After every client call I spend an hour writing notes and follow ups. Can AI do it?
Short answer: Yes, AI can draft most of it: a summary, a list of decisions, a list of tasks with owners, and a first version of the follow up email. It should not send anything or write to your records until a person has read it. Tools such as Microsoft Copilot in Teams (Microsoft's AI helper inside its meeting software) and Zoom's AI assistant already produce summaries and suggested action items, and a general AI model like Claude (Anthropic's AI assistant) can turn rough notes into the same output. The part that protects a consulting firm is the review step and the confidentiality rules around it.
You want the hour after each client call back, without a wrong note or a confidential detail ending up where it should not.
I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.
If you want to know where AI would pay off across your whole firm, not just after meetings, the AI Strategy Sprint finds it in 3 days.
What AI can and cannot do after a client call
The work after a call has four parts: write what was said, pull out what was decided, list who must do what by when, and tell the client. AI handles the first three well when it has a transcript or decent notes. It handles the fourth as a draft.
What it does badly is knowing what mattered. It does not know that the client's one throwaway comment about their budget is the most important thing said. It also sometimes attributes a task to the wrong person, or invents a due date that nobody agreed. So the rule is simple: AI drafts, a person decides.
Vendors confirm the same split in how they describe their own products. Microsoft says Copilot in Teams summarizes key discussion points, including who spoke and what they said, and suggests action items. The word is suggests. Zoom describes its AI assistant, ZoomMate, as producing summaries and lets you create custom summary formats. These are drafts for a human to check.
The five outputs worth automating
- A short summary. Five to eight lines: purpose of the call, what was agreed, what is open.
- A decisions list. Things the client approved or rejected, in plain words.
- A task list with owners and dates. Each task names one person and one due date, or says "date not agreed".
- A follow up email draft. Sent from the consultant, in your tone, restating decisions and next steps.
- A record update. A short note for your CRM (customer relationship management tool, the software that keeps client history) or your project tool.
Do these in that order. If the summary is wrong, everything after it is wrong, so the review step sits between the summary and the rest.
Confidentiality first
For a consulting firm this is the most important part. Clients tell you things under an understanding that they stay private. Before you feed any call into an AI tool, settle these points.
- Read your client contracts. Many include confidentiality clauses, and some limit what you may share with outside services. Ask a lawyer if the wording is unclear. This is not legal advice.
- Tell the client you are using it. If a tool records or transcribes the call, tell them at the start. Rules about recording consent vary by place, so check yours.
- Know whether your data trains the model. Anthropic states that by default it does not use inputs or outputs from its commercial products, such as Claude for Work and the API, to train its models. Its consumer plans are covered by a different policy, so check which plan you are on. Microsoft states that prompts, responses and data accessed through Microsoft Graph in Microsoft 365 Copilot are not used to train its foundation models. Whatever tool you pick, read the same page for it.
- Check who can see the output. Microsoft notes that Copilot only surfaces organizational data a user already has permission to view, which means your permissions need to be tidy first. A summary saved in a folder everyone can open is a leak.
- Remove what you do not need. NIST, the US standards agency, publishes a free AI Risk Management Framework if you want a structure for a written AI policy. Before pasting notes into a general AI tool, strip names, account numbers and anything the task does not require.
- Are the decisions what was actually decided?
- Does every task have one owner, and is each due date one that was agreed?
- Does anything in the summary mention another client or confidential detail that should not be there?
- Does the email sound like us, and does it promise anything we did not promise?
- Is the CRM note factual, with no guesses about the client's mood or intentions?
- The call covers legal advice, health data or anything regulated, unless your contract and tool settings clearly allow it.
- Your client has told you in writing not to use AI tools on their material.
- The call is a sensitive one, such as a dispute or a layoff discussion. Take notes by hand.
- Nobody on your team has time to review. An unchecked draft is worse than no draft.
One more limit worth knowing: Microsoft says Copilot in Teams will not work in meetings hosted outside the participant's own organization. If your clients host the call on their account, your built-in Teams AI may simply not be available, and you will need another way to get the transcript.
Three ways to do it
Option 1: use your meeting tool's built-in summary
If you use Teams or Zoom, check which plan includes the AI summary and suggested action items (Zoom's free plan lists meeting summaries for 3 hosted meetings per month, so check the current limits). This is the quickest start. The limits are that the output goes in the vendor's format, tasks do not automatically land in your task tool, and it only works on calls hosted on your account.
Option 2: paste notes or a transcript into an AI model
Take your own rough notes, or a transcript exported from the call, and give them to a model with a fixed prompt (see the template below). You control the format. The cost is a few minutes of copying, and you must handle confidentiality yourself. If you prefer not to have a bot in the room at all, read my guide to getting meeting notes without a note taker joining.
