Six 360 feedback questions that actually surface a blind spot

A sales vice president opens his 360 report in a meeting room in Gurugram. Eleven people rated him. On "listens actively" he scored 3.4 from direct reports, 4.0 from peers, and 4 from himself. He reads the page, closes the folder, and says the sentence that ends most 360 cycles: "I do listen. I just do not agree with the number."

Nothing changes before the next cycle. Same item, same 3.4.

The problem is not his defensiveness. It is the item. "Listens actively" asked eleven people to supply their own definition of listening, score his behaviour against that private definition, and hand the result to a spreadsheet that averaged eleven different definitions into one number. The number looks precise. It cannot answer a single question he might ask it.

A blind spot is a disagreement, not a low score

Fix the meaning first. A blind spot is a gap between how you rate yourself and how others rate you, with you higher. A person rated 2 out of 5 by everyone and 2 by himself has a weakness, and probably knows it. A person rated 4 by himself and 2.5 by six direct reports has a blind spot. The number that matters is the delta, and most reports bury it under a competency average.

If the blind spot is a disagreement, the instrument's job is to produce disagreements you can inspect, not averages you have to trust.

Why most 360 items cannot produce one

Look at the verbs. "Understands the business." "Is aware of how his tone lands." "Values other people's input." "Considers the impact on the team." Bracken and colleagues, in the journal Industrial and Organizational Psychology, put it directly: a rater should not be asked to report what is happening inside someone else's head, which is what "understands", "believes", "knows" and "considers" demand.

The rater complies anyway. They guess. The peer who once saw you cut someone off guesses low. The report who watched you sit silent through a ninety-minute review guesses high. Both are answering honestly, and both answers are about a different behaviour, because the item never named one.

This is measurable. Dai and colleagues analysed 737 responses and found that when a leadership competency was abstract, self-other agreement was lower than when it was concrete. Abstraction does not just blur the data. It manufactures disagreement, and the report then sells that disagreement as insight about you.

Agreement is thin even with concrete items. A meta-analysis by Conway and Huffcutt put the correlation between two supervisors at .50, two peers at .37, two subordinates at .30. So "peers: 3.6" is an average of partially disagreeing observers, and the average hides exactly what you need.

The test: could a rater answer this without leaving the room

Before rewriting anything, apply one question to each item in your instrument. Could a rater answer it from memory of a specific event, without inferring motive or character? If the answer is no, the item is an opinion poll with a five-point scale.

Three rules do most of the work.

1. A visible verb. If the item can be true while nothing observable happened, it is an inference item. "Explains the reasoning behind a decision" is visible. "Is transparent" is not.
2. A window and a count. "In the last 30 days" beats "generally". A window forces retrieval of instances instead of a summary feeling.
3. The instance, not the impression. Ask what happened, in what meeting, with what words. Impressions collapse into ratings. Instances survive as evidence.

The six questions

Six is enough for one behaviour set. More items lower completion. Keep the format identical across all six so answers read together.

1. The counted frequency. "In the last 30 days, how many times did [name] do [one specific, visible act] in a meeting you attended? 0 / 1 / 2 to 3 / 4 or more / I did not observe this." The last option is not a courtesy. Forcing a rating from a rater who never saw the behaviour is a main source of noise.

2. The last instance. "Describe the most recent time you saw [name] do this. What was the situation, and what did they say or do?" This produces the quotable example, which is what actually changes behaviour. A number invites argument. A Tuesday afternoon invites a reply.

3. The missed instance. "Describe a time in the last month when the situation called for this and it did not happen. What happened instead?" This is the blind spot detector. Nobody can defend "I do listen" against three people naming three meetings. The framing matters: the situation called for it and it did not happen, not "he failed". One asks for an event, the other for a verdict.

4. The cost. "When this happens, or does not happen, what does it cost your work? Name the delay, the rework or the decision that went the wrong way." This turns a personality complaint into a business case, and it decides which answer deserves a development goal.

