AI acquisitions and what buyers actually pay for

Retention is the first thing buyers check

Revenue that leaves next quarter is worth little. Buyers look at gross revenue churn, net revenue retention and how cohorts behave over time. Strong net revenue retention, where expansion from existing customers offsets losses, is one of the clearest signals of a product people rely on.

AI products face a specific doubt here. Many grew quickly on curiosity, and some of that early usage fades once the novelty is gone or once a general purpose assistant covers the same task. Buyers therefore look past the headline growth curve to the cohorts. They want to see customers who signed up six or twelve months ago still paying and still using the product.

• Monthly and annual churn, both by logo and by revenue.
• Cohort curves that flatten instead of sliding toward zero.
• Share of revenue on annual plans, which locks in a year at a time.
• Usage depth, such as active seats or tasks per account, not just log-ins.

Our article on SaaS churn rate goes deeper into how churn changes price.

Gross margin after inference costs

Classic SaaS is prized for high gross margins, because serving one more customer costs very little. Many AI products break that assumption. Every request to a large model costs money, and heavy users can cost more than they pay.

Buyers therefore recompute gross margin with inference, hosting, vector storage and third-party API fees counted as cost of revenue. They look at margin per plan and per customer segment, because a single power user on a flat price can quietly erase the profit of dozens of light users.

• Inference cost as a share of revenue, overall and by plan.
• Usage limits in the pricing that cap the cost of heavy users.
• Evidence of cost work already done, such as caching, routing simple requests to smaller models or batching.
• The trend, because model prices have tended to fall over time, while usage per customer often rises.

A product whose margin improves as it grows is easier to buy than one whose margin shrinks with every new heavy user. On Acquisitions, sellers can disclose inference cost share as a listing field, and buyers can filter by it.

Model dependency and platform risk

Almost every AI product depends on an AI model provider, and buyers price that dependency. The questions are practical.

• Can the product switch to another model provider without a rewrite?
• Does quality rely on a feature only one provider offers?
• Would a price change or a policy change at the provider hurt margin or break the product?
• Could the provider ship the same feature natively and absorb the market?

A product with an abstraction layer, evaluation sets that compare models, and a record of switching or running several providers carries less risk. A thin interface over a single model, with no proprietary workflow, data or distribution, carries the most. Buyers tend to apply a lower multiple to the second kind even when current revenue looks the same.

Dependency on distribution platforms matters too. An app that lives inside one marketplace, or a tool that relies on one platform's API, can lose access overnight. Buyers read the terms of those platforms as carefully as the financials.

Data rights and intellectual property

In many AI acquisitions the most valuable asset is data: fine-tuning sets, labeled examples, evaluation suites, customer usage history and the prompts and workflows built around them. Buyers pay for that only if it can legally transfer.

• Was training and fine-tuning data collected with the rights to use it commercially?
• Do customer contracts and the privacy policy allow the data to move to a new owner?
• Are there licenses on datasets, models or code that forbid transfer or commercial use?
• Who owns the code, prompts and designs created by contractors?

Gaps here do not always kill a deal, but they lower the price or push part of it into an escrow holdback or an earn-out until the issue is fixed. Clean documentation of data sources and contributor agreements is one of the cheapest ways a seller can raise what a buyer will pay.

Team, knowledge and transferability

Buyers ask who actually runs the business and what happens when the seller leaves. A product that depends on the founder's daily attention, personal relationships or undocumented know-how is worth less to someone who must run it without them.

• Written runbooks for deployment, monitoring and support.
• Clear ownership of accounts, domains and repositories in the business, not in a personal name.
• Hours per week needed to operate it, broken down by task.
• Whether key engineers or contractors are willing to stay through a transition.

For larger AI companies the team itself can be the main asset, and part of the price may be tied to key people staying. For small products, buyers mostly want a clean handover and a reasonable transition period with the seller available for questions.

Distribution and customer acquisition

A buyer inherits not only the product but the engine that brings new customers. That engine is often more defensible than the technology.

• Organic search traffic and its trend.
• Direct and returning visitors, which suggest brand pull.
• Integrations or marketplace listings that bring users in.
• Paid acquisition cost compared with the value of a customer over its lifetime.
• Concentration, meaning how much depends on a single channel.

An AI product with steady organic acquisition and low dependence on paid ads usually earns a better multiple than one that grows only while ad spend rises. Verified traffic data helps here, because buyers discount numbers they cannot check.

Other factors that move price

• Customer concentration. If one customer brings a large share of revenue, the risk of losing it weighs on the price.
• Category. B2B products with clear business value tend to be valued above consumer tools with fragile habits.
• Growth rate, adjusted for how it was achieved.
• Size. Larger, more established revenue tends to command higher multiples than very small revenue, because it is less fragile.
• Deal structure. Cash at closing, earn-outs, seller financing and holdbacks change the headline number without changing the real value much.
• Quality of records. Clean books, verified metrics and organized documents shorten due diligence and reduce the discount buyers apply for uncertainty.

What buyers usually do not pay for

Some things founders value highly carry little weight with acquirers.

• Hype, press attention and social media followers without matching revenue.
• Features nobody uses.
• Projections far above the current trend without a clear reason.
• Sign-ups and trial accounts that never convert into paying customers.
• A domain name on its own, unless it is short, memorable and in a strong category, in which case it can be a separate asset. See AI domains for sale.

How sellers can raise the price before listing

• Connect revenue and analytics through read-only access so buyers see verified numbers from day one. See Revenue verification.
• Recompute your margin with inference costs and publish the figure honestly.
• Reduce model dependency where it is cheap to do so, and document how you would switch.
• Collect proof of data rights, contractor agreements and licenses in one folder.
• Write runbooks and move accounts into the business name.
• Move customers to annual plans where it fits the product.

These steps take weeks, not months, and they speak directly to the questions every serious buyer asks. Buyers can use the same list as a lens when reviewing AI agents for sale or any other category.