The Ultimate Guide to 10x Better AI Outputs, for PE, VC and board leaders

The model answers before it understands. You would never hand a new analyst a task in one sentence and walk away. Most leaders do exactly that to AI a hundred times a week, then blame the model for the generic answer.

Here is what actually happens. You hand the model an empty brief, it fills every gap with a guess, and guesses average out. That is why so much AI output across our industry sounds identical. Not the same tool. The same missing context.

So I wrote down the method I run on my own work before anything real leaves my desk. I call it the Briefing Method, and the whole thing is below.

Why the brief beats the prompt

A prompt is an instruction. A brief is a transfer of judgment. The prompt tells the model what to do. The brief tells it what the work is for, who will read it, what it must not say, and what a strong version looks like. Every one of those is a decision only you can make, and every one you skip, the model makes for you, by guessing.

This is why prompt collections plateau. They polish the instruction and leave the judgment empty. You do not have a prompting problem. You have a briefing problem, and it has a fix that costs thirty seconds per task.

One boundary before we start. Mechanical tasks do not need a brief. Translate this, reformat that, fix the typo: just ask. The method below is for work with judgment and consequences in it, the memo, the analysis, the recommendation, the decision.

Rule 0: Where you run this

One rule before the first step. A confidential IC memo, an LP letter, a board pack, an unannounced number, does not go into a public chatbot. Everything here runs in a private workspace where your material never leaves your control and is never used to train a model.

That rule is set once and never bent, no matter how good the model gets. With it in place, everything below is safe to run on real work.

Step 1: Don’t let it answer yet

The first sign of a good analyst is the question they ask before starting. Hold the model to the same bar. On real work, its first job is to show you what it does not know, not to produce an answer.

So the first move of the method is simple. You give the rough task, and you forbid the answer:

Do not answer yet. First, brief me back:
1. The decision or outcome this work is for, as you understand it.
2. The context you are missing that would change the output.
3. The fewest questions that would close those gaps, five at most, and only questions whose answers change the result.
Start the work only after I confirm the brief.

Then answer its questions. Two minutes of back and forth, and you have handed over the judgment the output was always going to need.

That single pause converts the model from a fast typist into a careful analyst. Everything downstream inherits the quality of this step.

For leaders: if the first response is always an answer, your outputs are running on guesses. The fix is not a longer prompt. It is a shorter question list.

Step 2: The read-back

Boards read decisions back before they vote on them. Do the same here. After the questions are answered and before anything gets produced, make the model say the brief back to you:

Before you produce anything, restate the brief in your own words: the goal, who will read it, the constraints, what a strong version looks like, and what to avoid. One short paragraph. Then wait.

Read it. If the read-back is off, the answer was going to be off, and it is far cheaper to correct one paragraph of read-back than one page of confident error.

Most bad AI output happens because the model misunderstood the task and nobody caught it. The read-back is where you catch it.

For leaders: you already run this discipline with people. A direct report repeats the assignment back, you catch the drift in ten seconds. The model deserves less trust, not more.

Step 3: The gap review

The first version arrives. The instinct is to ask, can you improve this. Do not. “Improve” hands the judgment back to the model. Review it against the brief instead, the way you would mark up an analyst’s first pass:

Review your output against the brief:
1. Where it delivers on the brief.
2. Where it is still generic.
3. What is missing that the reader will ask for.
4. What should be cut.
5. The one change that would make it most useful to [the reader].
Then apply points 3 to 5.

The gap review turns the model into its own second pair of eyes, with the brief as the standard instead of its own taste.

For leaders: never grade work against “better.” Grade it against the brief. That is true for people and it is true here.

Step 4: The standing brief

If you keep typing who you are, who reads your work, what tone you use and what to avoid, you are writing the same brief by hand every session. Write it once instead. A standing brief is the permanent version, one per recurring job, saved where you run the work:

STANDING BRIEF: [the job, e.g. the quarterly portfolio review]
Reader: [who this is for]
Decision it serves: [the call this work feeds]
Inputs: [the documents it will receive]
Output: one page, answer first: the decision, the number, what changed.
Never: [assumptions and phrasings it must not make]
Always: flag anything inferred rather than read.

From then on, every session starts with the brief already loaded, and Step 1 only has to cover what is new about today’s task. Where does it live so it loads every time? In the project or custom instructions of whatever assistant you use, or as a saved note you paste to open the session. If you run your board and portfolio work in BoardLens, save it there once as a standing brief on the job itself.

The standing brief is where the method stops costing thirty seconds and starts paying it back.

For leaders: one standing brief per recurring job. The board pack has one reader and one shape. So should its brief.

Step 5: The swipe file, three briefs to steal

The three standing briefs I would set up first for anyone who sits on boards or runs a portfolio. Fill the brackets, save them, use them this week.

The board pack read:

Reader: a director walking into the meeting with 40 minutes to prepare.
Decision it serves: the votes and questions on this agenda.
Inputs: the pack, last quarter’s pack, the minutes.
Output: one page. The decisions being asked, the number that moved, what changed since last quarter, what is missing from the pack.
Never: summarise for length. Summarise for the decision.

The IC memo pressure-test:

Reader: the partner who will argue against this deal.
Decision it serves: invest, pass, or dig.
Inputs: the memo and its model.
Output: the three strongest objections, the assumption doing the most work, and the one number that would change the answer.
Never: soften. The meeting will not.

The portfolio update read:

Reader: the fund lead across [N] companies.
Decision it serves: where attention and follow-on go this quarter.
Inputs: every monthly update received.
Output: one page. Who moved, who slipped, the figure on page 40 someone hoped you would not rebuild, the two calls to make this week.
Never: average the portfolio. Rank it.

These three jobs eat more leadership hours than any others in this audience, and all three fail the same way, generic reads of specific documents. The brief is the difference.

For leaders: steal them as written, then edit the Never lines. That is where your judgment lives.

The boundary: when not to brief

One question decides it: would a different reader change this output. If the answer is no, translate the sentence, reformat the table, fix the typo, skip the method and just ask. If the answer is yes, there is judgment in the task, and judgment travels by brief.

The method, in one line

Rough ask. Brief back. Read-back. First version. Gap review. And for every job you repeat, a standing brief so you never explain yourself twice.

Everything in this guide runs in any capable assistant, as long as confidential material stays out of public chatbots. If the jobs you keep briefing are board packs, IC memos and portfolio reviews, that is exactly what we are building BoardLens for: a private analyst for boards and portfolios where your standing briefs live next to your real documents, never trained on your data.

Try it free at boardlens.ai. There is a demo company inside, so you can get a feel for it in under 5 minutes.