Option 3: connect it to your tools with a review queue
This is the version that removes the copying. The transcript goes to an AI step, which writes the summary, tasks and email draft into a review queue. A person reads the queue, edits, and approves. Only approved items go on to the task tool, the CRM and the email draft folder. A no-code automation tool can connect the pieces, and Microsoft's Power Automate, its workflow tool, can be one of them if you live in Microsoft 365. The pattern matters more than the tool. For how these pieces fit together, see my guides to AI workflows and business process automation.
The review queue is the whole point
A review queue is a simple holding place. Nothing the AI writes goes to a client or into a system of record until a person approves it. For a consulting firm, this is not optional, because a wrong sentence in a follow up email is a wrong sentence on your letterhead.
Make the review quick by giving the reviewer a checklist:
Aim for the reviewer to spend five minutes, not sixty. If review takes longer than writing from scratch, the prompt needs fixing, not the reviewer.
A prompt template you can start with
Paste this above your notes or transcript:
You are helping a consulting firm. Below are notes from a client call. Produce: (1) a summary of no more than 8 lines; (2) a list of decisions made; (3) a table of tasks with owner, task and due date, writing "not agreed" where no date was stated; (4) open questions; (5) a draft follow up email to the client in a plain, polite tone. Do not invent facts, dates or names. If something is unclear, list it under open questions instead of guessing.
The last two sentences matter. They tell the model to flag gaps instead of filling them, which cuts down the invented details that cause most embarrassing mistakes.
A worked example (made-up figures)
Imagine a 6-person operations consulting firm. Each consultant runs about 8 client calls a week. Before AI, each call needs about 45 minutes of write-up. That is 6 hours a week per consultant, or 36 hours across the firm.
With the review queue approach, assume the draft takes 2 minutes of the consultant's time to start and 10 minutes to review and edit. That is about 12 minutes a call, or roughly 1.6 hours a week per consultant, or about 10 hours across the firm. The saving in this example is around 26 hours a week. Those figures are assumptions for illustration. Your numbers will differ, so time three real calls before and after.
The hidden cost is the review. If consultants start approving without reading, you have swapped time saved for risk taken. I would make approval a required click by a named person, not a default.
Routing the tasks into the tools you already use
A task list in an email is only half a result. The useful part is that tasks land in the place your team already looks. Pick one task tool and one place for client history, and route to those only. A new tool just for meeting follow ups makes the problem worse.
For weekly internal meetings, the same idea applies: see my guide to running a weekly team meeting and following up on action items. For recurring client updates built from the same material, see client reporting automation. If your follow ups end with booking the next call, an AI scheduling assistant can handle that step. And if the same steps repeat every time, they can become a written procedure with an AI SOP generator.
When not to do this
How I'd approach this
I would start with one consultant, one call type and the paste-it-in method. Time it for two weeks. Write down every error you catch in review, because that list becomes your improved prompt and your checklist. Only then connect it to your task tool and CRM.
I would keep the confidentiality rules short and in writing: what may go into an AI tool, which tool, on which plan, and who approves. And I would never let an AI send a client email on its own. The first thing to try is usually the follow up email draft, since it saves the most time and has the clearest review step. If you want help choosing where to start across the firm, the AI Strategy Sprint is a paid 3-day first step where you and I find where AI could transform your business and generate new value, and you get the exact steps written down, starting with a free 15-minute AI brainstorm call.
Questions people ask
Can AI write meeting minutes automatically?
Yes, as a draft. Tools like Microsoft Copilot in Teams and Zoom's AI assistant produce summaries and suggested action items, and a general AI model can do the same from a transcript. A person should check the draft before it is shared.
Is it safe to put client meeting notes into an AI tool?
It depends on your client contract, the tool and the plan. Anthropic states that its commercial products do not use your inputs or outputs for training by default, and Microsoft makes a similar statement for Microsoft 365 Copilot. Read the vendor's page for your plan, and ask a lawyer if your client agreement is unclear.
Does Copilot in Teams work for calls hosted by my client?
Microsoft says Copilot will not work in meetings hosted outside the participant's own organization. If your clients host the call, plan another way to get a transcript. See also my guide to getting meeting notes without a note taker joining.
Can Power Automate create meeting notes?
Power Automate is Microsoft's workflow tool. The reliable pattern is to send the output to a review queue first. Check Microsoft's current documentation for which Teams and Copilot features your licence includes.
How do I stop the AI inventing tasks or dates?
Tell it plainly in the prompt: do not invent facts, dates or names, and list unclear items as open questions. Then have a person check every task owner and due date against what was actually agreed.
Where should the tasks go after the meeting?
Into the one task tool your team already uses, and a short note into your CRM. Avoid adding a new tool. For internal meetings, see how to run a weekly team meeting and follow up on action items.
More in this topic: Problem solving: automating service business work (21 guides) →