AI startup pitch deck
Investors see many AI decks that demo well and say little about the business. The ones that get funded show a specific job the AI does for a specific customer, usage that lasts, and margins that hold up once model costs are counted.
The numbers
A good demo earns a meeting. These numbers decide what happens after it.
Revenue minus the cost of inference, hosting and any human review. AI companies have often run lower margins than traditional software; a 2020 analysis put many in the 50 to 60% range[1].
On the slide: by quarter, with the plan to improve it: smaller models, caching, better prompts or pricing changes.
How many customers still use the product after three months, and how often. Many AI tools get tried once and forgotten.
On the slide: weekly active users or tasks per account, by monthly cohort.
The time or money the AI saves the customer, measured with them. It sets the price ceiling and the sales story.
On the slide: one customer's before and after, with how it was measured.
How often the output is right enough to use without edits, and what happens when it is wrong. Buyers in regulated work ask this first.
On the slide: on real customer work, with the human review step if there is one.
The same fictional company. Routing simple clauses to a smaller model and caching repeated ones cut the cost per review while acceptance stayed above 90%.
Defensibility
"What stops a larger company from adding this as a feature?" is the most common question in AI pitches. A strong deck answers it on a slide.
Answers that hold up are usually one or more of these: a workflow the product sits inside, so switching costs time; data the product gathers from customers' own work that improves results for them; distribution into a market large companies serve badly; and domain depth such as rules, formats or language that general tools get wrong.
Say which of these you have today with evidence, and which you are building. "Our prompts are better" or "we use the best model" are not answers, since anyone can switch models.
In the meeting
Slide by slide
The order follows the structure most investors expect, from Sequoia's original business plan outline[2] to Y Combinator's seed deck advice[3]. The notes are what changes for AI.
Red flags
Investors give a deck a few minutes. DocSend found average review time fell to under three minutes in 2022[4]. These are the slides that use those minutes badly.
Astitva writes your investor story free from a short brief, in this order, then designs each slide in your own brand for 5 credits. A 12-slide deck is 60 credits.
No card needed. Packs from ₹299 or $3, no subscription.
Twelve slides in investor order with a specific customer job, usage and retention after the first months, value delivered per customer, gross margin after model costs with a plan to improve it, and a clear answer on defensibility.
What stops a larger company copying the product, what gross margin looks like after inference and hosting costs, whether customers keep using it after the novelty fades, and how accuracy is measured.
Often, because inference, hosting and human review cost money on every use. A 2020 analysis put many AI companies at 50 to 60% gross margin. Founders should show how they will improve it.
Yes. Astitva writes the investor story free from a short brief and designs each slide in your brand for 5 credits. New accounts get 50 free credits.
Figures from these sources were current when this guide was written. Rules and benchmarks change, so check the latest version before you rely on one.