Employee ethics metrics that predict anything

Nobody can measure whether an employee is ethical. What can be measured is narrower, and most of what appears on ethics dashboards is not it.

This article covers measurement at the level of individuals and teams, where the temptation to score people lives and where the damage from doing it badly lands. It says what can be observed honestly, what cannot be observed at all, and what a team-level set looks like when it is not lying to anyone.

What to take away

What is actually observable

What someone did, in a situation that was seen. A person was asked to do something questionable and asked a question first, or did not. A person found a problem and escalated it, or sat on it. A person was told to be quiet and was not. These are events, they have dates, and they can be recorded.

What someone asked. Questions to the named ethics contact, requests for clarification of a rule, requests for exceptions. A person who asks is a person who noticed.

What someone reported, and what happened to them afterwards. Reports are events. So is what the reporter's career looked like twelve months later, compared with peers. In some countries what happens next is also a legal question rather than only a management one, as the UK's whistleblowing framework shows for one jurisdiction.

What a person says they would do. Give someone a specific hypothetical and ask, in writing, what they would do. The answer is not proof of conduct, but it reveals whether they know the route, and it cannot be gamed by clicking a box.

That is the list. It is short, and everything on it is a situation with a person in it. The same preference for evidence that something operates, over evidence that it exists, runs through the Department of Justice's guidance on evaluating compliance programs.

What is not observable

Intent. Whether the person who declared the conflict did so because they are honest or because they expected to be caught is unknowable and does not matter. Measure the declaration.

Private conduct. What people do outside work, with a handful of exceptions that are legally governed and role-specific, is not the organization's to measure. Attempts to do so are a different problem, and usually a legal one.

What did not happen. The corner not cut, the customer not misled, the number not adjusted. The whole point of good conduct is that it leaves no trace. A metric that counts incidents is counting the failures of the people who chose badly and is silent on the people who chose well, and there is no fix for that.

Anything from an attestation. A signed acknowledgment that a policy has been read measures that a reminder was sent and a link was clicked. The reasons completion rates carry no information are covered under ethics training programs and are not repeated here.

Why individual ethics scores go wrong

Some organizations put an "integrity" or "values" rating into performance reviews. The intention is good. The effect, within one cycle, is predictable.

The manager gives the rating with no observable basis beyond the events above. Most people in most years are not tested, so the manager mostly has no events. The rating then defaults to the manager's impression.

That impression correlates with likeability and with not having caused the manager difficulty. A person who raised a problem caused difficulty, and their rating reflects it.

Now the rating is known to affect pay. People learn what it rewards: not being the person who raised something. The metric has become an incentive against the one observable behavior that mattered.

If you have such a rating, remove it or replace it with a factual record: did this person raise anything, ask anything, decline anything this year, and what happened. A record of events cannot be inflated by charm and does not punish the reporter.

Measure the system instead

Almost every honest metric in this area measures the organization's machinery, not its people. The machinery is what determines whether the observable events happen at all.

Each row's third column is the honest limit. A report that presents the second column without the third is presenting a number as more than it is.

The honest team-level report

For a team, once or twice a year, the report that carries real information looks like this and fits on a page.

Three situations. Anonymized, specific, real. What came up, who noticed, what they did, what it cost them, what the team did about it. If there were no situations, say so, and say whether that is because nothing came up or because nothing was noticed. Be honest about not knowing which.

Questions and exceptions. How many questions the team asked of the ethics contact or the code's named person, and how many exceptions were requested and granted. Not as a score. As a description: a team that asked nothing and requested nothing is not necessarily a clean team; it may be a team that has stopped engaging.

What happened to anyone who raised something. Named to the reader with the authority to know, otherwise described. This is the item that determines whether next year's report has any situations in it.

One thing the team got wrong. A report with no such item is not believed by anyone who has run a team.

What is unknown. Explicitly. The corners not cut are not in the report because they cannot be. The private conduct is not in the report because it should not be. Saying so is what makes the rest credible.

When a number moves

If a team's figures change, run the same order as any conduct number. Did the definition change, did the team change, did visibility change, and only then did behavior change?

A team whose report count doubled after a new manager encouraged questions has not become twice as troubled.

A team whose exception requests fell to zero after the exception owner became hard to reach has not become compliant.

These interpretive traps are the same at every level. The organization-wide version, including how to read figures in pairs so that each corrects the other, is covered in the site's article on measuring whether frameworks are used, and the board-level reporting problem above that is in corporate responsibility. This article is the individual and team layer under both.

What to do on Monday

Find out whether your organization rates individuals on ethics in reviews. If it does, find out who raised something in the last two years and look at their ratings. That single check tells you whether the rating is measuring conduct or measuring convenience.

Then write the one-page report above for your own team, honestly, including the last section. If you cannot fill in the three situations, you have learned the exercise's most important lesson: you are not seeing what your team is dealing with.

The person deciding whether to tell you weighs the costs described in employee ethics. Making that decision easier is in your hands, and it is in leadership ethics, not in a dashboard.

Common questions

Is there any individual ethics measure that works?

A factual record of events: raised, asked, declined, and what followed. It is not a score and cannot be ranked, which is why it works.

How do I present a report with no situations in it?

Say there were none, and say whether you believe that is because nothing happened or because nothing reached you. The second is the more likely answer for most teams, and admitting it is the beginning of fixing it.

What should I do with a dashboard that already exists?

Add the third column: what each figure does not tell you. If the dashboard cannot accommodate that, it is the dashboard that is wrong.