AI and machine-learning pitch decks — teardowns on data advantage, model cost, and the difference between a product and a feature.
AI & Machine Learning Pitch Deck Examples
AI decks have to answer two questions they rarely pose to themselves: what is defensible beyond the model, and what inference actually costs.
200 teardowns in this category
200 teardowns in this category, spanning decks from 2005–2026. Of the 200 listed below, 183 list a funding stage and 177 disclose an amount raised. Every teardown includes the original deck, slide by slide.
What this collection contains
Measured across the 200 decks on this page. Decks that never disclosed a fact are left out of that figure rather than estimated.
Decks in this collection: 200
Median slide count: 14 (200 of 200 decks report a slide count)
Deck years covered: 2005–2026 (188 of 200 decks state a year)
Median disclosed raise: $13M (154 of 200 decks disclose an amount)
Most common stages: Seed (64), Series A (21), Series B (19)
Most common headquarters: San Francisco, United States (8), United Kingdom (6), United States (6)
The median AI deck in this cohort is 14 slides; half sit between 11 and 19, and the range runs from 5 to 57.
There is no single correct length. The middle of the distribution is tight, and the long tail is almost entirely data rooms and SPAC decks rather than first-meeting decks.
What to do with it: Aim for the 11–19 band for a deck you send cold. If yours is over 25 slides, the surplus is usually appendix material that should move behind the ask.
Product is the most frequently labelled slide, found in 77 of the 196 labelled decks, and it typically lands around slide 5 — after problem (slide 4) and alongside solution and market.
AI founders reach the product demonstration early, because the product is the claim. The abstract capability argument tends to come after, not before.
What to do with it: Show the working thing by slide five. If a reader has to wait until slide nine to see what you built, the earlier slides are doing market-education work your investor probably does not need.
Limitation: Slide labelling is partial, so these are counts of where a labelled slide was found, not evidence that the other decks lacked it.
156 of the 200 decks have a verifiable round size, and the median is $14.25M — but the band spread is wide: 19 under $1M, 47 between $1M and $10M, 39 between $10M and $50M, 41 between $50M and $250M.
"AI deck" is not one fundraising context. The corpus mixes pre-product seed decks with growth-stage and listed-company material.
What to do with it: Benchmark against your own stage, not against the industry median. The stage pages linked below are the honest comparison.
What investors say they want from an AI pitch
The points below come from the investors and accelerators themselves, each linked to the page it appears on. They are guidance, not measurements — treat them as the reader's expectations, and use the cohort numbers above for what decks actually do.
Lead with one declarative sentence. Sequoia's own business-plan guide opens by asking founders to define the company in a single declarative sentence, and warns that this is harder than it looks. In AI, that sentence is where most decks leak, because the model is described instead of the job it does. (Sequoia Capital)
Answer "why now" explicitly. The same guide asks for a clear why now — why has this not been built before? For an AI company that means naming the capability, cost or data change that made your product newly possible, rather than gesturing at the field moving fast. (Sequoia Capital)
Slides are read fast and read badly. Y Combinator tells founders their slides must be legible to someone in the back row, and that investors invest in teams, not slides — the deck's job is to be understood at speed, not to be complete. (Y Combinator 2023)
The market wants AI that completes work. Sequoia's 2025 AI 50 commentary frames the shift as AI moving from answering questions to taking on whole enterprise workflows. A deck positioned around task completion is arguing on the axis investors are currently sorting on. (Sequoia Capital 2025)
A slide order that matches how these decks are read
This ordering follows where labelled slides actually land in the cohort, tightened by the concision guidance investors publish themselves.
One-line definition. What the company does, in a sentence a reader can repeat to a partner who was not in the room. (Sequoia Capital)
Problem, with the person who has it. In the labelled decks, problem sits around slide 4 — early, and named concretely rather than as an industry trend.
Product, shown. Product is the most common labelled slide here and lands around slide 5. A screenshot of the real output beats an architecture diagram.
Why now. The specific capability, cost or data shift that made the product possible in the last 18 months. (Sequoia Capital)
Evidence, in numbers a reader can check. Recurring revenue, usage, retention or design partners — a16z's metrics guide is explicit that recurring product revenue is valued differently from services revenue, so label which you have. (Andreessen Horowitz 2015)
Team, and why this team. Team lands around slide 4 in the labelled decks — earlier than founders expect. Say what each person did before that makes them the right builder here.
