goal quality is measurable, and most goals fail the test

Ask ten managers whether a goal is a good one and you will get ten answers. Ask them whether it can be reviewed, and the answers converge. That difference is the whole subject. Goal quality is normally treated as a matter of writing style, something a manager either has a feel for or does not. It behaves much more like a data quality problem, and it can be scored the same way.

The underlying research is unusually settled. Edwin Locke and Gary Latham spent five decades building it. Across the hundreds of studies they synthesised, specific and difficult goals beat vague or easy ones in roughly nine cases out of ten. Effect sizes in the supporting meta-analyses run from about 0.4 to 0.8, which is large for anything in organisational psychology (Locke and Latham, 2002). Their 2019 retrospective sets out the conditions under which this holds. The person must be committed to the goal, must have the ability to reach it, must receive feedback on progress, and must not be carrying conflicting goals.

Feedback is the condition most goal systems quietly drop. It is not an optional addition to goal setting. It is part of the mechanism that makes goals work at all. A goal nobody can review generates no feedback, so it cannot produce the performance effect the theory predicts. That gives a usable definition to work with. Goal quality is the degree to which a goal can be reviewed by someone who was not in the room when it was written.

the reviewability test

Take any goal in your system and hand it to a competent colleague from another function. Ask them to state, without asking a single question, what evidence would show the goal was met and where that evidence would come from. If they cannot, the goal is not reviewable. At year end it will be settled by argument, seniority or memory, and the rating will say more about the relationship than about the work.

The test is deliberately harsh, because the alternative is worse. Kluger and DeNisi meta-analysed 607 effect sizes and found that feedback interventions raised performance on average, but that more than a third of them lowered it (Kluger and DeNisi, 1996). Feedback that cannot point at evidence tends to point at the person instead. That is the variety that does damage.

SMART, which came from a two-page article by George Doran in Management Review, was an attempt at the same idea (Doran, 1981). It has aged badly in practice because four of its five letters can be satisfied by wording alone. A goal can be specific, measurable, achievable, relevant and time-bound on paper and remain unreviewable, because nobody ever agreed where the number would come from.

why cascading fails

The standard response to weak goals is more cascading. Translate strategy into objectives, push the objectives down the hierarchy, measure progress, reward performance. Donald Sull, Rebecca Homkes and Charles Sull surveyed 7,600 managers in 262 companies across 30 industries, a third of them based in emerging markets, and found that the cascade mostly works as designed (Sull, Homkes and Sull, 2015). More than 80 per cent of managers said their goals were limited in number, specific and measurable, and that they had the funds needed to achieve them.

The same survey found that only 55 per cent of middle managers could name even one of their company's top five priorities. Fewer than a third of senior executives' direct reports clearly understood how corporate priorities connected to one another. Among frontline supervisors and team leaders the figure fell to 16 per cent. The machinery ran. The understanding did not arrive.

The reason is structural. Cascading moves goals vertically, and most work moves horizontally. Only 9 per cent of managers said they could rely on colleagues in other functions all of the time. Thirty per cent named failure to coordinate across units as the single greatest challenge to executing strategy, close behind failure to align at 40 per cent. Managers were three times more likely to miss a commitment because another unit did not deliver than because their own team did not.

Sull and his co-authors name the reflex that follows: the alignment trap. Execution stalls, so leaders tighten the cascade, add metrics and add review meetings. Coordination does not improve, because it was never the thing being measured. Ordóñez and colleagues documented the further cost in Academy of Management Perspectives, where narrow goals were shown to crowd out unmeasured work, distort risk taking, inhibit learning and raise unethical behaviour (Ordóñez, Schweitzer, Galinsky and Bazerman, 2009).

The fix is not to abandon cascading. It is to stop treating vertical alignment as evidence of goal quality, and to require every goal to declare what it needs from other teams.

a rubric for scoring goal quality

Here is a rubric you can run this quarter. Score each goal on six dimensions, 0 to 2 on each, for a maximum of 12.

1. Outcome, not activity. Score 2 if the goal names a change in the world. Score 1 if it names a deliverable that stands in for one. Score 0 if it names an activity.
2. Measure with a named source. Score 2 if the goal names both the metric and the system it comes from. Score 1 if it names a metric with no source. Score 0 if it names neither.
3. Baseline. Score 2 if the current value is stated. Score 1 if it is knowable but not stated. Score 0 if no baseline exists.
4. Target and date. Score 2 if both are explicit and the date falls inside the review period. Score 1 if one is missing. Score 0 if both are.
5. Single accountable owner. Score 2 if one named person can move the number. Score 1 if the owner is a group. Score 0 if nobody is named.
6. Declared dependency. Score 2 if the goal names what it needs from other teams, or states explicitly that it needs nothing. Score 1 if a dependency is implied. Score 0 if the goal is silent on the question.

Goals scoring 10 to 12 are reviewable. Goals scoring 6 to 9 need rework before the period starts. Goals below 6 should not be in the system at all, because they will consume review time and produce a rating nobody can defend.

Two rules keep the score honest. Score at the moment of setting, before any outcome is known, so results cannot contaminate the judgement. And score with two independent raters, then check how often they agree. If your raters disagree on more than a fifth of goals, the problem is your definitions rather than the goals.

running it across an organisation

You do not need to score every goal. Draw a stratified sample of 200 to 400 goals across functions, levels and countries. Two raters score each one. The output is a distribution rather than an average, because averages hide the tail, and the tail is where disputed ratings come from.

Report three figures to the executive team. The share of goals that are reviewable. The share that declare a dependency, which is the best available proxy for horizontal coordination. And the functions where those two figures diverge most, because that is where the cascade is producing paper rather than direction.

Then set a target on the score itself. Goal quality behaves like any other measured process. It improves when someone owns it, when the definition is public, and when managers can see their own distribution next to their peers.

The first run is usually uncomfortable. In most organisations the majority of goals fail on baseline and on declared dependency, the two dimensions nobody thinks to check. Those two are also the cheapest to fix, because neither requires a new strategy. One requires a starting number. The other requires an honest sentence about who else has to deliver.

Much of this can be automated. Checks for outcome wording, baselines, named sources and declared dependencies can run at the moment a goal is saved rather than at the end of the year, which is the only point at which the score can still change behaviour. That is the design principle behind entomo's performance intelligence work: inspect the goal before the year runs on it.

Goal quality is not taste. It is a property you can define, score, track and raise. Most organisations have never measured it, which is why they keep redesigning the performance process instead.

see how entomo does this: AI performance management.
in practice: top-5 Indonesian miner, 360° performance.