The 90-Second Investor Pitch

Capital for AI ventures is abundant and concentrated at the same time. In 2025 AI absorbed $211 billion of venture funding, up 85% year over year, with 58% of that money in megarounds of $500 million or more (Crunchbase News, 2025). The implication for a founder is blunt: the money is there, but it flows to teams who can articulate a defensible wedge in one breath. The plan exists to make that breath repeatable on paper.

Use this fill-in structure as the spine of your executive summary. Investors read it in roughly the order below, and they stop reading the moment one clause rings hollow.

The single clause investors interrogate hardest is the wedge. A pitch that says "we use AI to do X" is not a wedge; "we hold the only labelled dataset of Y, and accuracy improves with every customer" is. The rest of this guide gives you the numbers to back each blank with evidence instead of adjectives.

A second clause that trips founders is the raise amount. The instinct is to ask for as much as the market will bear, but a seed cheque is priced against a milestone, and asking for $3 million to reach a milestone a disciplined team would hit on $1.2 million reads as either inexperience or dilution you will regret. The plan should tie the number to a concrete next inflection: a specific ARR figure, a model-accuracy threshold that opens a regulated use case, or a count of reference logos that de-risks the Series A. When the milestone is legible, the amount defends itself, and you keep the negotiating room that vague asks throw away.

Finally, sequence the summary so the reader never has to hold a question in their head. Problem, then the buyer who feels it, then your solution, then why it is hard to copy, then the economics, then the ask. Each sentence should resolve the doubt the previous one raised. When an investor finishes the paragraph with no open loops, you have earned the second meeting, which is the only thing an executive summary is actually for.

Market Size, Funding & Demand

The global artificial intelligence market was valued at roughly $390.9 billion in 2025 and is projected to grow at a 30.6% compound annual rate to about $3.5 trillion by 2033 (Grand View Research, 2025). North America held the largest regional share at 35.5%, and the services segment took 36.3% of spending, a signal that buyers are paying for applied outcomes, not just raw models. For a founder, the headline number matters far less than which slice of it you can credibly serve.

Funding tells the more useful story. In 2025 more than $15 billion went into AI-focused seed rounds alone, up about 50% on the prior year, and 27 seed rounds exceeded $100 million (Crunchbase News, 2025). At the same time, deal counts below $10 million fell as money concentrated into larger cheques. The practical reading: capital is generous to credible AI teams and stingy to undifferentiated ones, so the plan's job is to land you in the first bucket.

One more point most guides skip: the addressable market for an applied AI company is not the trillion-dollar total. It is the budget your specific buyer already spends on the workflow you replace. Plans that anchor on the $390.9 billion figure as their serviceable market lose credibility instantly. Anchor instead on the line item your product displaces, then show how many of those buyers exist. For neighbouring tech-sector framing, our free business plan templates library covers SaaS and platform models you can borrow structure from.

Reading the Competitive Field

An AI startup competes on three fronts at once, and a plan that maps only one of them looks shallow. The first front is the foundation model layer: OpenAI, Anthropic, Mistral AI, and Cohere set the floor for raw capability, and any product that merely passes a prompt through to them inherits their economics without their scale. The second is horizontal tooling: Hugging Face and the developer platforms that make models easy to deploy, which lower the barrier for everyone, including your future imitators. The third, and the only one you can win, is the applied vertical where a product like Perplexity carved out a defensible position not by training a better base model but by owning a use case and a distribution channel.

The lesson for your plan is to compete where the giants will not bother to follow. OpenAI is not going to build claims-adjudication software for mid-market insurers; Anthropic is not going to chase dental-practice scheduling. Those narrow, unglamorous verticals are where proprietary data accumulates and switching costs harden. When you name competitors, name them honestly: list the foundation providers you depend on, the horizontal tools a copycat would use, and the two or three vertical players actually fighting for your buyer. Then show the specific reason a customer picks you, accuracy on their data, an integration nobody else has built, or a compliance posture a generalist cannot match.

Investors have grown allergic to the phrase "no direct competitors." It almost always means the founder has not looked hard enough, or the market is too small to have attracted anyone. A credible competitive section acknowledges the crowded base layer, concedes what the incumbents do better, and then makes a sharp, evidenced case for the slice you defend. That honesty reads as confidence; the alternative reads as naivety.

