← PerspectivesKareem Soliman · 12 min read

The organisation reads you back

Firāsa is the practice of reading what cannot be seen from what can. I have written about it as a habit of attention, as a research discipline, and as a thing that goes wrong on a basketball court. I want to write about it now in the place where I was actually trained to practise it, and where I think it is practised more relentlessly than anywhere else: inside organisations, by people who would never use the word.

Nowhere is the gap between the seen and the unseen wider, or the tools for crossing it more imperfect, than in organisational life. Every consequential decision an organisation makes about a person is an inference from an indicator to something the indicator is not. We hire on a conversation and a task. We promote on a performance record. We judge a leader by a team. We reward against a number. And at each of those points, someone is doing firāsa, with a tool that carries less of the decision than the decision assumes.

Consider the first instance most of us meet.


A person applies for a job. They are interviewed, and perhaps given a task that simulates the work. From an hour of conversation and a few hours of simulation, a hiring manager is asked to predict how this person will perform in the role over the next several years.

That prediction is already a reading. The interview is a surface. The task is a surface. The thing the manager wants to know, whether this person will be good at the job, is not on either surface. It has to be inferred.

Then the person starts, and the prediction meets everything it never measured. How they get on with the people beside them. How they respond to the particular leader they have been given. Whether the way they work fits the way the place works. Whether they have capabilities nobody asked about because the advertisement did not list them, which they then quietly lean on to get things done. The meta-analytic literature on person-environment fit says these things matter for performance and for whether people stay, and almost none of them are visible in a selection process. Kristof-Brown, Zimmerman & Johnson, 2005.

Even the parts we do measure carry less than we tend to assume. For decades the standard figures in personnel selection came from a 1998 meta-analysis that put cognitive ability tests and work samples near the top of the predictive league table. In 2022 a team led by Paul Sackett went back through those estimates and found that a statistical correction for restricted range had been applied far too generously. The revised numbers are humbler. Structured interviews sit around .42, job knowledge tests around .40, work-sample tests around .33, cognitive ability around .31. Sackett, Zhang, Berry & Lievens, 2022.

When I first drafted this paragraph, I wrote that simulated work is about as close to the real thing as selection can get. The revision corrected me. A well-built structured interview edges it out. I mention this because it is exactly the kind of assumption a psychology-trained person carries around for years without checking, and I would rather show the correction than pretend I did not need it.

A correlation of .4 is a real relationship and a professional selection process should use it. It also leaves most of the variation in later performance unexplained. Those two sentences are both true, and it is surprisingly hard to hold them at the same time while making a hiring decision.

Capability makes it harder still. Each role has its own set of requirements, and each requirement breaks into smaller ones, each of which would need its own demonstration to be measured well. Nobody does that, because nobody could. So the manager looks at a small sample of behaviour and infers a large territory of capability from it. If they do everything a careful manager could reasonably do, they are still, at the end, deciding whether to trust a reading.

The same thing happens again at every internal move. Is this person suitable for a different role in the same organisation? The evidence is a record of them doing a different job. The inference is the whole decision.


Suppose the reading holds and the person gets the job. The first thing they meet is the culture.

Culture in the organisational literature has acquired more definitions than most concepts survive. For the purposes here I mean the informal one that practitioners have used for decades: the way we do things around here. Deal & Kennedy, 1982. The unwritten code. What I mean by unwritten is precise: none of it appears in a policy, a handbook, or a procedure. It lives in the attitudes, the interactions, the things that are frowned upon but never punished and the things that are smiled at but never directly rewarded.

A new employee reads all of this from observable indicators, in their first weeks, with no instrument at all. Who talks first in meetings. What happens to the person who disagrees with the director. Whether leaving at five is noticed. They are practising firāsa on the organisation while the organisation practises it on them.

