Why does every AI agent in a revenue team need an owner?
The count of agents in a revenue team is set to climb. In July 2026, Gartner predicted that by 2028 AI agents will outnumber sellers 10 to 1, and that fewer than 40% of sellers will say agents improved their productivity. Treat both as analyst forecasts. The same release warned that sales leaders risk agent sprawl.
Agents also make mistakes. In a May 2025 benchmark from Salesforce AI Research, leading AI agents completed about 58% of single-turn CRM tasks and about 35% when the task ran across several turns. That is vendor research, run on synthetic data with the models of that time, and newer models may score higher. The planning point holds. Some share of agent work in your CRM and in your customers' inboxes will be wrong.
When it is, someone answers for it. In a revenue team that person is the CRO, whether or not anyone wrote it down.
What should you ask about every AI agent in the revenue team?
Ask these of every agent in the revenue team, whether you built it, bought it or a seller switched it on inside a tool you already pay for.
-
Which agents act on customers, records or the forecast today, and who set each one up?
Agents do not all come through procurement. A vendor adds one to a product you already own, or a seller signs up for one without asking. In Microsoft and LinkedIn's 2024 Work Trend Index (vendor research, 31,000 knowledge workers surveyed in early 2024), 78% of AI users said they bring their own AI tools to work, and that figure covered AI tools in general. NIST's voluntary AI Risk Management Framework treats an inventory of AI systems as part of governance. Listen for a list that includes what sellers and managers turned on themselves, with a setup name beside each entry.
-
Whose name is on each agent's work?
A team name or a vendor name leaves nobody to answer. NIST's framework asks that roles and responsibilities for AI risk be documented and "clear to individuals and teams throughout the organization," and a February 2026 NIST draft concept paper names linking user identities to AI agents, to "maintain accountability for the actions of automated systems," as an area to explore. The name needs authority behind it. In 2019, researcher Madeleine Clare Elish drew on accidents in complex automated systems to name the "moral crumple zone," where blame lands on a person who had limited control over the system. Listen for one person per agent who knows what it does and can change it.
-
Which actions need a person to approve them before they happen?
The OWASP Top 10 for LLM Applications (2025) advises human-in-the-loop control "to require a human to approve high-impact actions before they are taken." Approval works only when the approver reads. A 2022 peer-reviewed review of 41 human oversight policies for government algorithms found that "evidence suggests that people are unable to perform the desired oversight functions," a warning that carries to a manager clearing a queue of agent drafts. Listen for specific actions named, such as a price sent to a customer or a change to a committed deal, and for an approver who has said no recently.
-
How would we know if an agent changed a number in the forecast?
An agent that updates close dates or amounts from email can move the forecast with no person touching the deal. The NIST concept paper lists linking each agent action "to the identity of the non-human entity" as a focus area, and OWASP notes that logging limits damage without preventing it. Listen for a record that shows agent edits apart from human edits, and a person who reads those edits before the forecast goes to the board.
-
What happens when an agent gets it wrong with a customer?
In Moffatt v. Air Canada (February 2024), the airline argued its chatbot was "a separate legal entity that is responsible for its own actions." A British Columbia tribunal disagreed. "It makes no difference whether the information comes from a static page or a chatbot," it wrote, ordering C$812.02 in damages, interest and fees. That is one small-claims decision, and other courts may see the question differently. Listen for a named person who repairs the relationship and a way to find every customer the same error reached.
-
Who can switch an agent off?
NIST's framework expects responsibilities to be "assigned and understood" for disengaging or deactivating AI systems whose outcomes are "inconsistent with intended use." Agents keep running on weekends and at month end, when the person who set one up may be away. Listen for a name with the authority and the access to pause each agent, and a plan for the work it was doing.
"For every agent that touches a customer or the forecast, whose name is on it?"
This question sits under the others, because each of them ends at a person's name. Where the answer is blank, the number still has an owner, and it is the CRO.
What are the signs AI agents in your revenue team are well governed?
Good signs
- Every agent that touches customers or the forecast has one named owner who can explain what it does.
- The agent list includes tools that sellers and managers turned on themselves.
- Approvers sometimes reject an agent's work, and the team learns from it.
- The CRM shows which changes came from an agent.
- Someone has paused an agent before and knows how to do it again.
Warning signs
- The owner named for an agent is a vendor, a team or "IT."
