Quick answerWhen your AI agent is embedded inside one partner's product under a co-brand, negotiate the commercial upside terms as their own section, separate from IP and liability: who gets attribution and how visibly, whether revenue share is calculated on the partner's uplift from having the feature or on total usage volume, who controls joint marketing claims and case studies, and what happens to the revenue-share terms if the partner's own product pivots away from the use case the embedding was built for. These are negotiated once at deal signing and rarely revisited, so get the calculation basis and pivot clause right before signing, not after the first disputed payout.
A commercial question, not an IP or operations question
Two related questions about embedded partner deployments already have their own answers. The IP and liability terms cover who is liable for a customer-facing mistake and who owns the resulting conversation data, and the ongoing operational support obligations cover the SLA and communication commitments owed after launch. Neither of those addresses the commercial upside: how the partnership actually makes money for both sides beyond the base licensing fee, and who benefits when the embedded agent becomes a genuine selling point for the partner's own product.
This is also distinct from agent-to-agent integration safeguards, which governs two separate companies' agents interacting with each other across an API boundary. Here there is only one agent, embedded inside a single partner's product surface, presented to that partner's end customers as part of their offering.
The four terms worth negotiating explicitly
Attribution and branding visibility. Decide upfront whether your company gets visible credit inside the partner's product, in their marketing, or in their sales materials, and how prominent that credit is. A fully white-labeled embedding with zero visible attribution is a legitimate choice, but it should be a deliberate one reflected in a lower revenue share or higher base fee, not a default that gets negotiated away silently.
Revenue-share calculation basis. Choose between sharing a percentage of the partner's total revenue from the product tier the agent lives in, or sharing revenue specifically attributable to the uplift the agent creates, such as increased conversion or retention measurably tied to using the feature. Total-revenue splits are simpler to administer and harder to dispute. Uplift-based splits reward the vendor more fairly for genuine impact but require an agreed measurement methodology written into the contract itself, not left to a good-faith conversation after the fact.
Joint marketing and case study rights. Specify who can publish what, and whether either party needs the other's sign-off before naming them in a case study, press release, or sales deck. This matters more than it sounds: a partner's marketing team publishing a customer story that overstates the agent's capabilities creates real liability exposure for the vendor, who did not review the claims before they went out.
A pivot or discontinuation clause. Partner products change direction. Define what happens to the revenue-share and co-marketing terms if the partner deprioritizes or sunsets the specific feature the agent was embedded to support, including a defined wind-down period and whether any minimum revenue-share floor survives a reduced usage volume that was not the vendor's fault.
FAQ
Should revenue share replace or supplement a base licensing fee? Supplement it in almost every case. A pure revenue-share model with no base fee leaves the vendor fully exposed to the partner's own product execution, which the vendor does not control. A base fee plus a smaller revenue-share upside aligns incentives without transferring all the commercial risk.
Who should own the end-customer relationship for support purposes? That is answered by the ongoing operational support terms linked above, not by the revenue-share agreement. Keep the two negotiations conceptually separate even when they happen in the same contract, since conflating "who gets paid" with "who answers the support ticket" tends to produce worse terms on both.
Does this apply to a free or bundled embedding with no separate fee? Yes, with the revenue-share question reframed as a value-attribution question: even without a direct payment, you want contract language establishing what credit and data-sharing rights your company gets in exchange for the strategic value the embedding provides the partner.