November

See times for Thu 19 Nov

Times show in your own time zone

Pick a day

Let’s chat about your project.

A 30-minute video call, no strings attached. We’ll get to know your idea, give you our honest opinion and suggest the next steps.

Free 30-min video call

Thursday 19 November

See times for Thu 19 Nov

Times show in your own time zone

DTAIL®

Based in Barcelona, working globally.

Offices

  • Barcelona
  • California