Recommendations that go nowhere
The analysis was sound and the recommendation was politely ignored. It usually never named the decision, offered no alternative, or buried the uncertainty that would have made it usable. Decision, alternative, condition.
Most analysis that goes nowhere was not wrong. The numbers held up, the method was sound, and the reader said it was helpful. Then nothing happened: no decision, no objection, no follow-up question. A month later, the thing it argued against went ahead anyway.
The instinct is to make the next analysis more rigorous: more data, a better model, a cleaner chart. Occasionally that is the problem. Far more often the analysis was fine and the recommendation was the wrong shape. It answered a question, and the reader needed a decision.
These are the lessons I would hand to anyone who writes recommendations, at any level. The boardroom version is in Data in the boardroom, where the stakes make the problem obvious. The quieter version happens every week, on the last slide of a deck or in the last paragraph of a document, and that is where most of us learn, or fail to learn, how influence actually works.
Name the decision, in the reader’s words
A finding tells someone what is true. A recommendation tells them what to do about a decision they own. “Trial users who connect a data source in their first week convert far more often” is a finding. “Make connecting a source the first step of onboarding, ahead of the product tour” is a recommendation, and only if someone owns onboarding and is deciding what to change in it.
If you cannot name the decision, its owner, and roughly when it will be made, you have a finding. Send it. Just don’t be disappointed when nobody acts on it, because there was nothing to act on.
Use the reader’s words, not yours. The analyst writes “improve activation.” The product lead is deciding whether the new onboarding ships this quarter or next. The first is the analyst’s frame. The second is the one that gets a reply.
Offer an alternative you would accept
A single recommendation asks for a yes or a no, and the easiest answer to a yes-or-no question nobody asked for is “not now.” That is how a recommendation dies without anyone rejecting it.
An alternative changes the question from “do you accept my answer” to “which of these do you prefer,” and it shows you weighed the reader’s constraints, not just your analysis. It has to be real: one you would support, with its own cost stated. A straw option that exists only to make yours look good is spotted at once, and it spends the credibility the real one needed. Chip and Dan Heath make the same case in Decisive, their 2013 book on how organizations decide: choices framed as “whether or not” fail far more often than choices made between real alternatives.
State the condition that would change your mind
Uncertainty is where most recommendations lose the reader, in one of two ways. Hidden, it makes the recommendation sound surer than the evidence, which works until the first time it is wrong. Listed as caveats at the end, it reads as hedging and hands the reader a reason to wait.
State it as a condition instead: what would have to be true for you to be wrong, and how everyone will find out. “Make connecting a source the first step, and if activation among new trials hasn’t risen in six weeks, revert: the users who connect early may simply be the keen ones.” That one sentence admits the weakness, gives the reader a safe way to say yes, and says in advance what failure looks like.
Jeff Bezos drew the distinction that makes this work in Amazon’s 2015 letter to shareholders. Some decisions are consequential and hard to reverse, and deserve slow, careful deliberation. Most can be undone, and should be made quickly: he called them two-way doors. A stated condition turns your recommendation into a two-way door. That is why it gets a yes, and why it is easier to be wrong in, which matters more over a career than any single call.
The same recommendation, rewritten
The analysis behind both columns is identical. Only the shape changed.
| Before | After | |
|---|---|---|
| Decision | Not stated | Whether the onboarding change ships this quarter. Owner: the product lead, at planning in three weeks |
| Recommendation | “Consider emphasizing data connection earlier in onboarding.” | Make connecting a data source the first step, ahead of the tour |
| Alternative | None | Keep the tour first and prompt a connection at its end: a smaller effect for less engineering |
| Uncertainty | Four caveats about correlation, in a footnote | If activation among new trials has not risen in six weeks, revert |
| Ask | “Happy to discuss.” | A decision at planning, between the two options |
Before, the reader had to work out what the analysis meant for them. After, they only have to choose. The six results in the Tradeoffs exercise are written in the same shape: the choice, what it deferred, and the condition that would reverse it.
Take it back to the same person
When a recommendation has gone nowhere, rewrite it as decision, alternative and condition, and take it back to the same person. Not a new stakeholder, and not a new analysis. Going back to the same person is what makes it a test. If the rewrite gets a decision, the shape was the problem.
If it still goes nowhere, the cause is somewhere else, and that is worth knowing too. The decision was made before your work arrived, it was never theirs to make, or the timing is wrong. Each has a different fix, and none of them is a better chart. Owner and timing are two of the six questions in What would this work change?, which puts them, and four more about the choice itself, the evidence, capacity and support, to one decision you have in hand.
Declined is not the same as nowhere
A recommendation declined with a reason has not gone nowhere. It informed a decision, and the reason is information you did not have. The failure this essay is about is the other one: no decision, no reason, no reply.
Count “declined with a reason” as the format working, and keep a record of what you recommended, what was decided, and what happened. Over a year, that record becomes the best evidence you have of your own judgment. It is usually more useful, and more humbling, than anyone’s feedback.
What it costs
It takes longer. Finding out who owns the decision and when it will be made means a conversation before the analysis, not after. A real alternative means doing part of the analysis twice.
It exposes you. Naming the decision invites “that isn’t yours to decide.” Offering an alternative means the reader may choose the one you like less. A stated condition gives everyone a precise way to show you were wrong. A finding with caveats is safer. It is also easier to ignore, and that safety is most of the reason it goes nowhere.
Some readers want the number, not the options. Not every request is a decision in disguise, and sometimes the number really is the deliverable. The skill is telling the two apart, and asking what the number is for is usually enough.