How to Turn Raw Thoughts into a Proposal for Your Manager with AI
How to brief AI when you don’t have a solution yet: share your raw notes, ask for clarifying questions, answer them, and assemble a concise proposal for your manager. A real dialogue and a six-step checklist.
You need to prepare a proposal for your manager, but all you have so far are scattered notes on a few points: what isn’t working, what your colleagues are suggesting, and why you have doubts.
How do you brief an AI when you haven’t settled on a solution yourself yet?
The answer is simple — though, as always, there are nuances. Paste those raw thoughts into the chat as they are. State your goal and the recipient (the person who will read the memo), and ask the AI to ask you questions whose answers could change the proposed course of action.
Then:
- answer the questions one by one;
- add whatever was missed;
- choose an option from those the AI helps frame;
- instruct the AI to assemble the memo.
You don’t need to know the solution in advance to get started — this is genuinely important, and in practice it is a habit that takes time to build.
For a real workplace task, first check what format of AI use your organization’s policies allow, which tool is approved, and what information may be shared with it (for more detail on this topic, see the article “AI Is Banned in Your Organization. What You Can Do Right Now”). If the policy is unclear, check with the person responsible for that area — and in the meantime, let’s practice on the example below.
Share Your Thoughts with AI as They Are
Consider a hypothetical equipment maintenance company. Clients wait a long time to receive commercial quotes, and an employee wants to propose a way to shorten that turnaround time to their manager. The AI responses in this walkthrough are completely real.
Here is the first message the employee sends to the AI chat:
My raw thoughts:
- Clients wait a long time for commercial quotes on equipment maintenance.
- Writing the quote text itself takes very little time.
- We have to follow up with the client several times to clarify which equipment needs servicing, what work is required, and when to start.
- The service department calculates the cost of the work. Then the finance department checks the price and discount.
- Colleagues suggest writing commercial quotes with AI so we can reply to clients faster.
I want to prepare a proposal for my manager on how to shorten the time from a client’s request to sending the commercial quote. Do not write the memo or pick a solution yet. Start with three questions whose answers could affect the final proposal. Do not invent causes of delays or numbers.
Every Question Must Influence the Final Outcome
Why is the prompt structured this way?
The goal here is to shorten the total turnaround time for the client. Writing the quote text with AI is the colleagues’ idea, but it is not yet clear whether that will actually help reduce total response time. In the prompt, the employee specifies the exact outcome they want to reach through the dialogue with AI: a proposal that enables the manager to make a decision.
Asking the model not to write the memo yet sets the workflow order:
- first establish the grounds for choosing an approach;
- then draft the proposal.
That way, the employee does not have to invent the proposal upfront just to give the AI a prompt for a finished document.
A question is useful if the answer to it can change the outcome. For instance, knowing where the main waiting time occurs is essential for choosing which step to analyze first. That is why you should ask AI to pose questions on which the proposal actually depends, and forbid it from filling gaps with fabricated causes or numbers.
The AI returned three questions. Below are exact excerpts from its response:
Where is time actually being lost? For recent requests, do you have timestamps for each step: request received, each clarification with the client, handoff to the service department, cost estimate issued, finance department review, quote sent? If no such data exists, the first step might be measurement rather than changing the process.
Why are there multiple rounds of clarification instead of one? Does the client not know what they need on the first contact, do we fail to ask everything upfront, or do new questions emerge only during cost calculation in the service department?
What does the finance department check, and is that review needed in every single case? Is every quote reviewed, or only those with non-standard pricing or discounts?
The three questions target three distinct grounds for a decision: 1) turnaround time data, 2) how intake information is collected, and 3) approval rules. Different answers would lead to very different proposals for the manager.
Add What Was Missed and Instruct AI to Assemble the Memo
While answering, the employee notices one more gap: who confirms that the collected information is sufficient to start the cost calculation? They add that detail on their own, define a small pilot sample, and ask the model to put the memo together.
They send all of these pieces in a single message inside the same chat, repeating each question before its answer so neither the human nor the AI loses track of earlier turns:
Answering in order.
Where is time being lost? We haven’t measured the duration of individual stages yet. Email threads show the date a request arrived and the date the quote went out, as well as bounce-backs for clarification. So I cannot name the primary cause of delays yet.
Why are there multiple clarifications? The account manager forwards the request to the service department without checking whether the equipment, required work, and desired start date are specified. The service engineer spots the gaps during cost calculation and sends the request back to the manager.
What does finance check? Right now they check the price and discount on every quote. In this pilot I am not proposing to change the finance approval flow: first I want to sort out how we collect initial details from the client.
While answering, I realized that a checklist of questions alone isn’t enough. We need to define who verifies that the answers are sufficient for calculation. My proposal: the account manager collects the details from the client, and the service engineer confirms their sufficiency before calculation begins.