Defining the Buyer You Actually Serve

The strongest AI plans describe a buyer so specific you could find ten of them by the end of the week. A weak plan describes "enterprises" or "businesses adopting AI," which tells an investor you have not yet talked to a real customer. Pin down who signs the cheque, who feels the pain, and what event triggers the purchase, because those three rarely sit in the same chair.

• Economic buyer: the person whose budget the product comes out of, and the metric their bonus depends on. Your value has to map to that metric.
• End user: the person who lives in the product daily. If they resent it, adoption stalls no matter who bought it.
• Trigger event: the moment the status quo becomes intolerable, a compliance deadline, a headcount freeze, a competitor's launch, or a cost line that finally grew too large to ignore.

Quantify the segment. How many organisations fit the profile, what do they already spend on the workflow you replace, and how fast can you reach them through search, partnerships, or outbound. An AI product that saves a 40-person operations team 12 hours a week has a different pitch from one that improves a metric by two percentage points; both can be excellent, but the plan has to make the value legible in the buyer's own language. Positioning clarity is what converts a curious prospect into a paying account, and it is the cheapest lever a pre-revenue company has.

What It Costs to Build & Raise

A lean AI startup that builds on existing foundation models and managed cloud can reach a working product on $29,000 to $50,000 (about £22,000 to £40,000). A funded build with two or three engineers, real data work, and twelve months of runway runs $120,000 to $250,000 (£95,000 to £200,000). The wide spread comes almost entirely from one decision: whether you fine-tune or train your own models, or call a provider API and spend your capital on product and distribution instead.

Where the First Cheque Goes

• Compute and cloud credits (training + inference): $8K–$90K (£6K–£72K). The most volatile line; managed APIs keep it low, self-hosted GPUs push it high.
• Founding engineering / ML salaries (first 6 months): $10K–$80K (£8K–£64K), often deferred or equity-heavy pre-seed.
• Data acquisition, labelling and licensing: $3K–$35K (£2K–£28K). This is where defensibility is bought.
• Product, design and developer tooling: $4K–$25K (£3K–£20K).
• Legal, IP, incorporation and AI compliance review: $4K–$20K (£3K–£16K).

How AI Founders Actually Fund This

Three routes dominate. Equity is the default: 2025 was a record year, with over $15 billion in AI seed rounds and eight- and nine-figure seeds becoming routine for strong teams (Crunchbase News, 2025). Cloud credit programmes from major providers offset $5,000 to over $100,000 of compute for qualifying startups, effectively non-dilutive runway. Grants and accelerators round it out: in the UK the British Business Bank Start Up Loan offers up to £25,000 at 6% fixed with mentoring, and Innovate UK runs AI-specific competitions. A credible plan names which mix you are pursuing and times the raise to a milestone, not to a month on the calendar.

Modelling Burn and Runway Honestly

Compute makes AI burn behave differently from classic software burn, and investors know it. A SaaS company's costs are mostly fixed payroll; an AI company's costs scale with usage, so a viral launch can torch your runway faster than a quiet one. The plan should show a monthly burn that separates fixed costs (salaries, tooling, rent) from variable inference cost, and it should model what happens to runway if usage triples. A raise that buys eighteen months at flat usage might buy only ten at the growth rate you are promising, and naming that tension before the investor does is a mark of seriousness.

Build the model with a sensitivity table on cost per token or cost per call. When a provider cuts inference prices, which happens regularly, your margin improves and your runway extends; when you push model quality up, the opposite. Showing both directions tells an investor you have actually run the numbers rather than pasting a template. The founders who raise fastest are the ones whose financial model survives the first hard question, and the first hard question is almost always about compute.

How AI Startups Make Money

Three pricing models cover almost every applied AI company. Seat-based SaaS charges per user per month and reads cleanly to investors who know software. Usage or token-metered billing ties revenue to consumption, which protects margin but makes forecasting noisier. Outcome-based pricing charges for a result, a resolved ticket, a qualified lead, a closed claim, and commands the highest willingness to pay when you can prove the result. Most vertical AI software lands an average contract value between $6,000 and $60,000 a year.

