A 360 report changes nothing until someone follows up

Twenty-six managers completed a 360 in January. The reports arrived in February, were opened in a two-hour workshop, and each participant left with a development plan. The plans were one page each and all of them were written. By June, nineteen said the same thing in different words: improve stakeholder communication, build stronger cross-functional relationships, be more visible to senior leadership.

In September the re-rating window opened. Direct report scores had moved by a tenth of a point. The programme head asked whether the instrument was broken. The instrument was fine. Nobody had run the part of the cycle that changes behaviour, because that part does not come in the vendor's box.

A 360 is a measurement. It produces information about how a group of people currently experience one person's work. Information does not alter behaviour on its own, and the research on this is unusually clear and unusually ignored.

The ratings barely move, and we know why

The most thorough summary of the evidence is Smither, London and Reilly's 2005 meta-analysis of 24 longitudinal studies of multisource feedback. It computed how much ratings actually improved after a feedback cycle, corrected for measurement error. The corrected mean effect sizes were 0.15 for direct reports, 0.15 for supervisors, 0.05 for peers, and −0.04 for self-ratings.

Two of those numbers deserve attention. A peer effect of 0.05 is statistically near nothing: peers reported no meaningful change in their colleague's behaviour a year later. And self-ratings going slightly negative is not a failure. It is the most reliable thing in the whole table. People read their report, discover the gap between how they see themselves and how others see them, and revise their self-image downward. That is learning. It is just not behaviour.

This is not a quirk of 360s. Kluger and DeNisi's 1996 meta-analysis of feedback interventions, 607 effect sizes covering 23,663 observations, found an average improvement of d = 0.41, and that more than a third of the interventions made performance worse. Their explanation is the useful part: feedback moves attention. When it moves attention to the task, performance improves. When it moves attention to the self, to how one is being judged, it consumes exactly the capacity the task needed.

So the honest mechanism is this. A 360 does not install a behaviour. It creates a fortnight of heightened self-awareness, during which the person notices their own patterns, and then the quarter closes, a customer escalates, and the report sits in a folder. By the time the next cycle opens, the memory of the specific examples has gone, and the person is left with a theme. Themes cannot be acted on. Only situations can.

One more design flaw guarantees the decay. The rater window closes in January, the report lands in February, and the next measurement is twelve months later. Smither's analysis found improvement was larger when the two administrations were separated by less than a year. The shorter the gap, the more the person has to work with. Most programmes build the longest possible gap and then wonder why nothing accumulates.

Three dates, one page, ninety days

The follow-up that works is not a coaching programme. It is three short meetings and one line in a record, scheduled before the report is sent. The Centre for Creative Leadership's review of 360 practice puts follow-up as the sixth of six required steps, after action planning, and cites Goldsmith and Underhill's finding that managers who follow up on their formal feedback improve significantly more, with the frequency of follow-up tracking perceived positive change.

Here is the structure, with the owner and the artifact for each step.

Total manager time per participant is about 65 minutes, plus 30 minutes of a peer's time. That is the entire intervention. It costs no licences and no external facilitator.

The table has one non-obvious rule: the report is not the input to the day 0 meeting. It was written about a period that has ended. The input is the participant's own reading of it, prepared by highlighting the three items whose examples they can still remember. If a behaviour cannot be recalled with a specific incident, it is not the one to work on, however low the score.

The commitment that survives is an if-then, not a goal

Most development plans fail at the wording. "Improve stakeholder communication" is a category. "Be more assertive in escalation meetings" is a goal. Neither tells the person what to do at 11:40 on a Tuesday when the situation is actually in front of them.

An implementation intention is a plan in the form "if situation X occurs, then I will do Y". Gollwitzer and Sheeran's 2006 meta-analysis pooled 94 independent studies with more than 8,000 participants and found an average effect on goal attainment of d = 0.65, against control groups that had already formed the goal. The mechanism is that the situation, not the person's memory, triggers the action.

Three tests decide whether a commitment is worth the 90 days.

• Observable by someone who was not in the room. "I will be more present in reviews" fails. "In every review I chair, I will ask the two quietest people first" passes.
• Fires at least weekly. A behaviour that comes up once a quarter will not build a habit in 90 days, and the day 60 count will be one.
• Survives a bad week. If the plan requires twenty minutes of preparation, it dies in the first busy fortnight. If it requires one sentence, it lives.