The ask, and what it buys. Y Combinator's guidance is to tell the investor how much money you need and what it gets you — a number without a milestone is an unanswered question. (Y Combinator 2018)
Material that usually belongs in an appendix rather than the main deck: model architecture and evaluation detail; full customer list and logo wall; detailed cohort and unit-economics tables; hiring plan by function.
Decks in this cohort worth reading in full
Each links to our teardown of the original deck. We show covers and commentary only; the slides remain the company's.
Ditto. A compact deck that puts the working product early and keeps the technical argument short. Read the Ditto deck teardown
Scribe. Useful for seeing how a product-led AI company frames a workflow it removes rather than a model it trained. Read the Scribe deck teardown
Profound. A recent deck that argues from a category shift, then evidences it — the why-now structure investors ask for. Read the Profound deck teardown
Patterns that separate the strongest decks here
The demo is the claim. Decks that show output land the value argument in one slide; decks that describe capability spend three slides and still leave the reader unsure what the product does.
Distribution is treated as a slide, not an afterthought. In a field where model access is broadly available, how you reach the buyer is the differentiator most decks under-argue.
The market slide names a budget line. The strongest decks in this cohort size the spend they displace rather than the size of "AI".
Mistakes this cohort makes repeatedly
Opening with the model instead of the job. It asks the reader to do the translation from capability to value, and most readers will not. Instead: Open with the declarative sentence Sequoia's guide asks for, then show the product. (Sequoia Capital) Basis: Investor guidance.
Sizing the whole AI market. It signals the founder has not identified a specific buyer, which is exactly what the market slide is meant to demonstrate. Instead: Identify your customer, and if you are inventing a market say so — Sequoia's guide explicitly allows for that case. (Sequoia Capital) Basis: Investor guidance.
A 30-slide first-meeting deck. Half the decks measured here sit between 11 and 19 slides; a 30-slide deck is competing for attention against decks half its length. Instead: Cut to the 11–19 band and move the rest into an appendix you send after the meeting. Basis: Measured in this cohort (n=200).
Ignoring where the product falls under the EU AI Act. The Act sets a risk-based framework with strict obligations for high-risk systems before they reach the market, so a European buyer or investor will ask. Instead: Add one line naming your risk tier and what compliance work it implies. (European Commission 2024) Basis: Regulator.
Checklist before you send an AI deck
The company is defined in one sentence on the first slide.
The product is visible by slide five, with real output.
Why now names a specific change, not a general trend.
Recurring revenue is separated from services revenue.
The deck is inside 11–19 slides, with everything else in an appendix.
Your EU AI Act risk tier is stated if you sell into Europe.
The ask states the amount and the milestone it funds.
See where your AI deck sits in this cohort
Pitch Score reads your deck and places its length, slide order and evidence against the same 200 decks measured on this page.
In this cohort of 200 AI decks the median is 14 slides and the middle half sits between 11 and 19. Y Combinator's own advice is tighter still for a live pitch — around 5 to 7 ideas, legible from the back of a room. (Y Combinator 2023)
What goes on the first slide of an AI deck?
A single declarative sentence defining the company. Sequoia's business-plan guide opens with exactly that instruction, and it is the slide most AI decks spend on architecture instead. (Sequoia Capital)
How much do AI companies raise on these decks?
Of the 200 decks here, 156 have a verifiable round size, and the median is $14.25M. The spread is very wide — 19 rounds under $1M and 10 above $250M — so the stage-specific pages are the more useful comparison.
Do I need to mention regulation?
If you sell into the EU, yes. The European Commission's AI Act sets four risk levels and imposes obligations on high-risk systems before they go to market, so a reader will want to know which tier you are in. (European Commission 2024)
Should the team slide come early or late?
In the labelled decks in this cohort the team slide typically lands around slide 4, earlier than most founders place it.
Methodology and limitations
The cohort is the 200 published deck teardowns on this page (published AI and machine-learning deck teardowns). Nothing outside that set is counted, and the set is frozen so the numbers on this page do not move under you.
Slide counts come from the original deck file, not from a summary or a re-typed copy.
Slide-type positions are only computed over the 196 decks in this cohort whose slides are labelled. That is why every row shows "found in X of Y decks" rather than a percentage — a percentage would imply the other decks lack the slide, when in fact we simply have not labelled them.
Round sizes are the amounts publicly reported for the round the deck was used for; decks with no verifiable amount are excluded from that statistic rather than estimated.
External claims on this page link to the publisher that made them, with the date we checked the page. We do not restate a third-party number without a link to its own source.
We publish deck covers and link to each teardown. We do not republish slides from these decks.