The number that separates a serious AI plan from a generic SaaS plan is gross margin net of inference. Every query a customer runs costs you compute. At launch, before you optimise prompts, cache results, and route to cheaper models, margins often sit near 40%. Once inference is engineered down, well-run applied AI companies reach 55% to 80%. If your plan shows 90% margin like a pure software business, an investor will assume you have not modelled compute and will discount everything else you wrote.

A Worked Example

Take a vertical AI SaaS with 120 paying accounts at a $750 monthly ACV. That is $90,000 in monthly recurring revenue, or roughly $1.08 million ARR. At a 68% gross margin after inference and cloud, the company keeps about $734,000 in gross profit before headcount and go-to-market. Layer in a sales-assisted motion and that gross profit funds two or three commercial hires while leaving the engineering team intact. Show the investor this arithmetic explicitly, including the inference line, and the conversation shifts from "is this real" to "how fast can you grow accounts."

Retention and Expansion Are the Real Story

New logos get the attention, but net revenue retention is what compounds an AI company into something valuable. Because applied AI products improve as customers feed them data, a well-built product should get stickier and more useful over time, which shows up as expansion revenue: more seats, more usage, higher tiers. A plan that projects 115% to 130% net revenue retention, and explains the mechanism, more accuracy, more workflows, more departments, tells a far stronger story than one chasing an ever-growing top of funnel. The mechanism matters; investors have learned to discount retention claims that rest on hope rather than on a product that genuinely deepens with use.

Watch the failure mode too. If your product solves a one-time problem, retention collapses no matter how good the technology is, and the plan needs an answer: a recurring trigger, an expanding surface area, or a data asset that keeps the customer locked in. Naming the retention risk honestly, then showing how the product design defends against it, is exactly the kind of detail that separates a fundable AI plan from a polished pitch deck with nothing underneath.

Three Ways to Structure the Company

Most AI ventures fall into one of three structural archetypes, and investors price them differently. Your plan should declare which one you are, because each carries a distinct cost base, margin profile, and moat argument. Picking a lane up front prevents the most common pitch confusion, claiming the capital efficiency of one model and the defensibility of another.

The applied SaaS lane is where most first-time founders should start, because it reaches revenue fastest and lets you accumulate the proprietary data that later becomes the moat. The infrastructure lane needs the largest raise and the strongest technical pedigree. The services lane is underrated: it generates cash early and quietly builds the dataset that can convert it into software. State your lane, then defend the margin and moat that go with it.

Reaching the First 100 Customers

Distribution kills more AI startups than technology does. The product works, the demo dazzles, and then the company spends eighteen months discovering it has no repeatable way to put the product in front of buyers. Your plan should treat go-to-market as a first-class engineering problem, with a stated motion and a cost of acquisition you can defend.

For applied AI selling into a vertical, three motions dominate the early stage. Founder-led sales gets you the first ten to thirty accounts and, more importantly, the objections that reshape the product; no SDR can do this learning for you. Wedge-led product adoption works when the product solves one painful task so well that a single team adopts it without procurement, then expands; this is how most modern software lands. Partnership and integration distribution borrows reach from a platform your buyer already trusts, an EHR, an accounting suite, a CRM, in exchange for a revenue share.

Put a number on it. If your average contract value is $17,000 and a blended acquisition cost is $4,000, you recover the cost in roughly three months and the payback story sells itself. If acquisition costs $14,000 against the same contract, the plan has to explain how that compresses with scale, or investors will assume the unit economics never work. The honest version, naming the channel, the cost, and the payback period, always beats a vague claim that "the product sells itself."

AI gives you one unusual advantage worth flagging in the plan: the product itself generates proof. Every task it completes is a data point you can turn into a case study, a benchmark, or a referral. Founders who instrument this from day one build a compounding marketing asset; those who treat marketing as a separate function pay full price for every customer. Show the investor you understand that the product and the go-to-market are the same flywheel.

Compliance & Legal Requirements

AI is not a licensed trade in the way a restaurant or a clinic is, but it carries fast-moving compliance exposure that investors now expect to see addressed in the plan. The reach is what catches founders out: obligations attach to where your output is used, not only to where you are incorporated.

European Union

The EU AI Act is the binding framework, and it applies on a territorial-effect basis, so a US or UK company whose product is used inside the EU can be in scope (European Commission, 2025).

