The Review That Made Things Worse
The founder had prepared for two weeks. He had examples, dates, quotes and a development plan. He opened the conversation by saying he wanted to be honest, then delivered forty minutes of accumulated observation.
The manager nodded. He thanked the founder for the candour. He left the room and updated his CV that evening.
Six months later the performance was unchanged and the relationship was gone. The founder concluded that the manager could not take feedback. The manager concluded that the founder had been storing complaints for a year and had chosen the review as the moment to empty the drawer.
Both conclusions were reasonable. The feedback had failed before it reached the content.
Avraham Kluger and Angelo DeNisi conducted a meta-analysis of feedback interventions and found a result that most managers have never heard. Feedback improved performance on average, and in over a third of the studies it made performance worse (Kluger & DeNisi, 1996, Psychological Bulletin). The average hides a wide distribution, and the distribution is not random.
Feedback is not a transfer of information. It is an intervention in a person's self-image, and the self-image defends itself.
Three Conversations in One
Douglas Stone and Sheila Heen describe feedback as three distinct conversations that people confuse: appreciation, coaching and evaluation (Stone & Heen, 2014, Thanks for the Feedback, Viking).
Appreciation is about relationship and recognition. It says: I see what you did and it mattered. Coaching is about improvement. It says: here is how the work could be better next time. Evaluation is about ranking and consequence. It says: here is where you stand.
Most feedback conversations mix all three without announcing which one is happening. The manager thinks they are coaching. The employee hears evaluation. The appreciation gets lost because it arrives wrapped in the same sentence as the criticism.
The confusion is expensive. A person who hears evaluation while being coached becomes defensive, because the stakes feel like their position rather than their technique. A person who hears coaching while being evaluated becomes confused, because the conversation seems to be about improvement when the decision has already been made.
The first discipline is to name the conversation. This is coaching, and there is no evaluation attached. Or: this is an evaluation, and I want to be direct about the conclusion before we discuss the path. Naming the mode removes the ambiguity that produces the defence.
Most founders skip the naming because they think it sounds clinical. The alternative is a conversation where the other person spends the whole time guessing which one it is, and the guessing consumes the attention the feedback needed.
What the Evidence Says About Specificity
John Hattie and Helen Timperley reviewed the feedback literature and produced a framework that separates four levels: feedback about the task, about the process, about self-regulation, and about the self (Hattie & Timperley, 2007, Review of Educational Research).
Feedback about the self, which includes praise and personal judgement, is the least effective for improving performance. Feedback about the task and the process is the most effective, because it gives the person something to act on.
This finding challenges the instinct to soften feedback with personal reassurance. You are talented sits at the self level, and it does not tell the person what to change. The task was late and the client had to reschedule twice sits at the task level, and it does.
Hattie and Timperley also distinguish feed-up, feed-back and feed-forward. Feed-up clarifies the goal. Feed-back describes where the current performance sits relative to the goal. Feed-forward describes the next step. A feedback conversation that contains only feed-back produces awareness without movement.
Kluger and DeNisi's finding fits this framework. Feedback that threatens the self elicits defence, and feedback that specifies the task elicits adjustment. The same content can be delivered at either level, and the level determines the response.
NLP's Meta Model makes the same demand in a different vocabulary. A vague judgement such as you need to be more strategic contains a deleted comparison and an unspecified criterion. The recovery questions are: more strategic than what, measured by which decision, and according to whose standard (Bandler & Grinder, 1975, The Structure of Magic I). The questions convert an evaluation into a specification.
The Delivery Mechanics
Content and delivery are separate skills, and most founders develop one at the expense of the other.
Timing. Feedback delivered months after the event is received as a case file rather than a coaching moment. The longer the gap, the more the recipient experiences the feedback as accumulated judgement. Deliver close to the behaviour, in private, and keep it short.
Quantity. One topic per conversation. A list of seven observations is processed as an attack, and the recipient remembers the emotional tone rather than the items. Choose the one change that would make the largest difference and leave the rest for later.
Behaviour and impact. Describe what happened and what followed. In the meeting on Tuesday, the client asked three questions and each answer started with a caveat. The client then asked for a written proposal instead of agreeing. The description gives the person the sequence, and the sequence is actionable.
Ownership of the judgement. Say I noticed rather than everybody thinks. The first is a report from one person. The second recruits a crowd the recipient cannot question.
The closing move. Ask what the person heard. The answer reveals the gap between the message sent and the message received, and the gap is usually larger than the sender expects. Correct the gap before the conversation ends.
The Founder's Own Receiving Problem
A founder who cannot receive feedback will not give it well, because the giving is modelled on the receiving.
Stone and Heen describe three triggers that make feedback hard to hear: the truth trigger, when the content seems wrong; the relationship trigger, when the source seems unqualified; and the identity trigger, when the content threatens a story the person holds about themselves (Stone & Heen, 2014). The identity trigger is the strongest, and founders have a large identity investment in being right about the business.
The practical response is to separate the signal from the source. A junior person can be wrong about the solution and right about the problem. A competitor can be wrong about your strategy and right about a customer complaint. The useful question is not whether the person has standing. It is whether the observation contains information you did not have.
There is a second discipline that matters more for founders. Ask for the feedback you do not want. The team already knows which topics are dangerous, and they will not volunteer them. A founder who asks what are you not telling me, and then waits through the silence, gets more information than a founder who asks for general feedback and receives a polite summary.
The receiving skill is also a cultural lever. A team that watches the founder take a difficult observation without defending will offer the next one. A team that watches the founder explain why the observation is mistaken will stop offering.
The Annual Review Trap
The annual review is the worst-designed feedback instrument in common use, and most companies keep it because it is familiar.
It compresses a year of observation into a single conversation, which guarantees that the recipient experiences it as a verdict rather than a coaching moment. It arrives with a rating, which converts every observation into evidence for the rating. It is delivered by a manager who has been storing the material, which means the recipient hears a case being built rather than a colleague being helpful.
The information problem is worse than the emotional one. A year of stored examples is a biased sample, because the manager remembers what was salient and forgets what was ordinary. The recipient is being evaluated on a highlight reel assembled by someone else's memory.
None of this means evaluation is unnecessary. It means evaluation and coaching need separate containers. Put the ranking in a short, explicit conversation with a clear conclusion. Put the coaching in frequent, specific conversations close to the work. Mixing them produces a review that is too late for coaching and too soft for evaluation.
A founder who moves to monthly twenty-minute coaching conversations and a separate quarterly evaluation will find that the annual review has almost nothing left to do. That is the sign the system is working.
The Written Version
Written feedback has different mechanics and different failure modes.
It removes tone, which is an advantage for precision and a disadvantage for relationship. The same sentence that lands as direct in a room can land as cold on a screen. The compensation is to add the context that tone would have carried: why the observation matters, what the standard is, and what support is available.
It creates a record, which changes how the recipient reads it. A message that might be coaching in conversation becomes evidence in a file when it is written down. The writer needs to be explicit about which one it is.
It is quick to send and hard to retract. The founder who writes a long critique at eleven at night is producing a document that will be read more times than any conversation. Write it, leave it overnight, and cut it by half in the morning.
The useful written form is short and specific. One behaviour, one impact, one next step, one offer of support. Four sentences can carry more than four paragraphs, because the recipient can act on four sentences.