OKR vs KPI: The Difference That Decides Your Quarter

A director I worked with opened a quarterly review with eleven metrics on one slide. Uptime, ticket volume, cycle time, NPS, cost per transaction, and six more. Every number had a target. Every number was green.

Her VP looked at it for a while and asked: "What are we trying to change?"

Nobody had an answer. The slide showed the business running. It did not show anyone steering it. That is the entire OKR versus KPI question, and most teams get it wrong in exactly that direction.

The short version

• A KPI is a health indicator. It runs continuously, it has a threshold rather than a finish line, and it is somebody's standing responsibility. Uptime is a KPI. It is never "done."
• An OKR is a change commitment. It has a start, an end, and a number you are deliberately trying to move within a window. "Cut median onboarding from 14 days to 5 by December 31" is an OKR.

Most teams publish KPIs and call them OKRs. The result is a quarter of measurement with no direction: everything is tracked, nothing is targeted.

What a KPI is for

A KPI answers "is this still working." It is a standing instrument, and its job is to be boring. You want uptime flat at 99.9%. You do not want a story about uptime. A KPI that moves is usually bad news.

The useful property of a KPI is that it does not care about your quarter. It runs across reorgs, strategy changes, and three generations of leadership. That continuity is exactly what makes it a poor planning instrument and an excellent alarm.

KPIs also protect you from your own OKRs. When a team commits to moving one number hard, the risk is that they move it by breaking something adjacent. The KPI set is what catches it. Cut onboarding to five days by removing a verification step, and the fraud KPI is where that shows up.

Who owns a KPI

A function, permanently. Not a program, not a quarter. If a KPI's owner changes when the quarter changes, it was never a KPI.

What an OKR is for

An OKR answers "what are we changing, by when, and how will we know." The objective is the direction in plain language. The key results are the two to four numbers that would have to move for anyone to believe the objective happened.

The discipline in an OKR is not the format. It is the finish line. An OKR without an end date is a KPI with ambition attached, and it will still be on the slide in eighteen months with a yellow dot next to it.

The second discipline is subtraction. Three OKRs per team is a real commitment. Nine OKRs per team is a list of everything the team was going to do anyway, relabeled. If your OKRs would be identical had you never run the planning session, you did not plan.

Who owns an OKR

One named person, for one window. "Priya Reddy, VP Platform Engineering, through December 31" is an owner. "The platform team" is a hope.

The honest comparison table

The four-question test

When someone hands me a metric and asks which one it is, I ask four things:

1. Is there an end date? No end date, it is a KPI.
2. Is flat a good outcome? If holding steady counts as success, it is a KPI.
3. Would you staff a project to move it? If yes, it is an OKR. If the answer is "someone watches it," it is a KPI.
4. Does it survive a reorg? KPIs do. OKRs should not.

Most metrics that get called OKRs fail all four. They are KPIs with a quarter stapled on.

The most common failure mode

Publishing the KPI set as the OKR set, then wondering why the quarter had no shape.

It happens because it is safe. A KPI you already track has clean data, a known owner, and no risk of missing. Turning it into an "OKR" produces a green slide in week two and a review with nothing to discuss. Nobody has committed to changing anything, so nobody can fail, and the planning session that produced it was theatre.

The tell is a key result that was already going to happen. If the number moves whether or not anyone acts, it is not a key result. It is a forecast.

The second tell is a full green board in month one. Real OKRs are uncomfortable at the midpoint. If yours are not, you set thresholds, not targets.

Practical setup for a quarter

1. Write the KPI set first, and stop calling it planning. Eight to fifteen numbers, function owners, weekly alarm review. This is the instrument panel, and it takes an afternoon.
2. Pick three things to change. Three per team. Write each as a sentence a sponsor would repeat without the slide.
3. Attach two to four key results to each. Each one a number with a start value, a target value, and a date. A key result without a start value cannot be measured, only asserted.
4. Name the guardrail KPIs. For each OKR, name the one or two KPIs that would degrade if you achieved it the wrong way. Put them on the same page.
5. Review monthly, forecast rather than report. The question is not "where are we." It is "where will we land, and what changes that."

The guardrail step is the one most teams skip, and it is the one that keeps OKRs from being expensive. A team that hits its number by damaging an adjacent one has not delivered a quarter. It has moved a cost somewhere the slide does not show.

A worked example: one metric, three framings

Take support ticket volume. Same number, completely different instrument depending on how it is framed.

As a KPI: "Weekly ticket volume stays under 400." Owned by the support lead, permanently. Reviewed every Monday in about ninety seconds. Nobody presents it. If it holds, nothing happens. If it crosses 400 two weeks running, it becomes an agenda item. Good news is a flat line.

As a key result: "Cut weekly ticket volume from 380 to 250 by December 31, by shipping self-serve password reset and the billing FAQ." Owned by a named person, for one quarter. Reviewed monthly as a forecast. Good news is a falling line, and a flat line is now a problem.

The metric did not change. What changed is that the second version names a mechanism, a target, and a date, and somebody's quarter is attached to it.

As the failure mode: "Reduce ticket volume" as the objective, "weekly ticket volume" as the key result, no start value, no target, no date. It is green in week one because 380 is a number and nobody specified which direction counts. It is still on the board in April, yellow, with a note saying it is trending in the right direction.

All three of those are things I have shipped. Only one of them was a plan.

Questions PMs ask

Can a KPI become a key result?

Yes, and this is the healthiest version of the relationship. Take a KPI that has drifted, attach a start value, a target, and a date, and it is this quarter's key result. When the quarter closes it goes back to being a KPI at the new threshold. What it cannot do is be both at the same time.

How many OKRs is too many?

Three per team is the working cap. Five is a stretch. If you have nine, count how many would have happened without the planning session. That number is your real OKR count. The rest is inventory.

What about metrics we cannot influence this quarter?

Those are KPIs, and that is fine. Not everything worth watching is worth changing. The mistake is putting an uninfluenceable number on the OKR board and then explaining every month why it did not move.

Should key results be achievable or aspirational?

Set them so a good quarter lands around 70 to 80%. Hit 100% consistently and your targets are thresholds wearing a costume. Hit 30% consistently and nobody believes the board, at which point it stops driving anything at all.

The bottom line

A KPI is the instrument panel. An OKR is the destination. A program that publishes only KPIs is flying level with no heading. A program that publishes only OKRs is flying toward something with no idea whether the engines are on fire.

If your quarterly slide has eleven numbers and every one is green, you do not have a strong quarter. You have an instrument panel and no heading.

The panel tells you the plane is fine. The heading tells you where it is going. You need the panel. You are paid for the heading.