Sales teams are writing AI policies faster than they are naming people to answer for the agents, and the numbers Seismic published this week point the same way.
The ownership gap is measured
Seismic’s 2026 The New Go-to-Market Standard for B2B Selling Report, summarized in a Seismic blog post on October 26, found that 16% of enterprise go-to-market organizations have no single owner to determine and confirm AI agent accuracy. It also found that 41% of those teams say their customer-facing functions need regular security and compliance sign-off to use agents for day-to-day revenue execution. Seismic sells governance features and the figures come from its own survey. They match what other vendors describe.
Take SAP’s quarterly release notes. They say Joule assistants orchestrate agents “respecting established authorizations, and maintaining audit trails,” and they introduce an AI Agent Hub, a central inventory of SAP and non-SAP agents. An inventory is a list of things that need an owner. The release notes sell the list and the logs. The person who reads the logs is left for the customer to name.
Advertisement
Why the policy comes second
My position is that a sales organization should assign a named owner to every agent that touches a buyer or a record before it writes a line of policy. That means one person for each workflow, with a backup who can act on a weekend, and the authority to pause the agent without asking anyone.
The reason is how these failures happen. Seismic’s report describes teams that build their own AI workflows without a vendor platform or central governance, and it says: “The workflow does not scale beyond the team that built it.” Nobody wrote a bad policy there. The workflow outlived the person who understood it. A policy cannot answer the phone at 7 p.m. if an agent starts quoting a retired discount to every open lead, and a named owner can.
Agents in the CRM make this sharper. An agent that drafts a meeting recap gives a rep something to review. An agent that updates an opportunity stage or edits a quote changes the data that forecasts and compensation are built on. We have argued that a connector is a data boundary RevOps must own, and the same logic applies to every write permission an agent receives. Somebody has to hold the permission list.
Seismic’s own list of failure modes reads like a set of owner questions. It names agents contacting leads with claims that never passed brand approval, agents pulling restricted account files into routine workflows, agents revising quotes beyond negotiated limits, and agents overwriting opportunity stages until pipeline reports become unreliable. Each of those lands first on a specific team, and that team is the natural place to look for the owner.
The strongest objection
The best argument against me is that owners without a framework make arbitrary decisions. Ten owners will draw ten different lines for what needs human approval, and the company ends up with ten policies anyway. Security and legal teams are right to want one standard, and Seismic’s finding that 41% require regular sign-off shows they are getting it in some companies.
That objection is correct about consistency and wrong about sequence. A shared standard is easy to write once ten owners have each hit a real edge case, because the edge cases supply the rules. A standard written first tends to cover the cases its authors imagined, which are the drafting tasks. We made a related argument in pre-launch testing is not a governance plan for sales AI: the checks that matter run after launch, and they need a person assigned to run them.
What I would do this quarter
Make a list of every agent your sales and RevOps teams have switched on, including the ones built by a single admin in a spare afternoon. Put one name next to each. Write down the three actions the agent can take that a buyer would notice: sending a message, changing a price, changing a record. Give the named owner a shutoff they can use without a ticket, and test it once with a drill. Then hand the list to security and legal as the starting draft of the standard.
If the 16% figure is close to right for your company, the first meeting is short. It takes one slide and ends with a name on every row.
Source: Seismic, AI agent governance: Best practices for go-to-market teams
