Key takeaway
An AI business valuation method is a way of turning what is known about an AI product into a price. The methods are the same ones used for software generally, but AI products strain each of them, because their costs, their suppliers and their customers' behaviour are less settled. This guide compares the methods. For the inputs and the evidence buyers ask for, see how to value an AI business.
Why AI businesses are harder to value
Four things make any single method less dependable here than for ordinary software:
- Model-provider dependence. Many AI products run on a model owned by another company. That company sets the price, the usage limits and the terms, and can release a feature that competes with the product.
- Gross margin. Each request costs money to process. A product can have healthy revenue and keep much less of it than a typical software business, and the share can change when usage patterns or model prices change.
- Short history. Many AI products are one or two years old. Twelve months of figures say less about the next three years than they would for an older business.
- Churn. Early sign-ups driven by curiosity often cancel. Churn that looks acceptable as an average can hide early customers leaving quickly.
Each method below handles these problems differently, and none handles all four.
The five methods at a glance
| Method | Starts from | Best used for | Main weakness |
|---|---|---|---|
| Risk-adjusted ARR multiple | Recurring revenue, discounted for margin and dependence | AI products with subscription revenue | Short history makes ARR less reliable |
| Profit multiple | Annual profit | Profitable AI tools with stable costs | Model costs can change the profit quickly |
| Cost to rebuild | What a competitor would spend to copy it | Thin products; setting a floor or ceiling | Ignores customers and distribution |
| Strategic or acquihire value | What it's worth to one specific buyer | Products with unique data, team or users | Can't be estimated from the outside |
| Discounted cash flow | Forecast cash flows | Larger AI businesses with long retention data | Forecasts are weakest where change is fastest |
1. ARR multiple with a margin and dependence discount
What it is. ARR multiplied by a figure, with that figure reduced where gross margin is below software levels or where the product relies on one model provider.
When it's used. For AI products sold on subscription, which is most of the small ones.
Strengths. It uses revenue that can be verified from billing records, and it makes the AI-specific risks explicit instead of hiding them in a lower headline number.
Weaknesses. It assumes today's ARR will repeat. For a product less than two years old with high churn, that assumption carries most of the risk.
2. Profit multiple
What it is. Annual net profit multiplied by a figure, as for a website or a small software business.
When it's used. By buyers who want income now, and for AI tools whose costs have been stable for a year or more.
Strengths. Profit already reflects model costs, so a thin margin shows up without a separate adjustment.
Weaknesses. Profit in an AI business is exposed to prices the owner doesn't control. A change in the provider's pricing, in either direction, changes the figure the multiple is applied to.
3. Cost to rebuild
What it is. An estimate of what a capable team would spend, in time and money, to build an equivalent product.
When it's used. For products that are mostly an interface on top of a general-purpose model, for products with little revenue, and as a check on any other figure.
Strengths. It asks the question every technical buyer asks: could I build this myself for less?
Weaknesses. Code is often the cheapest part to copy. The customers, the integrations into their work, the data and the brand are not in the rebuild estimate, and they are usually where the value is. For a product with paying customers, rebuild cost tends to be a floor. For one without, it can be a ceiling.
4. Strategic or acquihire value
What it is. The value of the business to one particular buyer: a company that wants the team (an acquihire), the proprietary data, the user base or the technology for its own product.
When it's used. When the buyer is a company in the same or a neighbouring market, not an individual buying for income.
Strengths. It can be well above any earnings-based figure, because the buyer is pricing what the product does for their business, not what it earns alone.
Weaknesses. It depends entirely on who the buyer is and what they need at that moment. It can't be calculated from the seller's figures, and a seller can't count on it. VALUERAQ's tool doesn't estimate it.
5. Discounted cash flow
What it is. A forecast of future cash flows, each reduced for risk and delay, then added up.
When it's used. Rarely for small AI products. More often for larger businesses with their own models and several years of data.
Strengths. It lets a buyer model specific scenarios, such as model costs changing or churn settling at a lower level.
Weaknesses. It needs a view of the product's market several years out, which is the hardest thing to know about an AI product. Small changes to the assumptions produce very different results.
Which method VALUERAQ uses, and why
VALUERAQ's AI Business Valuation Tool uses the first method: an ARR multiple, adjusted for risk, with the implied profit multiple shown as a cross-check.
VALUERAQ's default assumption is 1.5× to 4.5× ARR, with a midpoint of 3.0×. That is slightly below its SaaS default, reflecting shorter histories and less certain margins. The tool then adjusts for revenue trend, churn, gross margin, model-provider dependence, customer concentration, age and owner hours. Adjustments multiply together and are capped at −50% and +40%. The range and every adjustment are editorial assumptions, not measured market data, and are published in the methodology and data sources.
This method was chosen because it works from figures the owner can produce and a buyer can verify, and because it shows each discount separately. It doesn't capture strategic value or the quality of the technology, which need expert review.