• Prohibited AI practices have been banned since 2 February 2025.
• General-purpose AI model obligations applied from 2 August 2025.
• High-risk Annex III systems (recruitment, credit scoring, and similar) must comply by 2 December 2027.
• Penalties reach EUR 35 million or 7% of global turnover, whichever is higher.
• SMEs and startups get priority access to regulatory sandboxes if they qualify.

United Kingdom

The UK runs a principles-based, regulator-led approach rather than a single statute. The AI (Regulation) Bill was reintroduced on 4 March 2025, proposing a central AI Authority, and an AI and Digital Hub advisory service helps founders navigate sector rules. In practice, your binding obligations today come through the Information Commissioner's Office under UK GDPR: expect to run a Data Protection Impact Assessment for any product that processes personal data, typically a £3,000 to £16,000 piece of work with counsel.

United States

Federal policy is deliberately deregulatory under the 2025 America's AI Action Plan, but the action has moved to the states. The Colorado AI Act and California's transparency rules impose duties on high-risk and consumer-facing systems, so a US-facing AI startup should budget $5,000 to $25,000 for compliance counsel and map which state regimes touch its buyers.

Data, IP, and the Questions in Due Diligence

Beyond the AI-specific statutes, two ordinary issues sink more deals than founders expect. The first is data provenance: an investor's lawyers will ask where your training data came from and whether you have the rights to use it. A proprietary dataset built on scraped content you do not own is a liability dressed as a moat. Document your data sources, your licences, and your consent basis before you raise, not during the data room scramble. The second is IP assignment: make sure every founder, contractor, and early engineer has signed their work over to the company. A capable model that legally belongs to a freelancer who built it on a weekend is a problem that surfaces at the worst possible moment, usually mid-acquisition.

The practical takeaway for the plan is a short, honest compliance posture: name your EU AI Act risk tier, state your UK and US data obligations, confirm your IP is assigned and your data is licensed, and note the budget you have set aside. Founders who address this in a paragraph look in command of their business; founders who omit it entirely invite the investor to assume the worst.

Mistakes That Kill the Raise

We have reviewed enough AI plans to see the same five failures repeat. Each is avoidable, and each is the kind of thing a sharp investor spots in the first two minutes.

• The thin wrapper. A plain interface over a public model with no proprietary data, workflow, or distribution. If a competitor can rebuild your product in a weekend, you do not have a company yet, you have a feature.
• Round-number projections with no inference line. Revenue that climbs in suspiciously clean multiples, no compute cost modelled, no source cited. This is the single fastest way to lose a lender or an investor.
• Ignoring EU AI Act exposure. Founders assume a US or UK base keeps them clear. It does not; the Act follows the output. A plan that never mentions risk tiering looks naive.
• Underpricing inference. Charging a flat low price while every power user erodes your margin. Without metering or tiering, growth makes the unit economics worse, not better.
• No moat beyond being early. "We were first" is not defensible when a better-funded team can copy the idea next quarter. Name the compounding advantage, or build one before you raise.

Sample Business Plan Preview

Here's an extract from an AI startup plan written by our team, so you can see the level of specificity investors expect:

What's in the Template

Every Avvale AI startup plan template includes these sections, pre-structured for a fundraising audience:

• Executive Summary, the 90-second pitch frame, built to land the wedge in one paragraph
• Problem & AI Solution, the quantified pain and exactly how your model addresses it
• Market & Wedge, serviceable market anchored on the budget you displace, not the trillion-dollar total
• Moat & Defensibility, proprietary data, compounding workflow, or distribution, with evidence
• Business & Pricing Model, seat, usage, or outcome pricing with a stated gross margin net of inference
• Go-to-Market, the specific buyer, the wedge motion, and the path to repeatable acquisition
• Compliance Posture, EU AI Act risk tier, UK/US data obligations, and what you will do about them
• Team & Milestones, founder credibility plus the milestone the raise actually buys

The optional Financial Forecast add-on (included in our $300/£250 and $1,000/£800 packages) provides a five-year Excel model with ARR build, gross margin net of inference, burn, runway, and a sensitivity table on compute cost, the exact view a seed investor asks for. You can also pair this with our market research & content service to have the whole narrative written for you.

Frequently Asked Questions