Delegating to AI: what managers can drop from the calendar
A manager who hands decisions to AI and keeps building reports by hand has delegation backwards. The criterion for choosing what to pass to AI, what to keep with a person and how to write the instruction so the work comes back ready to review.
Starting this week, a sales manager can delegate to AI the work of compiling and writing first drafts: meeting summaries, the Monday report, one-on-one prep, the briefing before a customer meeting, the list of stalled deals and drafts of emails and proposals. They should keep the decisions about people, price, revenue projection and the customer relationship, because there a mistake is expensive and responsibility does not transfer to a system. If you lead sales, marketing or post-sale, your decision is to pick three compiling tasks to take off your managers’ calendars and write down, for each one, who reviews the output before it reaches anyone. Delegating to AI follows the same rule as delegating to a person: a clear task, enough context, an agreed format and a check at the end.
What I see in companies is delegation turned upside down. The manager asks AI for a recommendation on which salesperson to promote, pastes the answer into a presentation and spends the afternoon building the board meeting spreadsheet by hand, copying numbers from the CRM cell by cell. Or the mirror image of the same mistake: they forbid the team from using AI for anything and find out months later that half the salespeople use personal tools with customer data. In both cases nobody decided, with a criterion, which work leaves the manager’s hands and who checks what comes back.
What does it mean to delegate a task to AI?
Delegating to AI means handing an assistant such as Claude, ChatGPT, Gemini or the assistant built into the CRM a task with a beginning, middle and end, and getting back work the manager reviews before using it. The difference from “using AI” loosely lies in the agreement: the task always comes back in the same format, from the same sources, and someone answers for the result. A manager who asks “help me with the report” every Monday is chatting with a tool. A manager who has a saved instruction, receives the report in the format the executive team uses and checks three numbers before sending it has delegated.
The uses that work in sales management, such as the pipeline meeting, the projection review and coaching, are in AI for sales management. This text covers the criterion that comes first: what to hand over and what to hold on to.
What can a sales manager delegate to AI today?
The rule I use has two questions. First: if the AI gets it wrong, will I notice the error before it causes harm? Second: if the error slips through, can it be undone at no cost to the customer or to someone on the team? When both answers are “yes”, the task can leave the manager’s calendar today. When either one is “no”, AI prepares the material and a person makes the decision.
| Task | Delegate to AI? | Who reviews | Why |
|---|---|---|---|
| Meeting summary for the CRM | Yes | Salesperson who owns the deal | Whoever was in the meeting spots the error when reading |
| Weekly report for the executive team | Yes, with checks | Manager checks the key numbers | The data comes from the CRM and can be verified |
| One-on-one prep | Yes | Manager, before the conversation | The manager uses it as a script, without passing it on |
| Pre-meeting customer briefing | Yes | Salesperson attending the meeting | Public sources and account history, all verifiable |
| Draft of a proposal or email | Yes, as a draft | Salesperson and, if there is a price, the manager | The customer only receives it after review |
| List of stalled deals | Yes | Manager in the pipeline meeting | The list points where to look and each case gets checked |
| Discount approval | No | Manager, within approval levels | A margin decision that affects the customer |
| Revenue projection for the executive team | Only the prep | Manager and director | AI suggests, leadership owns the number |
| Performance review and promotion | No | Manager, director and HR | A decision about a person, with legal and career effects |
| Message sent to the customer without review | No | Nobody should skip the review | The error reaches the customer under the company’s name |
The first four items in the table have their own guides: meeting summary with AI, Monday report with AI, pre-meeting briefing with AI and stalled deals in the pipeline. Starting with them has a practical advantage: the manager feels the time savings in the first week and learns to review with no risk to the customer. When the task involves spreadsheets exported from the CRM and documents in the manager’s folder, the step by step is in Claude Code for sales managers.
What should a manager not delegate to AI?
AI writes with the same confidence when it is right and when it makes things up, and what the market calls hallucination shows up in exactly the questions where the manager wanted a definitive answer. I describe how it happens and how to reduce the risk in AI hallucination. That is why four types of decision stay with a person, even when AI helps prepare the material.
- Decisions about people. Hiring, reviews, promotions, dismissals and portfolio distribution. AI can organize CRM data and one-on-one notes, and the manager decides and signs.
- Price and terms. Discounts, payment terms, special conditions and price adjustments touch margin and contract. AI can simulate scenarios and the manager approves within the approval levels.
- The number that goes to the executive team. The revenue projection needs an owner who can explain why they believe in it. AI points out inconsistencies and at-risk deals, and the manager owns the number.
- Whatever reaches the customer. Proposals, emails, complaint replies and WhatsApp messages go through a person before sending. An assistant that talks to customers without review calls for another conversation, about AI agents and governance.
This limit also protects the manager. When a decision about a person or a price goes wrong, the company asks who decided, and “the AI did” does not work as an answer for the customer, the employee or the executive team.
Why does CRM quality limit what you can delegate?