People have tried to measure culture. There are validated instruments, and there is meta-analytic evidence that some of what they capture relates to organisational outcomes. Hartnell, Ou & Kinicki, 2011. But the thing itself is so pervasive and so multifaceted that passing it through a survey produces a score that is at best a compressed shadow of it. Whether you read culture by observation alone or through an instrument, you are inferring something intangible from what can be observed. That is not a reason to throw the instruments out. It is a reason to know what the score can and cannot stand for.

Of all the things that shape a culture, leadership arguably has the widest reach, which makes it worth a section of its own.


How do organisations decide who should lead?

In a great many cases, technical ability is enough to earn someone the opportunity. The best engineer becomes the engineering manager. The best salesperson gets the team. Leadership, though, runs on a different set of capabilities: communication, judgement about people, conflict, the capacity to make others better rather than to be the best in the room. The observable indicator, a strong individual performance record, is frequently a poor guide to how a person will lead, even when the leadership role still needs some of the technical skill.

This one has been measured directly. Alan Benson, Danielle Li and Kelly Shue looked at more than fifty thousand sales workers across 214 firms and found two things. Sales performance strongly predicted who got promoted into management. And a worker's sales performance before promotion negatively predicted how their subordinates performed afterwards. The better the salesperson, on average, the worse the manager. Benson, Li & Shue, 2019. Firms were reading leadership potential from the wrong surface and getting it wrong in a direction you could measure. The authors suggest some firms accept this knowingly, because promotion is also an incentive for the people below, which is its own kind of reading: the promotion is telling the sales floor what to do, not only who should lead it.

Once someone is leading, the reading does not get easier. The most common indicator of a leader's efficacy is the performance of their team or department. But the number of things that drive a team's results, beyond the person nominally responsible for them, is very large: the market, the people they inherited, the budget they were given, the decisions made two levels up. Attribution research has a name for our tendency to credit and blame leaders for outcomes they only partly control; it is called the romance of leadership. Meindl, Ehrlich & Dukerich, 1985. Judging a leader strictly by outcomes is unfair even when they are, in the end, responsible for them.

More specialised tools exist. A 360-degree assessment gathers ratings from a leader's manager, peers and direct reports, and gives a much richer picture than a results line. But look at what those ratings are made of. Personal history with the leader. The last interaction, which may have gone badly. Each rater's own capability to observe leadership behaviour and their own idea of what good looks like. Very few of the people asked to rate a leader have been trained to evaluate leadership, and yet their judgement is what the evaluation rests on.

The size of that problem has been estimated. When Scullen, Mount and Goff decomposed the variance in multi-rater performance ratings, the largest single component was not the ratee's performance. It was the rater. More than half of the variance in ratings was idiosyncratic to the person doing the rating. Scullen, Mount & Goff, 2000. A 360 is partly a reading of the leader and partly a reading of everyone who filled it in.

That is not a reason to throw the tool out. I notice I have now said that twice.


Then there are KPIs.

Of all the tendencies I have watched in organisations, the heaviest reliance is on key performance indicators: for establishing whether someone is performing, and then for the bonus, the promotion, or the demotion that follows. KPIs will always be prime targets for what economists call Goodhart's law, usually rendered as: when a measure becomes a target, it ceases to be a good measure. Goodhart, 1975; Strathern, 1997.

I want to tell you what that looks like from the inside, because I have been the inside.

I once worked in a sales role where what I cared about was giving people good service and making sure they left with what they needed. Service was not on the KPIs. Sales figures were. When my numbers started falling short and management made it clear that the numbers were what mattered, the other things started to go. I stopped spending as long on the phone with a customer if I did not have to. I stopped helping the people around me as much, because that time was not counted anywhere. I was still on the phone. I was still, technically, serving. But the thing I had been doing well was invisible to the thing I was being measured on, so it was the thing that gave way first.