Worked example: three methods, one business
A hypothetical AI document-summary tool, two years old, with $8,000 MRR and $2,000 monthly net profit. Revenue grew 5% over the last year, monthly churn is 4%, the largest customer is 5% of revenue and the owner works 15 hours a week. Gross margin is 60%, and the product relies on one model provider.
Risk-adjusted ARR multiple. ARR is $8,000 × 12 = $96,000. Most factors sit in the tool's baseline band. Two don't: gross margin between 50% and 70% takes off 7%, and single-provider dependence takes off 10%. The adjustments multiply: 0.93 × 0.90 = 0.837, a combined reduction of 16.3%.
| Default multiple | Adjusted multiple | Value | |
|---|---|---|---|
| Low | 1.5× | about 1.26× | about $121,000 |
| Midpoint | 3.0× | about 2.51× | about $241,000 |
| High | 4.5× | about 3.77× | about $362,000 |
Profit multiple. Annual profit is $2,000 × 12 = $24,000. At the 3× to 5× reference that VALUERAQ uses for its profit cross-check (also a platform assumption), that is $72,000 to $120,000. The ARR midpoint of $241,000 is about 10× annual profit.
Cost to rebuild. Suppose an equivalent product would take one developer four months at $10,000 a month, plus $5,000 for design and setup. These are illustrative figures. The total is 4 × $10,000 + $5,000 = $45,000.
Three methods give roughly $45,000, $72,000 to $120,000, and $121,000 to $362,000. The gap between rebuild cost and the other two is the value of customers who already pay. The gap between the profit and ARR figures is the value of growth that, at 5% a year, hasn't shown up yet. A careful buyer would expect to pay near the bottom of the ARR range, where it meets the top of the profit range.
Why big-company AI multiples do not apply to a small AI tool
Headlines about AI company valuations usually describe venture-funded companies with very large revenue, large teams and investors pricing in years of expected growth. A small AI tool with a few thousand dollars of monthly revenue, one owner and one model provider is a different kind of asset. It is usually bought by an individual or a small company for the income it produces, so the price follows earnings and risk, not a funding-round narrative.
That is why VALUERAQ's default range for AI businesses is 1.5x to 4.5x ARR, not a venture-scale figure. Treat any multiple quoted for a large AI company as a statement about that company. It is not a ceiling or a target for yours.
Which method to start with
| Your situation | Start with | Then check with |
|---|---|---|
| Subscription revenue, at least a year of history | Risk-adjusted ARR multiple | Profit multiple and rebuild cost |
| Profitable, steady costs, little growth | Profit multiple | ARR multiple |
| Little or no revenue, mostly an interface on a model | Cost to rebuild | Ask what customers would pay for |
| Unique data, team or users, and a likely corporate buyer | Strategic value (get expert advice) | Rebuild cost as a floor |
| Larger business with several years of data | Discounted cash flow | ARR multiple |
Common mistakes
- Using one method only. Run at least two and explain the gap.
- Ignoring model costs. Revenue that looks healthy can leave little after inference and hosting.
- Counting one-off revenue as recurring. Setup fees and credit top-ups are not ARR.
- Copying a headline multiple. See the section above.
- Hiding provider dependence. Buyers find it in the first hour of review. Say it first and show the fallback.
- Treating a range as a price. A range starts a conversation; the buyer's checks finish it.
Frequently asked questions
Which method is best for a small AI business?
For one with subscription revenue, a risk-adjusted ARR multiple checked against profit and rebuild cost. If the three are far apart, find out why before relying on any of them.
Why is the ARR multiple lower for AI businesses than for SaaS?
In VALUERAQ's tool the default range starts lower, as an editorial assumption, because AI products tend to have shorter histories and less certain margins. A specific AI business with strong retention and margins can still reach a higher multiple than a weak SaaS business.
Can an AI business with no revenue be valued?
Not with an earnings-based method. It would be priced on rebuild cost or on strategic value to a particular buyer, and VALUERAQ's tool doesn't estimate either.
How do I check whether my AI business is priced fairly?
Run the ARR method and the profit method, work out the rebuild cost, and compare. If the asking price sits well above all three, be ready to explain what a buyer gets that the numbers do not show, such as proprietary data or contracts. Then enter your figures in the AI Business Valuation Tool.
Should I value the AI business on revenue or profit?
Both. Revenue (ARR) shows the size of the customer base. Profit shows what is left after model and hosting costs. A gap between the two is information, so show it. See revenue vs profit multiples.
Does the choice of model provider change the value?
It can. In VALUERAQ's tool, relying on one provider lowers the range by 10%, being able to switch between several leaves it unchanged, and running your own models raises it by 5%. These are default adjustments from the methodology, not market data.
Next step
See the risk-adjusted ARR method applied to your own figures with the free AI Business Valuation Tool.
Written by VALUERAQ Editorial Team. The VALUERAQ editorial team writes and maintains the platform's guides, methodology and glossary. Every guide is reviewed against the published methodology before publication and when the methodology changes.