Applied to the manager with the low stakeholder scores, the plan might read: if a delivery date changes after the client has been told, then I post the new date and the reason in the account channel the same day, and Priya checks on Friday. One behaviour, one trigger, one witness, one check.

The witness is the part everyone skips

An unwitnessed commitment is a private intention, and private intentions are cheap. The witness is not accountability theatre. They are the memory of the plan and the person who can tell you whether the count is real.

Choose one rater who was already inside whatever anonymity threshold your report uses, so nothing about their individual ratings is revealed by the role. Two 15-minute conversations, day 30 and day 60. They are not evaluating. They are asking one question: how many times did it come up, and how many times did it happen. If you are still designing the rater pool, the rules for a pool that survives a friendly list matter here, because a witness chosen by the participant is a witness who will be kind.

Two choices destroy the witness role. Making the manager the witness makes the participant censor, because the manager holds the appraisal. Making HR the witness turns a development conversation into compliance, and the counts become optimistic.

What the manager does, and what they must not do

The manager's job is the calendar and the decision, not the coaching. Three tasks: protect the 45 minutes on day 0 and refuse to leave until the page names one behaviour; ask the day 90 question as a question about a plan, not about the person; write keep, change or drop into the record and mean it.

Walker and Smither's five-year study of upward feedback, summarised in Atwater and colleagues' field experiment paper, found that managers who held a follow-up meeting with their subordinates to discuss the results improved more than managers who did not. That is the cheapest lever in the entire cycle, and it is free.

What the manager must not do is let the follow-up become evidence for anything else. The moment a 360 commitment is scored in an appraisal, raters adjust what they say, and the next cycle measures the adjustment rather than the behaviour. The anonymity decision has to be made before the window opens, not after the report lands. If the organisation needs the 360 to carry consequences, it should stop calling it development and staff it like an appraisal.

A worked example, with illustrative numbers

A 400-person services firm runs a 360 for 26 managers. The instrument and platform cost 91,000 rupees, call it 3,500 per head. Rater time is six raters at 25 minutes each, 65 hours in total. That is the cost of the measurement, paid whether or not anything follows.

Option A is the common one: a two-hour report workshop and a development plan, then nothing. Add 130 hours of participant time and zero follow-up hours.

Option B adds the 30/60/90 structure: 26 managers at roughly 65 minutes each, about 28 hours, plus 26 witnesses at 30 minutes, about 13 hours. Call it 41 hours, and at a loaded cost of 1,100 rupees an hour, about 45,000 rupees.

The numbers below are illustrative, not measured. In this shape of programme, Option A typically lands a direct report mean moving from 3.4 to 3.5 across the year, while Option B lands around 3.8, with 17 of 26 participants able to show a day 60 count for the behaviour they committed to. The ratio is the argument: the follow-up costs about half what the instrument costs, and it is the only part of the cycle with a body of evidence behind it.

Where this breaks

The structure assumes the 360 is developmental and that its data is worth acting on. Four situations break it.

If the report secretly feeds promotion or increment decisions, raters are already managing impressions, and a 90-day follow-up is polishing contaminated data.

If teams are small, three raters cannot support anonymity and a witness at once. The witness becomes identifiable, and the counts get polite. In that case use observed work samples instead of a 360.

If the person being rated is the manager of the people who hold the calendar, the manager cannot run the follow-up, because they are the subject. L&D or a peer has to hold the three dates.

If the gap is a genuine skill deficit rather than a missing cue, an if-then plan will fire reliably and fail reliably. That person needs practice with feedback, not a trigger. The two problems look identical in a report and are completely different in a room.

Before the next window closes

Take the last 360 your organisation ran. Pick five participants. Book their day 30 fifteen-minute slot now, before anyone reads a report, and add one line to the report's cover page: bring one behaviour and the name of one rater to the handover meeting.

Five is enough to learn whether the structure survives your managers' calendars. If five hold, run the next cycle this way and let the tool do what it is good at, which is holding the three dates, the one line and the count in one place. That is what a system that records commitments rather than scores is for. If two of the five never make the day 30 call, the problem is not the 360. It is that nobody owns the ninety days after it.