Why does it matter which manager work AI takes on?
Front-line managers shape how a team works each day. Gallup reported in 2015 that managers account for at least 70% of the variance in employee engagement scores. Much of their week goes elsewhere. In a 2016 survey of 1,770 managers in 14 countries, Accenture researchers writing in Harvard Business Review found that administrative coordination and control took more than half the time of managers at every level. The survey was self-reported and predates today's models.
AI helps most when the work sits inside what it does well. In a field experiment with 758 Boston Consulting Group consultants (a 2023 working paper), those using GPT-4 on tasks within its reach worked 25.1% faster at more than 40% higher quality. On a task outside its reach, they were 19 percentage points less likely than consultants without AI to get it right. The authors call this a "jagged technological frontier."
What questions separate the work AI can do from the work managers must keep?
These questions sort the hours a machine can take from the hours that belong to your managers. Ask them before you choose a tool, and ask them again once your managers have lived with it.
-
Which part of the manager's week is reading and summarizing, and which part is judgment?
A pipeline review means little until someone has read the deal history and service records behind each opportunity. That reading is work a model does fast. Deciding which seller needs help, and what to say to them, is judgment. In the 2016 HBR survey, 86% of managers wanted AI to help with monitoring and reporting, and the authors found the same managers undervalued "the deep social skills critical to networking, coaching, and collaborating." Listen for managers who can point to the hours on their own calendar that go to each.
-
What data would the AI read, and how current is it?
A summary is only as good as the record under it. In a services business, much of what matters about an account lives in service tickets and technician notes, outside the CRM. In February 2025, Gartner predicted that through 2026 organizations will abandon 60% of AI projects unsupported by AI-ready data. Treat that as an analyst forecast. Listen for every source named, with when each was last updated and what the tool cannot see. A tool reading a stale CRM gives a confident answer that is out of date.
-
How will we know when its answers are wrong?
Wrong answers read as smoothly as right ones. A 2025 paper by researchers at OpenAI and Georgia Tech argues that language models guess when uncertain, "producing plausible yet incorrect statements instead of admitting uncertainty," because training and testing reward the guess. People also check less as they trust more. A Microsoft Research survey of 319 knowledge workers, published in 2025, found that higher confidence in generative AI was associated with less critical thinking. That finding is self-reported and shows association. Listen for a habit of checking answers against the source record and a named person who reviews the misses.
-
Which decisions stay with a person, and who decided that?
Some decisions are about people. Choosing who gets coached on what, or who goes on a performance plan, draws on context no record holds. Others are trade-offs with a customer, such as a price concession on a renewal or a change in scope. The EU AI Act lists AI systems used to "monitor and evaluate the performance and behaviour" of workers among its high-risk uses, so for some companies this line carries legal weight. Listen for a line someone wrote down and owns, with a name and a date beside it.
-
What will managers do with the time it gives back?
Saved time drifts unless someone decides where it goes. Economists at the Federal Reserve Bank of St. Louis reported in November 2025 that workers' self-reported time savings from generative AI equal 1.6% of all work hours in their U.S. survey, counting nonusers. They noted that saved time can go to "lower-value tasks" or to leisure on the job. Listen for a specific answer about where the hours move and who will notice if they move somewhere else.
-
What will sellers know about what it reads?
A tool that reads a seller's deals and email is reading that seller's work. If it counts as a high-risk system under the EU AI Act, the employer must inform workers' representatives and the affected workers before using it at work, a duty that applies from 2 December 2027 under the Act's consolidated text as at July 2026. Outside the EU, tell them first anyway. Listen for a plain explanation a seller could repeat, covering what the tool reads and what it will never be used for.
"Which hours of my managers' week could a machine take, and which hours must stay theirs?"
Every question above feeds this one. The reading hours can move once you trust the data and someone checks the answers. The hours with sellers and customers stay with the manager, and the time given back pays off only when it lands there.
What are the signs AI is helping sales managers, or hurting?
Good signs
- Managers can name the hours in their week that go to reading and the hours that go to people.
- Every summary points back to the record it came from and the date of that record.
- Someone checks answers against the source and can show the last miss and what changed.
- Decisions about territory and performance have an owner, and the owner is a person.
- Sellers can explain what the tool reads about them.
Warning signs
- Summaries reach sellers or customers before a manager has read them.
- The hours saved went into more internal reporting.
- A performance conversation started from a machine's ranking.
- Nobody can say when the data the tool reads was last updated.
- Managers quote the summary and no longer open the account.
