How to Build an Investor Pitch Deck With AI, and What to Check

Write the story yourself first in one page: the problem, your solution, proof it's working, the market, how you make money, the team and what you're asking for. Then use AI to turn that page into 10 to 12 slides, tighten the wording and draft charts from your real figures. Finally, check every number, claim and source yourself.

Investors test the numbers far harder than the design, and AI is at its most confident when it's wrong about market sizes, competitors and growth rates. The safe division of labour: AI drafts wording and design from facts you supply, every figure comes from your accounts or a source you've opened yourself, and nothing AI "researched" goes on a slide unchecked.

Follow me on Instagram@sagnikteaches

Before any slides: the one-page story

Ask an AI tool for "a pitch deck for my business" and you'll get slides that could belong to anyone, because it only has your prompt to go on. Investors see many of those. The fix is to give AI your thinking, not ask it to do the thinking. Spend an hour filling in this page in plain words; it's the most valuable hour of the whole process.

Connect on LinkedInSagnik Bhattacharya

ONE-PAGE STORY

The problem:     Who has it, how often, what it costs them today.
Our solution:    What we do, in one sentence a customer would use.
Proof:           Revenue, growth, customers, repeat rate, waiting list.
                 Real numbers with dates.
The market:      How many customers like ours could we realistically
                 reach, and what would they spend? (Count them.)
How we make money: Price, margin per job or customer, what it costs
                 to win one.
Competition:     Who else they could use, and why they choose us.
Team:            Who does what, and why we can deliver this.
The ask:         How much, what it pays for, what it achieves by when.

If you already have a business plan, most of this is in it; writing a business plan with AI covers building one. If you're approaching a bank rather than equity investors, the emphasis shifts to repayment and security, and preparing a business loan application with AI is the better guide.

Subscribe on YouTube@codingliquids

The slide list, and what AI can and can't do on each

Most investor decks follow a similar order. Guy Kawasaki's 10/20/30 rule (ten slides, twenty minutes, no font smaller than 30 point) is a useful sanity check: if you need 25 slides, the story isn't clear yet.

The one-liner on slide 1 shows the gap between AI's defaults and what works. Take a commercial kitchen-repair firm (an illustration). Ask for ten one-line pitches and most come back like this: "Next-generation equipment care for the modern hospitality sector." It's grammatical and says nothing. The version worth keeping was the owner's rewrite of option seven: "We fix restaurant fridges, ovens and dishwashers the same day, so kitchens don't lose a service." The test is whether a head chef would say it down the phone to another chef. Use the AI list for raw material, then pick and rewrite.

Stage 1: Draft the slide text from your one-pager

Paste the one-page story into a capable assistant with a prompt that stops it padding or inventing:

Below is the one-page story for my business. Turn it into text for an
11-slide investor deck in this order: [paste slide list].

For each slide give:
- A headline that states the point as a full sentence
  (e.g. "Demand already exceeds what we can install"), not a label.
- No more than 3 short bullet points.
- A note on what chart or image would support it.

Rules:
- Use ONLY facts and numbers from my story. If a slide needs a fact
  I haven't given, write [NEED: ...] instead of filling it in.
- No buzzwords or superlatives.
- Keep my plain wording where it's clear.

[paste one-page story]

The [NEED: ...] markers are the useful part: they show you where the story is thin before an investor does. Fill those gaps with real information, not by asking AI to "research" them.

Here's the kind of thing that comes back for one slide. This illustrative output is for a specialty coffee roaster whose story said it had 22 wholesale customers in March 2025 and 38 now, with the average monthly order rising from 14 kg to 19 kg:

SLIDE 4: TRACTION  (illustrative AI output)

Headline: Cafés that buy from us stay, and order more each year
- 38 cafés and offices on standing orders, up from 22 in March 2025
- Average monthly order up from 14 kg to 19 kg
- One of the fastest-growing specialty roasters in its market
- [NEED: how many customers have left in the past 12 months]
Chart: wholesale kilograms per month, March 2025 to now

Three fixes before it goes near a slide. The third bullet is a superlative the prompt banned; rules are followed most of the time, not always, so read for them. The headline says customers "stay", which only the [NEED] figure can prove: the roaster checks its invoicing system and finds 3 of the original 22 have stopped ordering, so the headline becomes "19 of our first 22 wholesale customers still order every month". And the chart note is fine, but the chart itself gets built from the sales export, not from these bullets.

Stage 2: Build the numbers in a spreadsheet, not in the chat

Three sets of numbers carry an investor deck, and all three belong in a spreadsheet you control:

• Traction: revenue by month or quarter, customer counts, repeat rates, straight from your accounts and CRM.
• Unit economics: average sale, gross margin per job, and what it costs to win a customer. Investors often go straight to this slide.
• Projections: three years, driven by a few assumptions you can defend (jobs per month, average value, margin, overheads). Show a likely and a downside case; scenario planning with AI shows how to build both.

