How to Evaluate an AI Company’s Business Model Before Investing

Before investing, test whether an AI company can turn a real customer problem into repeatable revenue, deliver the product at a sustainable cost, and fund its operations. Start with the company’s latest filings and financial statements—not its AI label, product announcements, or growth rate alone.

Start by identifying what the company sells and who pays

Describe the business in one sentence: “The company sells

to [buyer] to solve [problem], and charges by [subscription, seat, usage, license, project, or other method].” Use the company’s filings and product materials to fill in each blank. If you cannot, treat that as an unresolved diligence question.

Separate the actual offer from the broad label “AI company.” A business might sell a foundation model, an application built on a model, cloud or compute infrastructure, consulting and implementation, or a bundle of these. The buyer, purchasing budget, and reasons to renew may differ substantially between them. For a diversified software company, also distinguish revenue specifically attributable to AI products from revenue elsewhere in the portfolio.

• Buyer: Who signs the contract, and who uses the product day to day?
• Problem: What task, cost, risk, or delay does the product address?
• Budget: Which existing budget pays for it, or what spending must the customer add?
• Deployment: Is the product sold as software, hosted service, infrastructure, or work that depends on people and implementation?

Reconstruct how revenue is earned

Do not treat all revenue described as “AI revenue” or “subscription revenue” as equivalent. Read the latest annual and quarterly reports, including the business description, management’s discussion and analysis, revenue-recognition footnotes, contract obligations, and risk factors. Look for what customers are charged, when revenue is recognized, and what the company has promised to deliver.

Revenue type	What to establish	What can make it less predictable
Subscription or seat-based	Contract length, renewal terms, seats purchased, and whether customers can expand or reduce use.	Nonrenewals, discounts, unused seats, or a contract that does not reflect actual customer value.
Usage-based or consumption	What is metered, whether a customer commits to a minimum, and how usage charges are recognized.	Volatile demand, customer limits on usage, or a mismatch between the price and cost of serving heavy use.
License	Whether payment is upfront or recurring, what rights the license grants, and what support is included.	Irregular purchases or future support and upgrade costs that are not evident from the license sale alone.
Services and implementation	How much revenue depends on consulting, integration, training, or custom work, and whether those services recur.	Revenue that requires proportional hiring or extensive work for each deployment.
Bundled offering	How the company allocates revenue among software, hosting, support, and services, and whether those items are priced separately.	A headline subscription figure that obscures variable usage, hosting obligations, or labor-intensive delivery.

For example, C3.ai’s fiscal 2026 Form 10-K describes subscriptions with ratable or consumption-based recognition, runtime fees, customer-hosted and vendor-hosted options, and cloud-provider hosting costs. It reported that subscriptions were 91% of total revenue in fiscal 2026, 84% in fiscal 2025, and 90% in fiscal 2024. Those are C3.ai figures for the stated fiscal years, not a benchmark for AI companies generally. C3.ai’s fiscal 2026 Form 10-K.

Check contract duration, renewal provisions, variable fees, bundled hosting or support, and remaining performance obligations or backlog where applicable. These disclosures describe different things: an announcement, booking, backlog, remaining performance obligation, and recognized revenue should not be treated as interchangeable evidence of sales already earned.

Look for proof that customers get value and keep using it

Strong customer evidence goes beyond a pilot, a press release, or a named logo. Look for production deployments, repeat usage, renewals, customer expansion, and outcomes that can be independently verified when disclosed. Ask how much implementation is needed before the product works and whether it becomes part of a workflow that customers would find costly or disruptive to replace.

Check renewals, concentration, and contract quality

Compare customer concentration disclosures across reporting periods. A small number of large customers can support rapid growth, but one lost or reduced contract may have an outsized effect. Read the risk-factor discussion alongside the concentration figures: C3.ai’s fiscal 2026 filing identifies concentration and renewals as risks. A concentration level alone does not determine whether a business is sound; it changes the potential consequences of customer loss. C3.ai’s fiscal 2026 Form 10-K.

• Are customers moving from trials to production, or does disclosed activity remain largely at the pilot stage?
• Do renewal and expansion indicators suggest the product remains useful after the first contract?
• How many customers account for a material share of sales, and is that concentration changing?
• Does the contract require substantial custom work, or can the company deliver a similar product repeatedly?

Map the costs of delivering AI as usage scales

Revenue growth does not show whether each additional customer or unit of usage makes the business more profitable. For each major product or segment, identify the costs that rise with delivery and compare them with the revenue earned from that customer or usage unit.

• Compute and hosting: inference, cloud infrastructure, GPUs or other accelerators, and data transfer.
• Data and model work: data licensing or acquisition, model development, and research and development.
• People: implementation, customer support, human review, and ongoing custom work.
• Commercial effort: sales cycles, customer acquisition, and the time required to move a customer into production.

