Khaled Azar 
Here is the question a buyer will ask about your AI, and almost nobody has an answer ready for it. If the AI stopped working tomorrow, what changes on the customer’s invoice?
If the answer is nothing, you shipped a feature. If the answer is that a line item disappears, you built a business. Buyers do not pay extra for AI you have shipped. They pay for AI money they can still collect after they own the company.
Most founders have never been asked this. Bain looked at roughly 200 B2B SaaS companies in August 2026 and found about 1 in 5 AI-native software companies still charges mostly per seat, with the AI folded in for free. Shipping AI and selling AI are 2 different businesses.
By the end of this post you will be able to sort your AI revenue into the 3 buckets a buyer uses, run their arithmetic yourself, and know what to have ready before anyone asks for it.
The Switch-Off Test
One question, asked about every AI feature you have built.
I have watched this land in management meetings more than once. The founder walks the AI roadmap for ten minutes. The buyer waits, then asks which customers pay separately for it, and the room goes quiet while somebody opens the billing export.
A buyer is not paying for technology. They are paying a multiple of revenue they believe will still arrive in year three. Every claim about AI is a claim about future revenue, so their job is to find out whether the AI is attached to money or to a slide.
The fastest way to find out is to imagine removing it. Switch the AI off and watch the invoice:
A line item disappears. The customer was buying AI as its own product. That is AI revenue.
A meter stops running. The customer was paying for AI by consumption. That is AI revenue too, priced differently.
Nothing happens. The invoice is identical. The customer bought your product for other reasons and the AI came along. Buyers credit this at zero until you prove otherwise.
That third case is where most founders live. It is not fatal, just unproven, and the rest of this post is about moving revenue out of it.
Three kinds of AI revenue, and what buyers pay for each
The big vendors have already picked sides, and their price pages are the clearest teaching material there is. Microsoft sells Microsoft 365 Copilot Business at $18 per user per month billed yearly, down from a regular $21, and a separate qualifying Microsoft 365 license is required underneath it (prices checked August 19, 2026). The AI is its own product a customer decides to buy.
Salesforce meters it instead. Agentforce pricing sells credits at $500 per 100,000, or $2 per conversation, on top of a free entry tier. Zoom does the same in miniature: its free AI tier is capped at 3 meeting summaries and 20 AI queries a month, and the paid tier carries 2,200 AI credits per user per month.
Among vendors introducing AI pricing, roughly four in five are choosing capacity models, where the customer buys a fixed entitlement, rather than pure pay-per-use. Source: Bain & Company, August 10, 2026, from an analysis of about 200 B2B SaaS companies.
Notice what none of the three did. Not one gives unlimited AI away inside the existing price and calls it a feature. Even the free tier has a meter on it. That is the market telling you what buyers already believe, and it sorts your own revenue into three buckets.
| Type | Switch-off result | What buyers credit | What they will demand |
|---|---|---|---|
| Priced AI | A line item disappears | Full multiple, same as core ARR | Attach rate, meaning the share of customers who bought it, and its renewal rate |
| Metered AI | A meter stops | Full multiple if usage is stable, discounted if spiky | Usage trend, and gross margin after the cost of running the models |
| Bundled AI | Nothing changes | Zero, until retention evidence exists | Proof that AI users renew better than non-users |
What a buyer’s model does with your AI ARR
Here is the arithmetic, one step at a time, with round numbers. A company does $5,000,000 of ARR and tells buyers that $1,500,000 of it is AI revenue.
Diligence opens the billing system. $400,000 sits on a separate AI product line customers bought on purpose. That survives.
Another $300,000 comes from AI credits customers consume and top up. That survives too, with a question attached about margin.
The remaining $800,000 is the AI tier of a plan customers were already on. Nobody bought AI. They bought the plan. That does not survive.
So the buyer credits $700,000 against a $1,500,000 claim. The company did not lose value. It lost the premium it asked for on the difference, and it lost credibility on every other number in the deck, which is the more expensive half.
It is the same sorting buyers run on how your pricing and packaging decisions show up in valuation. Packaging AI into a tier is a growth decision. It is also a decision to make that revenue invisible at exit.
Four AI claims that fail diligence, and one that does not
Three of these you have probably said out loud.
“Our AI drives 30% of engagement.” Engagement is not revenue. A buyer will ask what a customer would pay to keep it, and if that has never been tested with a price, it is a hypothesis. It moves nothing without retention data beside it.
“We added AI and raised prices 15%.” Now you are close. The buyer will ask what happened to churn and win rates after the increase. A price rise customers absorbed is evidence. One that quietly raised churn is a liability in a premium costume.
“AI is our biggest differentiator.” The buyer’s version is whether a competitor could ship the same thing in two quarters on the same public models. If your AI is a wrapper with no proprietary data behind it, the answer is usually yes, which is the subject of how buyers score AI risk.
“AI revenue is growing 200%.” Usually off a small base, out of a free beta, with no renewal completed. Buyers discount growth never tested by a renewal. One full renewal cycle at the AI price is worth more than a year of top-line growth.
Sometimes the premium is fully earned. Here is what that looks like, as an illustration rather than a specific deal. A vertical SaaS company sells a separate AI module at a set price per seat. Around 40% of customers have bought it, and that share is climbing. They renew several points better than customers without it, the module has completed two renewal cycles at full price, and its gross margin stays in normal software territory even after paying to run the models.
That company is not claiming an AI premium. It is demonstrating one, in numbers a buyer can check in an afternoon. Margin is the piece founders forget, which is why gross margin quietly adjusts your multiple more than sellers expect.
The evidence package to build now
All of this sits in your systems already. Build it before the first management meeting, not during diligence.
- AI revenue split three ways, straight from billing. If you cannot produce it, that is itself the finding.
- Share of customers on the AI line, over eight quarters. The trend matters more than today’s number.
- Retention delta, AI users against non-users. This is the only thing that converts bundled AI into credited AI.
- Renewal history at the AI price. Zero completed cycles is an answer, and better said by you than found by them.
None of it requires new engineering, just measuring AI as a business rather than a roadmap item, which is the shift our broader take on how AI is moving SaaS valuations keeps running into.
Frequently Asked Questions
Do buyers pay more for SaaS companies with AI features?
They pay more for AI revenue, not AI features. If customers never made a separate decision to pay for it, buyers price the company on its core metrics. The premium attaches to a billing line, not a capability.
Should I unbundle my AI features into a paid add-on before selling?
Only if you have time to complete at least one renewal cycle at the new price. Unbundling weeks before a process produces a number with no history, and risks churn in the exact period buyers examine. With 12 months or more, it is often the highest-return packaging change available.
How do buyers verify AI revenue claims in due diligence?
They go to the billing system, not the pitch deck. They look for a separate product line or a usage meter, then check how many customers bought it, how it renews, and what margin survives the model costs. Anything untraceable to an invoice line gets reclassified as core product revenue.
Does giving AI away for free hurt my valuation?
It does not hurt, but it does not help unless you can show the AI improves retention. Free AI that measurably lifts renewal rates is real value and buyers credit it. Free AI with no retention evidence is a cost in the buyer’s model, because they inherit the model bill without the revenue.
Next Steps
Not sure which bucket your AI revenue lands in? We will run the Switch-Off Test against your actual billing data and show you what a buyer would credit before you go to market.