The unit economics slide is a short sum, and it's worth doing by hand once so you can explain every line of it. Take a meal-prep kitchen selling weekly subscriptions (illustrative figures from one quarter):

The line investors probe is the seven months. If it's an average across everyone who ever subscribed, a handful of loyal early customers can hide the fact that most recent sign-ups leave after two. Build it from a single group, such as everyone who joined in one month a year ago, and say so on the slide. Ask AI to check the logic of the table if you like, but the figures come from the platform export.

For the market slide, build the estimate from the bottom up: how many customers like yours you could realistically reach, multiplied by what they'd spend with you. The usual labels are TAM (total addressable market, everyone who could conceivably buy), SAM (serviceable available market, those you could actually serve) and SOM (serviceable obtainable market, what you could win in a few years). A bottom-up SOM you can explain beats a vast headline figure you can't. AI is useful for checking the logic of your estimate; it's unreliable as the source of the figures, and its arithmetic needs checking too.

Back to the kitchen-repair firm, filled in. The counts come from a business directory search of its 45-minute service area; the spend figure comes from the firm's own invoices:

BOTTOM-UP MARKET ESTIMATE  (illustrative)

Food businesses within 45 minutes of the depot:  3,400
Average repair and maintenance spend per site:   $2,800 a year
   (median of 60 current customers' invoices, last 12 months)
Serviceable market (SAM):   3,400 x $2,800 =    $9.52 million a year
Share we could win in 3 years:                  6% = 204 sites
Obtainable market (SOM):    204 x $2,800   =    $571,200 a year
We serve today:                                 71 sites

Then paste the block into the chat and ask: "Act as a sceptical investor. List the three assumptions in this estimate you'd challenge first, and what evidence would answer each." An illustrative reply: the $2,800 comes from existing customers, who may be heavier users than the average site; 6% share needs a reason, such as how many sites the firm won last year; and the 3,400 includes businesses tied to a manufacturer's service contract. All three are fair, and each points to evidence you already hold rather than a figure AI should supply. Notice the reply doesn't contain a single new number. That's the right way to use it here.

Stage 3: Design it with an AI presentation tool

Once text and numbers are settled, design is where AI saves the most time with the least risk. Pick the tool that matches the software you present in:

With Copilot, the route is to save your slide text as a Word document and use "Create presentation from file" from the Copilot button in a new PowerPoint presentation, as described in Microsoft's guide. Whatever the tool, rebuild every chart from your spreadsheet rather than letting the tool redraw numbers, apply your own brand colours, and keep to one message per slide. Don't use AI-generated images that could be mistaken for real photos of your team, premises or completed work.

The redraw problem is easy to miss. In one illustrative case, a three-shop bakery pasted its yearly revenue into a deck builder: $610,000, $655,000, $720,000. The tool drew a bar chart with the vertical axis starting at $600,000, so an 18% rise over two years looked like the business had grown twelvefold. Nobody on the team noticed, because the bars "looked right". The first investor to read it asked where the axis started. Rebuilt in the spreadsheet with the axis at zero, the growth looks modest and honest, which is what gets a second meeting.

Stage 4: What to check before it goes to anyone

Print the deck and go through it with a pen. This checklist catches the errors that end meetings early:

Numbers

• Every figure traces to a named source: your accounts, CRM, or the projection spreadsheet.
• The same figure is identical everywhere it appears (revenue on the traction slide matches the financials slide).
• Use-of-funds lines add up exactly to the amount you're asking for.
• Percentages are calculated from the underlying numbers, by you, not copied from an AI draft.

Market and industry claims

• Every external statistic has a source you have opened and read, with its date. AI tools can produce convincing figures and even citations that don't exist; AI hallucinations explained covers why.
• If you can't find a source, delete the claim or replace it with your own data.

Competitors

• Check each competitor's current website for every claim you make about them. AI's knowledge of small competitors is often patchy or out of date.
• Describe them fairly. Investors know the market and will notice a straw man.

Permissions and confidentiality

• Customer logos and testimonials only with the customer's permission.
• No client names or details that were shared with you in confidence.
• Draft in a business AI plan that doesn't train on your content, or a personal plan after switching off model training in its privacy settings. Don't paste your full accounts or shareholder details into a free account.

Consistency and tone

• Company name, dates and the ask identical on every slide.
• Read every headline aloud. Anything that sounds like marketing copy rather than something you'd say across a table gets rewritten.

Rehearsal questions

Last, use AI as a rehearsal partner rather than a writer. Paste the finished slide text and ask:

You are an experienced investor reading this deck for the first time.
Ask me the eight questions you'd most want answered in the first ten
minutes of the meeting, hardest first. Don't soften them, and don't
answer them for me.