Then determine which party bears those costs. Does the company pass compute costs through to customers, absorb them in a fixed subscription, or bill variable usage separately? Can it improve efficiency, route tasks to less costly models, or raise prices without harming retention? Verify the answer in company disclosures rather than inferring a cost structure just because the product uses AI.

A 2026 SEC-filed risk disclosure from GridAI Technologies Corp. describes possible risks including volatile usage-based revenue, subscriptions that fail to capture heavy usage, pricing below inference cost, and commoditization that could pressure prices and gross margins. These are mechanisms to investigate, not proof that they apply to another company. GridAI Technologies Corp.’s Form 10-K for the year ended December 31, 2025.

Connect growth to margins, cash flow, and funding

Compare several reporting periods rather than judging the business from one growth figure. Read the income statement and cash-flow statement together: a company can grow revenue while losing money, consuming cash, or relying on further financing.

• Gross profit and margin: Are delivery costs growing faster or slower than revenue, and does the filing explain why?
• Operating expenses: How much is being spent on research and development, sales, and administration relative to revenue?
• Cash generation: What do operating cash flow and capital expenditure show about the cash needed to run and expand the business?
• Funding and dilution: Consider cash balances, debt, financing needs, and stock-based compensation when assessing how growth is funded.
• Path to self-funding: What needs to change for the business to cover its costs, and is that a management target or a result already demonstrated?

Historical figures help explain why this check matters but should not be mistaken for current performance. C3.ai’s fiscal 2025 Form 10-K reported net losses of $288.7 million in fiscal 2025, $279.7 million in fiscal 2024, and $268.8 million in fiscal 2023, plus an accumulated deficit of $1.4 billion as of April 30, 2025. These figures describe C3.ai in those periods only. C3.ai’s fiscal 2025 Form 10-K.

For a private company, public materials may not establish customer retention, gross margins, cash burn, or funding runway. Record those items as unknown unless reliable company disclosures establish them; do not assume the missing figures are favorable.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

Test whether the company has an advantage customers will keep paying for

Ask what makes the product hard to replace and whether that advantage is likely to persist. Possible sources include reliable customer outcomes, rights to valuable data, integration into important workflows, distribution, scale, or switching costs. A technical capability is not automatically a durable commercial advantage if competitors can offer a similar feature cheaply or bundle it into a product customers already buy.

Compare the company’s position against relevant alternatives, including open models, other application vendors, cloud providers, and in-house development. Microsoft’s fiscal 2025 Form 10-K describes significant AI development and operating costs and a rapidly evolving, competitive market. That broad context does not establish a target company’s particular risks; assess those using the target’s own disclosures. Microsoft’s fiscal 2025 Form 10-K.

Consider whether the company depends on a particular model, chip, cloud provider, or data supplier. Changes in access, pricing, or performance from one of those providers could affect the product’s cost or availability. Review target-specific disclosures on privacy, security, intellectual property, regulation, and execution rather than assuming these risks are identical across issuers.

Run downside cases before forming a view

Use the company’s disclosed economics and customer evidence to examine how the business might respond if conditions worsen. For each case, identify the financial or operational mechanism that would transmit the shock to results.

• Production adoption is delayed: Do pilots convert to revenue, or does the company incur sales and implementation costs without broad deployment?
• Inference or hosting costs rise: Can pricing adjust, can efficiency improve, or does the company absorb the increase?
• A large customer reduces use or does not renew: How much revenue and expected usage are exposed?
• A model or cloud provider bundles a competing feature: What customer-paid value remains distinct?
• Lower-cost or open alternatives improve: Is the product differentiated by outcomes, integration, data rights, or service rather than model access alone?
• A failure, privacy issue, or legal restriction occurs: What could it mean for costs, product availability, customer trust, or demand?

Use a repeatable filing-based diligence sequence

1. Write the one-sentence business description. Name the product, buyer, problem, and charging method; flag anything you cannot verify.
2. Rebuild the revenue picture. Separate subscriptions, usage, licenses, services, and bundles; check recognition rules, duration, and commitments.
3. Check customer evidence. Look for production use, repeat usage, renewals, expansion, concentration, and the effort required to implement.
4. Trace unit costs. Identify compute, hosting, data, support, implementation, and R&D exposure, then compare with revenue and pricing.
5. Assess cash and funding. Compare margins, operating costs, cash flow, capital needs, debt, and dilution across periods.
6. Challenge durability. Compare alternatives and dependencies, then test the downside cases against target-specific filings.

For a public company, begin with its latest annual report and quarterly reports, then confirm definitions and periods in the footnotes. For a private company, ask for comparable customer, margin, cash-flow, and concentration information; if it is not available, the uncertainty itself belongs in the assessment.

Keep business quality separate from investment attractiveness

A credible business model does not by itself establish that a company’s shares are attractively priced. Valuation, dilution, governance, investment horizon, risk tolerance, and the company’s latest reported results require separate analysis. Nor does one issuer’s filing establish an industry-wide success rate: the examples above illustrate disclosures to examine, not representative outcomes for AI companies.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.