The Framing Gap: why most bad decisions are well analysed

Most bad decisions are well analysed. That sounds like a contradiction. It isn't, and once you see why, you can't unsee it.

In fifteen years of reviewing decisions that went wrong, I can count on one hand the times the arithmetic was the problem. The models were sound. The numbers added up. The logic held. What failed was upstream of all of it: the question itself.

A sentence built in thirty seconds

A decision reaches your desk as a single sentence. "We need to decide on the second location." "We should hire a senior salesperson." "It's time to move to the bigger warehouse." That sentence was built in about thirty seconds, usually by one person, usually while they were doing four other things. It was never meant to carry weight.

But it does. Everything after it sits on top: finance models it, operations assesses it, someone builds the plan. Three weeks of competent, well-evidenced work, all of it inside a boundary nobody drew on purpose.

I call the distance between that sentence and the real decision the Framing Gap: the gap between the question you asked and the decision actually in front of you.

Three signs your Framing Gap is open

The question is binary. Real decisions rarely have only two options. When one arrives as a straight yes-or-no, options were removed before the meeting: silently, by one person, usually with no intent to narrow the field. They were thinking out loud, and the sentence stuck.

The question names a solution. "Should we hire a senior salesperson" already contains its own answer. Hiring is one candidate response. The decision underneath is usually about conversion, or pricing, or accountability, and the analysis, however sharp, never touches it, because it was commissioned to evaluate the hire.

The deadline has no source. Ask where it came from. Sometimes there is a real constraint: a lease, a filing, a season. Often nobody can trace it. Someone said "before the quarter" once, it got repeated, and it hardened into a fact. False deadlines are unusually good at holding a frame shut.

None of these are failures of analysis. The room is competent. The disagreement is real. The model is careful. Every visible signal says well-run process, and it is, inside the frame.

The verdict changed. The business didn't.

Here is the pattern, in the shape it usually arrives.

A decision to hire a senior salesperson. Revenue is flat. The sales effort today is the founder plus two juniors. The logic is clean: buy experience, buy pipeline, buy back the founder's calendar.

Read that question through the six lenses and it holds up reasonably well. Survivable financially at 60% of forecast. Operationally feasible. Consistent with the three-year direction. A defensible yes, with conditions.

Then one question does its work: what would have to be true for this decision to become unnecessary?

The honest answer, written down: if the two juniors were working a qualified list instead of a cold one.

That is not a hiring decision. It is a lead-qualification decision wearing a hiring decision's clothes. Read the same six lenses on that question, and the original doesn't survive contact:

The People lens stops asking "can we attract a senior hire" and starts asking who owns qualification today, and why nobody has said their name out loud.
The Financial lens stops modelling a salary and starts modelling twelve months of unqualified pipeline, which a new hire would inherit unchanged, at a higher cost per month.
The Risk lens names a different failure mode entirely: not "the hire doesn't work out", but "the hire works out, and the problem is still there".

Same business. Same evidence. Opposite verdict.

The uncomfortable part is what that says about the first analysis. It wasn't sloppy. Every lens was answered properly. It was rigorous work on a question shaped before anyone started.

Why more analysis doesn't help

Apply rigour to a mis-specified question and you get a more confident wrong answer, defended by better evidence.

Rigour applied to the wrong question is more dangerous than no rigour at all. It produces confidence, and confidence is what makes a decision hard to reverse.

Competence is the trap here. You can analyse brilliantly inside a frame and never notice the frame is wrong. The better you are at answering the question, the less likely you are to question it. And you can't out-think your own frame from inside it, because it is the thing doing the thinking. The more confident your read, the less visible its edges.

For an experienced operator, instinct is mostly a gift. Pattern recognition earned over years lets you read a situation in seconds that would take a novice a month to map. I would never tell a seasoned founder to distrust it wholesale. Your instinct reads the situation fast. It just can't tell you when it's reading the wrong situation.

Three questions that open the frame

I rely on a few deliberately awkward questions to break a frame open before analysing anything:

How would someone with no stake in this describe the decision?
What would have to be true for this decision to become unnecessary?
If I'm wrong about this, what's the most likely thing I'm wrong about?

None are clever. All are uncomfortable, which is exactly why they work. They force the frame into view.

Avoided conversations wearing a decision's clothes

The second question does the most damage, because of what it tends to reveal. The most expensive decisions I see aren't decisions at all. They are avoided conversations wearing a decision's clothes.

A business brings a big call to the table. Whether to restructure a team. Whether to bring in a new senior hire. Whether to invest in a system to fix a recurring problem. It gets analysed properly: costs, options, risks. And underneath a striking number of them is the same buried truth: the decision would evaporate if one person did their job differently, and nobody wants to say so.

The restructure is often a way of not addressing one manager who isn't performing.
The new hire is a way of not having the accountability conversation with the team you already have.
The system investment is a way of not telling two departments to actually talk to each other.

Each is a real expense, carefully justified, standing in for a conversation that would cost nothing but courage.

If the honest answer to "what would make this unnecessary?" names a person rather than a number ("if X actually owned this", "if Y and Z stopped working around each other"), you're not looking at a spending decision. You're looking at an accountability decision that has been converted into a spending one, because spending is easier than confronting.

The conversion is expensive twice over. You pay for the decision: the restructure, the hire, the system. And you still haven't fixed the thing that caused it, so it comes back, and you pay again. The money buys you a delay, not a solution.

This is why "what will it cost?" is the wrong first question on a lot of decisions. The right one is "what is this decision actually standing in for?"

None of this means the spend is always wrong. Sometimes the restructure is genuinely right. But you want to know you're buying the solution, not buying your way around a conversation, because only one of those two actually ends the problem.

AI inherits your framing too

This mattered before AI. It matters more now.

Type "should we open a second location" into an AI chatbot and it will help you brilliantly. It weighs the costs, models the footfall, lays out the risks, structures the whole thing. What it will never do is stop and ask whether "second location" was the right question: whether the real decision was about capacity you can't yet fill, or a sales problem a new site won't touch, or an accountability gap no amount of floor space will close.

It accepts your framing and optimises inside it. The more precise your question, the better the answer, and the narrower the box nobody checked. A vague question gets a weak response. A sharp question gets an excellent response to a decision you may have already mis-specified.

This isn't a flaw in the model. It is what the tool is for. A conversational system is built to give a useful answer to the prompt in front of it. Your framing is the prompt. Accepting it is the product. That is fine when the framing is right and expensive when it isn't, and you rarely find out which, because the answer is equally fluent either way.

So organisations racing to put AI into decision-making are, in many cases, industrialising their framing errors: faster, more fluently, at scale. The AI doesn't fix the wrong question. It answers it with more confidence than a human would have dared.

Start with the question

In any decision I work on, the first step isn't analysis. It is interrogating the framing before a single lens is applied. Checking the question first is also the first thing Viveka, the AI tool I built on the framework, does on any decision.

Get the question right and an average analysis serves you well. Get it wrong and the best analysis in the world, human or machine, just helps you be wrong on schedule.

Strategy quality is question quality. It always was. AI only raised the stakes on getting it right.