For the coffee roaster, the top of the list might come back like this (illustrative):

1. Green coffee prices swing a lot. What happens to your margin if
   they rise by a third, and how quickly can you pass that on?
2. Your three largest wholesale customers: what share of revenue
   are they, and are any on contracts?
3. You're asking for a second roaster. What's your current capacity
   use, and when do you run out?
4. Why would a café switch from its current roaster to you?

Each is a fair question, and the deck answered none of them. The roaster could answer the second and third from its own records within ten minutes, so they went onto a backup slide kept after the ask. The first needed real work: a line in the projections showing margin at current prices and at a third higher. Questions you can't answer in rehearsal are the ones that come up in the meeting.

Worked example: an electrical contractor raising $500,000

An illustration. Say a 14-person electrical contractor turns over $2.1 million. Home EV charger and battery installations have grown from $210,000 to $380,000 of revenue in a year, and the firm turns away about 60 of these enquiries a month for lack of qualified installers. The owner wants to raise $500,000 to set up a dedicated installation team: four electricians, two vans, training and marketing.

• One-page story: written by the owner in 90 minutes. The proof section uses the firm's own figures: 81% growth in the installation line, the turned-away enquiries, and a margin per installation from job costing.
• AI draft: the Stage 1 prompt produces 11 slides of text with four [NEED] markers, including the cost of winning a customer, which the owner hadn't worked out. The office manager pulls it from the marketing spend and enquiry records.
• Spreadsheet: projections driven by installs per month (ramping from 15 to 45 over 18 months), average installation value and margin, with a downside case at two-thirds of the ramp.
• Design: slide text saved as a Word document and turned into a draft deck with Copilot in PowerPoint, then restyled with the firm's colours; every chart rebuilt from the spreadsheet.

The Stage 4 check catches three problems. An early AI draft had described the home charging market as "growing 30% a year", with no source; it's removed and replaced with the firm's own growth figure, which is more persuasive anyway. The competition slide said a rival offered no battery installations; the rival's website shows it does, so the slide is corrected. And the use-of-funds lines add up to $520,000, not $500,000; the marketing line is trimmed. Each of those would have cost credibility in the first meeting.

Why decks built with AI get put aside

• They sound like everyone else's. Phrases like "disrupting the industry" and "innovative solutions" signal an unedited AI draft. Your own plain words are more convincing.
• Projections with no drivers. A line that goes steeply up and to the right, with no jobs-per-month or margin assumptions behind it, invites the question "why?" you can't answer.
• Numbers that disagree. The fastest way to lose an investor's confidence is two different revenue figures in the same deck.
• A hidden ask. State the amount, what it buys and what it achieves, on its own slide, near the end.

The same approach works for other persuasion decks: for winning customers rather than investors, see writing a sales deck with AI.

Questions about using AI for an investor deck

Do investors mind if a deck was made with AI?

They mind decks that read as generic, and AI-drafted text often does unless you rewrite it. Nobody objects to AI helping with layout or wording. What damages credibility is an invented statistic, a competitor description that's out of date, or numbers that don't match between slides. If you can explain every figure and its source in the meeting, how the slides were produced doesn't matter.

Can AI write my financial projections?

It can help you structure the model, list the assumptions to include and challenge ones that look optimistic. It shouldn't produce the numbers. Build projections in a spreadsheet from your own accounts, with each year driven by a small number of assumptions you can defend, such as jobs per month and average job value. Then check that every figure in the deck matches the spreadsheet exactly.

Which AI tool is best for building a pitch deck?

Use the one inside the software you already present with. Copilot in PowerPoint needs a Microsoft 365 Copilot licence; Gemini in Google Slides comes with Google Workspace Business Standard and above. Standalone deck builders such as Gamma produce designed slides quickly and export to PowerPoint. As of September 2026 features change often, so try the free tier or trial with your one-page story before paying.

Further reads

• How to Use ChatGPT as a Business Adviser Without Being Misled — Get tough questions from AI before investors ask them.
• How to Run a SWOT Analysis of Your Business With AI — Sharpen the competition and risks slides.
• How to Forecast Next Quarter's Sales With AI Using Your History — Ground the near-term projections in your own history.
• How to Build a 13-Week Cash Flow Forecast With AI Help — Show investors you manage cash week to week.
• How to Create Customer Personas With AI From Real Data — Describe your customers from data, not guesswork.
• How to Track Profit per Project in a Service Business With AI — Reliable job margins make the unit economics slide credible.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Microsoft Support (Create a presentation with Copilot in PowerPoint); Google Workspace plan pages; Guy Kawasaki's 10/20/30 rule. The worked example is illustrative.