I am not proud of that, and I do not think it makes me unusual. The goal-setting literature is one of the most replicated bodies of work in organisational psychology, and it says clearly that specific, difficult goals raise performance on the goal. Locke & Latham, 2002. What the same literature's critics documented, in a paper whose title I have always enjoyed, is what the goals do to everything else: narrowed attention, neglect of whatever is not measured, more risk-taking, less cooperation, more corner-cutting. Ordóñez, Schweitzer, Galinsky & Bazerman, 2009. So the honest answer to whether KPIs make people perform better is: better on what the KPI counts, and often worse on what it does not, with the net effect depending on how much of the job the KPI captures. I used to think of that as an open question. I now think it is two questions, with different answers.

Organisational psychology has a name for the things that went first in my case. Organisational citizenship behaviour: helping colleagues, taking on what is not strictly your job, the small discretionary acts that make a workplace function. It has been studied for decades and it relates, at the unit level, to performance outcomes organisations actually care about. Podsakoff, Whiting, Podsakoff & Blume, 2009. You will rarely find it on a formal KPI. Managers do seem to weigh it when they form their own impressions of people, which is something, but an impression is not a target, and when the target tightens the citizenship is what people stop doing.

This is where reading people as if they were texts starts cheating. A page does not change its words because you are grading it. A person does. The indicator we chose to read did not just miss the unmeasured parts of my performance. It reached in and changed them. I had learned what the reader was looking for, and I had started to become that, at the expense of the things the reader could not see.

The KPI, in other words, is an interview that never ends, and the candidate has worked out the questions.


I keep saying that none of this means we should throw the tools out, and I want to ask what that sentence is protecting.

Part of it is true. Interviews, work samples, culture surveys, 360s and KPIs are all better than the alternative, which is an unrecorded impression with no structure and no way to be checked. Some degree of accountability is necessary; people do occasionally waste an organisation's time and money, and a measure that catches it is doing its job. Measurement without firāsa is counting, but firāsa without measurement is projection, and projection with hiring authority is worse than a mediocre test.

But the sentence is also protecting me from a harder question, and the harder question is the one I actually care about. Regardless of what we observe on the surface, how fair is it to infer a person's true capability, their true performance, their real impact on the organisation, from the indicators we have? To what degree are those indicators subjective lenses into a complex world? And how often do we take a decision the indicator was never designed to carry, firing or promoting someone, and rest it on that indicator anyway because it is the number we had?

I do not think that question has one answer. It has a shape. The rigour we bring to a reading should be proportionate to the decision it feeds. A .4 correlation is fine for deciding who to shortlist and terrible for deciding, alone, who to let go. A 360 is useful for a development conversation and dangerous as the sole basis for a demotion. A KPI is a reasonable way to notice that something is drifting and an unreasonable way to conclude what it means. In every case the tool tells you where to look. It does not tell you what you will find when you look, and it does not excuse you from looking.

That, for me, is the link between the word I named my work after and the field I was actually trained in. Organisational behaviour is firāsa at scale, with instruments, under time pressure, with money and livelihoods attached. Everyone at work is being read, by someone, from a surface that does not contain them. And everyone at work is reading, with tools they did not design, about people they cannot see all of.

The indicator is not the person. It is where we agreed to stop looking.

I think about the customers I hurried off the phone. They never appeared on the dashboard. Neither did the colleague I stopped helping. The number I was being read by went up a little, eventually, and somebody somewhere read that as improvement.


This essay was composed in collaboration with AI from Kareem Soliman's account of how firāsa is practised inside organisations and his own experience of being measured. The organisational argument, the examples and the sales story are his; the AI helped structure the piece, checked the claims against the literature, and corrected two of his assumptions along the way, which he has left visible. Kareem retains responsibility for the account and for the decision to publish it.

Continue the enquiry

The Reading Before the Verdict ↗What Would Change My Mind? ↗Putting a Number on It ↗I Can Build Things I Don’t Understand ↗How Sure Are You ↗What the Eyes Don't Tell You ↗