Case study Fictitious numbers
A PE-backed commercial HVAC services company has four branch sales managers, each leading seven sellers. The company gives the managers an AI assistant that reads the CRM and the service history, then drafts account summaries and prep for each pipeline review. Each manager reports that it saves 6 hours a week of reading and summarizing.
A quarter later, the VP of sales looks at the managers' calendars to see where those 6 hours went. One hour went to more time with sellers on open bids and renewals. Three went to new internal reports requested from above. The last two went to checking the summaries, after several of them quoted renewal dates from agreements that had since been re-signed. One plus three plus two accounts for the full 6.
- Time with sellers on bids and renewals
- New internal reports
- Checking the summaries
- Two hours a week
- Summaries quoted renewal dates from re-signed agreements
- Part of the price of trusting the tool
- Three hours a week
- New internal reports requested from above
- The VP cuts the ones nobody reads
The tool did what it promised. The time went where no one had decided it should.
The checking hours are part of the price of trusting the tool, and they stay. The reporting hours were a choice nobody made on purpose. The VP takes the report requests to the operating partner, cuts the ones nobody reads, and asks each manager to name where the returned hours will go. The tool did what it promised. The time went where no one had decided it should.
FAQ
Will AI replace sales managers?
Gartner's May 2026 research on sellers points to a split. It said AI is well suited to "account research, personalized messaging, signal monitoring and next best actions," while sellers stay differentiated in "empathy, judgment, contextual understanding and value framing." In its survey of 645 B2B buyers, buyers were 39 percentage points more likely to say a rep understood their needs than to say generative AI did. These are analyst survey findings. Building those strengths in sellers stays a manager's job.
Who gains most when a team starts using AI?
A study of 5,172 customer-support agents, published in the Quarterly Journal of Economics in 2025, found AI assistance raised issues resolved per hour by 15% on average. Less experienced and lower-skilled agents improved in both speed and quality. The most experienced and highest-skilled saw small gains in speed and small declines in quality. These were support agents working with a chat assistant, so test the pattern on your own team before you plan around it.
Is it legal to use AI to evaluate sales rep performance?
It depends on where you operate. The EU AI Act classes AI systems that monitor and evaluate the performance and behaviour of workers as high-risk, and employers must inform the affected workers and their representatives before using one at work. Under the consolidated text as at July 2026, those high-risk duties apply from 2 December 2027. Check the dates with counsel. In any country, a person should own the final decision on someone's job.
Why do AI tools make confident mistakes?
Researchers call these mistakes hallucinations. The 2025 OpenAI and Georgia Tech paper argues that language models are "optimized to be good test-takers," and that guessing when uncertain improves test scores. The result is a fluent answer that can be wrong. Your defense is a person who checks answers against the source record on a regular rhythm. A summary that cannot be traced to its source cannot be checked.
Sources
- Employees Want a Lot More From Their Managers, Gallup, 8 Apr 2015 checked 9 Oct 2026
- How Artificial Intelligence Will Redefine Management, Kolbjørnsrud, Amico and Thomas (Accenture researchers), Harvard Business Review, 2 Nov 2016 checked 9 Oct 2026
- Navigating the Jagged Technological Frontier, Dell'Acqua et al., Harvard Business School Working Paper 24-013, September 2023 checked 9 Oct 2026
- Generative AI at Work, Brynjolfsson, Li and Raymond, Quarterly Journal of Economics, 2025 (peer-reviewed) checked 9 Oct 2026
- Why Language Models Hallucinate, Kalai, Nachum, Vempala and Zhang, arXiv preprint, September 2025 checked 9 Oct 2026
- The Impact of Generative AI on Critical Thinking, Lee et al., Microsoft Research, CHI 2025 (self-reported survey) checked 9 Oct 2026
- Lack of AI-Ready Data Puts AI Projects at Risk (Gartner, analyst prediction, 26 Feb 2025) checked 9 Oct 2026
- The State of Generative AI Adoption in 2025, Bick, Blandin and Deming, Federal Reserve Bank of St. Louis, 13 Nov 2025 checked 9 Oct 2026
- Gartner Survey Finds Sales Organizations That Provide AI-Enabled Next Best Actions Are 2.6x More Likely to Achieve Commercial Growth (Gartner, analyst survey, 20 May 2026) checked 9 Oct 2026
- EU AI Act, Annex III, High-Risk AI Systems (EU AI Act Explorer) checked 9 Oct 2026
- EU AI Act, Article 26, Obligations of Deployers of High-Risk AI Systems (EU AI Act Explorer) checked 9 Oct 2026
- AI Act, Article 113, Entry into force and application, European Commission AI Act Service Desk (consolidated text as at 27 Jul 2026) checked 10 Oct 2026