- The forecast moved and nobody can say whether a person or an agent moved it.
- Approval queues get cleared in one sitting.
- A customer complaint is the first anyone hears of an agent.
- Nobody knows who can switch an agent off.
Case study Fictitious numbers
A PE-backed commercial HVAC services company has 22 sellers across four branches. Midway through a quarter, the renewal forecast drops, and no seller edited the deals that moved. The CRO asks for a list of every AI agent that writes to customers, changes CRM records or touches the forecast. She expects the 3 that revenue operations set up.
The list comes back with 11. Revenue operations set up 3. Sellers and branch managers turned on 6 inside tools the company already paid for. A vendor configured 2 during an implementation. That is 3 plus 6 plus 2, for 11 in all, and only the first 3 had a named owner.
- Named owner
- No owner
The next forecast change comes with a name beside it.
One of the seller-built agents read customer email and pushed renewal dates into the next quarter whenever a customer asked for more time. The seller who turned it on had moved to another branch. Nobody had reviewed it since.
The CRO puts a name on every agent. Two seller-built agents get switched off because no manager will own them. Branch managers take the other 4, and the revenue operations lead adds the vendor's 2 to the 3 his team already runs. That leaves 9 running, each with a name. The renewal-date agent now suggests a change, and the account owner accepts or rejects it. The next forecast change comes with a name beside it.
FAQ
Who is responsible when an AI agent gives a customer wrong information?
The business that deployed it, at least in Moffatt v. Air Canada (2024). A British Columbia tribunal rejected the airline's argument that its chatbot was a separate legal entity, and found the airline "did not take reasonable care to ensure its chatbot was accurate." That is one small-claims ruling, and courts elsewhere may decide differently. Inside your company, give each agent a named owner who answers for it.
Do customers have to be told they are talking to an AI agent?
In the EU, the AI Act sets a rule. Article 50, which has applied since 2 August 2026, requires providers of AI systems meant to interact directly with people to design them so those people are told they are dealing with AI, unless that is obvious from the context. A company that builds its own agent may count as the provider. Rules differ by country and channel, so check with counsel for each market you sell into.
What does human in the loop mean for AI agents?
A person approves the agent's action before it takes effect. NIST's February 2026 draft concept paper on agent identity describes a range of agent actions, "from controlled human-in-the-loop approval to autonomous action." A revenue team can run different agents at different points on that range. The work is deciding which actions sit at the approval end, and making sure approvers have the time to read what they approve.
Should revenue operations, IT or the CRO own AI agents in sales?
Each owns a different piece. Revenue operations or IT can own setup, access and upkeep. A business owner answers for what each agent does to customers and to the number. NIST's AI Risk Management Framework places responsibility for decisions about AI risk with executive leadership. In a revenue team that points to the CRO, with a named owner under the CRO for each agent.
Sources
- Gartner Predicts AI Agents Will Outnumber Sellers 10 to 1 by 2028 (analyst prediction, 28 Jul 2026) checked 10 Oct 2026
- CRMArena-Pro, Huang et al., Salesforce AI Research (vendor research, arXiv preprint, May 2025) checked 10 Oct 2026
- AI at Work Is Here. Now Comes the Hard Part, Microsoft and LinkedIn 2024 Work Trend Index (vendor research, 8 May 2024) checked 10 Oct 2026
- NIST AI 100-1, Artificial Intelligence Risk Management Framework (AI RMF 1.0), January 2023 checked 10 Oct 2026
- Accelerating the Adoption of Software and AI Agent Identity and Authorization, NIST NCCoE draft concept paper, February 2026 checked 10 Oct 2026
- Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction, Elish, Engaging Science, Technology, and Society, 2019 checked 10 Oct 2026
- LLM06:2025 Excessive Agency, OWASP Top 10 for LLM Applications checked 10 Oct 2026
- The Flaws of Policies Requiring Human Oversight of Government Algorithms, Green, Computer Law & Security Review, 2022 (peer-reviewed) checked 10 Oct 2026
- Moffatt v. Air Canada, 2024 BCCRT 149, British Columbia Civil Resolution Tribunal checked 10 Oct 2026
- AI Act, Article 50, European Commission AI Act Service Desk checked 10 Oct 2026
- AI Act, Article 113, European Commission AI Act Service Desk checked 10 Oct 2026