I want to propose a small experiment to my manager on the next five requests for the same service type. The account manager and service engineer will draft a short intake checklist and use it before calculation. We will see whether bounce-backs and total turnaround time from request to sent quote decrease, and how much time the new check takes. For comparison, we will pre-select completed requests for the same service of comparable scope. Before starting, we will align with the manager on what result would justify continuing. Five requests will provide material for the next decision, though they won’t prove a permanent effect.
Assemble this into a short memo to my manager, under 180 words: the problem, what I propose and why, how we will test it, and what decision is needed from the manager. I am asking to approve the experiment and the participation of the account manager and service engineer. Preserve what is unknown; do not invent improvement numbers or assume people have already agreed.
Add material details even if the AI didn’t ask about them. Here, the AI’s questions pointed out what needed clarification, while the employee defined the role split, designed the trial, and specified what approval was required. Those answers became the substance of the memo.
Bounce-backs for clarification do not by themselves prove that intake gaps are the main source of delay. If the longest wait happens at another stage, the new intake step might barely move total response time. That is why the employee proposes testing a specific change while explicitly keeping the primary cause of delays marked as unknown.
The closing paragraph of the prompt defines the shape of the document: what the manager needs to understand and what decision they are being asked to make. Now the AI has everything it needs to assemble the memo.
Verify the Facts and the Links Between Them
Here is the resulting memo in full, unedited:
Memo to Manager
Problem. Clients wait a long time for commercial quotes on equipment maintenance. Email records show that requests bounce back several times for clarification, even though drafting the quote text itself is fast. We have not measured the duration of individual stages, so I cannot yet name the primary cause of the delays.
What I propose and why. Currently, the account manager forwards a request to the service department without checking whether the equipment, required work, and desired start date are included. The engineer spots the gaps during calculation and returns the request. Proposal: the account manager and service engineer put together a short checklist of required initial details. The manager collects them from the client, and the engineer confirms they are sufficient before calculation starts. We do not change the finance department approval flow at this stage.
How we will test it. An experiment on the next five requests for a single service type. We will compare against completed requests for the same service of comparable scope: number of bounce-backs, total time from request to sent quote, and time spent on the new verification step. Before launching the experiment, we need to align on the form and threshold of results that will justify continuing. Five requests will provide data for the next decision, but will not prove a sustained effect.
Decision requested. I ask you to approve running the experiment and authorize the participation of the account manager and service engineer.
The memo makes clear what the proposal rests on and what decision is needed from the manager. The unknown primary cause of delays was not inflated into an established fact, and the participation of the manager and engineer is properly framed as pending approval.
One phrasing detail does need tightening: the completed baseline requests can only be compared on the number of bounce-backs and total turnaround time. The time spent on the new verification step has to be measured separately — there is no historical baseline for it in the inputs. Before sending, replace the comparison sentence with two distinct sentences:
Against completed requests for the same service of comparable scope, we will compare the number of bounce-backs and total turnaround time from request to sent quote. Separately, we will measure the time required for the new verification step.
Before sending, verify the facts, conditions, and causal links between them. Are all claims backed by your input data? Did a possible cause get rewritten as a proven one? Did compression strip out any material qualifiers?
In this dialogue, the AI’s questions helped sharpen the initial raw notes, the employee supplied the missing operational piece and chose the pilot design, and the AI assembled the memo. The result is a clear brief ready for discussion with the manager, while the actual impact of the proposed change will be tested in practice.
Try It on Your Own Task
Pick a note you have been putting off until you had time to “phrase it properly.” For your very first message, a rough list of bullet points, your goal, and the recipient are already enough. Depending on your situation, the next step might turn out to be gathering missing data, testing a different process tweak, or deciding not to change anything yet.
- Share your thoughts as they are. Name the recipient and goal; distinguish what is known fact from what is still an idea or assumption.
- Ask the AI to ask you clarifying questions before drafting the document — specifically questions whose answers could change the proposal. Instruct it not to invent missing causes or numbers.
- Answer the questions in the same chat and in the same order. Repeat each question before your answer, state explicitly what remains unknown, and add important details the AI didn’t ask about.
- Choose your proposal or define the first step (depending on the situation). If key data is missing, propose gathering it first. For an experiment, specify what will change and how you will evaluate the result.
- Prompt the AI to assemble the memo. Specify the required sections, word limit, and the exact decision you need from your manager. Instruct it to preserve unknowns and conditions.
- Verify the finished text against your source facts. Check facts, roles, causal claims, and promises. Fix ambiguities and make sure brevity hasn’t stripped away essential meaning.
Checklist: How to Turn Raw Thoughts into a Proposal for Your Manager