5. The stop question. "What does [name] do that they probably believe helps, that you would ask them to stop?" Overused strengths are the most common blind spot in senior people, and almost no instrument asks about them. A manager who prepares every deck himself believes he is protecting quality; his team experiences a bottleneck. Some practitioners use bipolar "too little to too much" scales for this reason, because a frequency scale cannot separate a strength from an excess of it.

6. The one change. "If [name] changed one thing in the next 90 days, which change would most improve your ability to do your job?" This gives you the development goal in the rater's own words, and it doubles as the re-measurement item at 90 days.

Run them in this order, self first

The ratee answers all six about themselves before seeing anything from anyone else. Timestamp it. That is the baseline for the delta, and collecting it later contaminates it.

Then send the same six to raters in the same order: frequency first because it primes recall, the two incident questions next while attention is high, then cost and stop, then the one change as the close.

Report per item, never as one score: the self-answer, the distribution of others' answers, and the spread. The spread tells you whether you have a shared view or two camps. An average tells you neither.

Reading the result without a statistician

Sort the six items into three buckets. No statistics needed.

The third bucket is where most reports go wrong. When a group splits, the average is an artefact of who was in the room. Nowack and Mashihi recommend reporting the range and distribution alongside the average. If a group has fewer than three respondents, do not report it as a group. Greguras and Robie calculated that .70 reliability needs roughly four supervisors, eight peers and nine direct reports, which almost no manager has, so treat small-group numbers as questions, not facts.

A worked example with illustrative numbers

The figures below are illustrative, not a case study. The structure is the point.

A regional sales manager, 11 raters: one manager, four peers, six direct reports. On the old item, "communicates decisions clearly", he scored 4.1 overall and the report declared communication a strength. On the six new items:

• Counted frequency, "restated the decision and the reason before ending a call": self 4 or more, direct reports median 1.
• Last instance: two reports name a pricing exception in the first week of the month. One quotes him: "just get it done, I will explain later".
• Missed instance: four of six name a different call, where a discount was approved and the reason never reached the account team.
• Cost: the account team quoted the wrong floor price on three accounts, and one order was reworked after a two-day delay.
• Stop question: three reports say he joins calls he need not join, which reads as checking up.
• One change: "say the reason in the same breath as the decision, or send one line after the call".

The old report gave him a strength. The new one gave him one sentence to say differently and a cost attached to not saying it. At 90 days the same frequency item is the re-measure.

What this does not fix

Better items do not make people honest. If raters fear a low score, they will rate high, and behavioural items do not change that. The evidence points the same way across cultures: a study of 964 managers in 21 countries found self and other ratings more positive where assertiveness and power distance were both high, and a 31-country study found larger self-observer gaps on decisiveness and composure in high power distance cultures. India sits in that pattern. Design for it by keeping the process developmental and treating the incident questions as the real data.

Three limits, plainly. Six items on one behaviour set do not cover a competency framework, so this supplements your instrument rather than replacing it. Behavioural items can miss a problem nobody has language for, which is why the open comment box stays. And none of this proves the behaviour matters commercially, which is a separate question with its own method.

One caution from the feedback literature. Kluger and DeNisi pooled 607 effect sizes from 23,663 observations: feedback improved performance on average, effect size .41, but performance declined in more than a third of the interventions. Attention pulled toward the self and away from the task is where it goes wrong, and a report full of trait language does exactly that.

Start with your own instrument this week

Open your current questionnaire and count the items with an inference verb: understand, believe, know, value, consider, be aware. If more than half do, your report is measuring your raters, and no coaching on the receiving end will fix it.

Then rewrite six items for one behaviour that matters this quarter, using the three rules above, and pilot them on a single manager with at least three raters in each group. Compare the answers against the old score for the same behaviour. If the new items produce named incidents and a cost, a manager can act on Monday. If they produce the same polite numbers, the problem was never the questionnaire, and you learned that cheaply.

BolderBrain reports 360 results as item-level distributions with incident capture, so the delta is visible without a spreadsheet exercise. The workflow is on the platform features page; a single-score tool is compared on the comparison page.