AI that reads the CRM delivers a summary as good as the records it finds. If the salesperson does not log the next activity, the stalled-deals list comes out full of false alarms. If deal values are guessed, the weekly report adds up guesses with nice formatting. In the 2024 Salesforce State of Sales report, which surveyed 5,500 sales professionals, only 35% said they fully trust the accuracy of their organization’s data.
With imperfect data, delegation is still possible and starts to require checks. The manager who delegates the weekly report should pick three numbers to check every week, and the first round tends to reveal where the records fail. The conversation about what to fix in the records is in CRM data quality, and the limits of AI on a disorganized base are in AI on an unstructured CRM.
How do you write a delegation instruction for AI?
A delegation instruction has the same parts as the request a manager would make to a new analyst: who is asking and for what, which material to use, what to deliver, in what format, what not to do and how to check it. A saved instruction becomes a team asset: the next manager inherits the template and the output comes out the same every week.
Two rules complete the instruction. The first is about data: the manager pastes only what is needed, with no customer personal data beyond what the task requires, and only into tools the company approved, as I cover in AI and data protection law. The second is about memory: an instruction that worked gets saved in a team library, with an owner and a review date, so nobody reinvents the request every Monday.
How much time does a manager get back by delegating to AI?
It depends on how much compiling they do today, and an honest answer requires measuring. I suggest the manager log, for two weeks, how much time goes into the tasks in the table that can leave the calendar, and repeat the measurement a month after delegating. Review time goes into the count; it is higher in the first weeks and drops as the instruction improves. The full calculation, with tool and training costs, is in AI ROI.
Time recovered only turns into results when it goes back into the blocks that drive revenue: one-on-ones, recording reviews, large deals and strategic customers. If the manager frees four hours a week and they vanish into internal meetings, the delegation worked and the management did not change. That is why I tie this work to the design of the sales leader’s calendar, with the blocks protected before the time appears. For someone just promoted from sales rep, this shift in tasks belongs in the 90-day plan for a new sales manager.
What do the executive team, marketing, post-sale and HR need to define?
The manager decides what to delegate within their area, and the company sets the outside rules. Three decisions belong to leadership: which tools are approved and with which data, who answers for an error that reaches the customer and how delegation shows up in managers’ reviews. These rules fit in a short AI use policy, published before the team gets access.
For marketing, the same logic applies to campaign reports, research summaries and first drafts of content, reviewed by whoever signs the piece. In post-sale, AI prepares the account history before the renewal meeting and the summary of the quarter’s tickets, and the account manager decides what to bring to the customer. For HR and L&D, the signal that the team needs training shows up when each manager uses AI their own way, with no saved instruction and no review criterion.
A 30-day plan to get started
- Week 1, choice. Each manager picks three compiling tasks from the table and measures the current time for each.
- Week 2, instruction. The manager writes the delegation instruction for each task, defines who reviews it and tests it with last week’s material.
- Week 3, use. The three tasks start coming out of AI, with the review recorded and three points checked on each delivery.
- Week 4, adjustment. The manager fixes the instruction where the review found errors and saves the final version in the team library.
- Day 30, read. Compare time spent with the baseline and decide where the recovered hours go.
If your leadership wants managers to delegate to AI with a written criterion and a defined review, that is the work of the in-company applied AI training. I run the program on the operation’s real tasks, with instructions written and tested in the room, review criteria for each one and the data rules the company needs, and every manager leaves with an initial delegation library.
Frequently asked questions
What is the first task a sales manager should delegate to AI?
The meeting summary for the CRM or the weekly report. Both eat time every week, use material the manager knows and let you check the output before it reaches anyone.
Can AI build the team’s sales projection?
It can prepare the analysis, point out deals with risk signals and compare them with history. The number that goes to the executive team needs a manager who explains it and answers for it.
Which AI tool should you use to delegate tasks?
The one the company approved for the type of data involved. Claude, ChatGPT, Gemini and the assistants built into CRMs such as HubSpot handle the tasks on this list, and the choice depends more on data rules and where the information already lives than on the model.
How do you know whether AI got the report wrong?
Pick three numbers you check every week against the CRM and ask the AI to mark what it calculated or inferred. When the three match for a few weeks in a row, the review can get lighter, without going away.
Does delegating to AI reduce the need for managers?
The compiling work shrinks, and the work of developing people and deciding remains. In my experience, a manager who delegates well spends more time with the team and with customers.
Want to bring this to your team?
The topics in this article also become in-company training: designed from a diagnosis of your operation, with real cases from your team and a deliverable that stays with the company.
More writing on the same subject.

Value selling: how to sell outcomes instead of price
A B2B customer buys the distance between where they are today and where they want to be. When the salesperson talks about the product, that distance drops out of the conversation and the buyer starts comparing prices.
16 min readRead more
In-company sales training: how to change behavior
One-day training earns good ratings and leaves the pipeline unchanged. To change behavior, a company needs a diagnosis first, practice on real deals, a deliverable that stays, reinforcement from sales, marketing and customer success leaders and measurement in the CRM.
17 min readRead more
AI training for teams: how to build capability by role
Generic AI courses bring enthusiasm and little use. Teams learn when they train by role, on the company's own processes, with usage rules, deliverables that stay and metrics agreed before the first session.
16 min readRead more