In this guide
Key Takeaways
- →Grades get assigned by seniority, budget and fear of being cut. The question they should answer is narrower: how much of this KPI depends on this priority?
- →A grade is a commitment to a level of scrutiny. Primary means monthly review and a full hypothesis; partial means a named component and an annual look.
- →Framed that way, inflation self-corrects. Many owners choose secondary once they see what primary costs them.
- →Three test questions, asked in order: the cancellation test, the help test, the scope test. If all three are no, leave the cell blank.
- →The KPI owner proposes the grade; the priority owner accepts or contests it through catchball. Record the dissent.
- →Every regrade needs a date and a reason, or the annual reset has nothing to learn from.
A correlation grade isn’t a measure of how much a priority matters. It’s a decision about where the review will spend its attention. Grade by feel and the attention goes everywhere, which means nowhere.
The matrix where everything is strong
Most X-matrices use three grades of correlation, traditionally drawn as a filled circle, an open circle and a triangle. Look at a finished matrix from almost any planning offsite and count the filled circles. They usually outnumber the other two combined.
The grades get assigned for reasons that have little to do with cause and effect. A priority sponsored by a senior leader gets strong correlations because grading it weakly feels like a slight. A priority with a large budget gets strong correlations because a large budget ought to buy a large effect. And a priority whose owner is worried about being cut gets strong correlations because weak ones look like an argument for cutting it.
None of this is cynical. People are answering the question they think is being asked, which is “how important is this work?” The question the grade is supposed to answer is narrower: how much of this KPI’s movement depends on this priority?
When every dot is strong, the review can’t tell which links are load-bearing. It treats them all the same, and the one that’s failing gets the same five minutes as the eleven that are fine.
Grades exist to allocate scrutiny
The useful way to think about a grade is by what it commits the team to. A primary correlation is a link the organization can’t afford to be wrong about, so it should carry the full apparatus of a hypothesis and be checked every month. A partial correlation is worth recording but not worth a monthly conversation.
| Grade | What it commits you to | Reviewed |
|---|---|---|
| Primary | A written mechanism, a contribution estimate, a lag window, a leading signal and a kill signal, agreed by both owners | Monthly |
| Secondary | A one-sentence mechanism and a rough contribution estimate | Quarterly |
| Partial | The component of the KPI it affects, named explicitly | At the annual reset |
| None | Nothing. A blank cell is an honest answer | Not reviewed |
Framed this way, grading inflation becomes self-correcting. An owner who wants a primary grade has to write the hypothesis, name a kill signal and be reviewed against it monthly. Many will choose secondary once they see what primary costs, and the ones who insist on primary are the ones who believe it.
Three questions, one per grade
Each grade has a test question. Ask them in order and stop at the first “yes”.
Primary: the cancellation test
If this priority were cancelled tomorrow, would we expect to miss the KPI target? The question is counterfactual on purpose. It forces the room to imagine the KPI without the priority, which is a different exercise from imagining the priority’s benefits. A priority can be valuable and still fail this test, because two others would carry the KPI anyway.
Secondary: the help test
Would the KPI target survive without this priority, but be harder to reach? Secondary correlations are the supporting cast. They matter in aggregate, and a matrix where every KPI depends on a single primary with no secondary support is fragile. But no single secondary link should be able to sink the target.
Partial: the scope test
Does this priority affect only one component of the KPI, such as one region, product line or customer segment? Partial is often misused as “weak”. It’s more useful as “narrow”. A site-level improvement priority may be decisive for that site’s share of an on-time delivery KPI and irrelevant to the rest. Naming the component turns a vague triangle into something the KPI owner can check.
If all three answers are no, leave the cell blank. Planning teams resist blanks, but a blank is information: it says this priority serves other results, and this KPI depends on other work.
Four rules that keep the grades honest
Questions alone won’t stop inflation in a room full of people with something to protect. Four rules applied to the whole matrix do most of the remaining work.
- Every KPI has at least one primary. A KPI with only secondary and partial links is a result nobody is directly working on.
- Primaries fit inside the gap. The contributions claimed by a KPI’s primary correlations can’t add up to more than the movement its target requires.
- Primaries per priority are few. A priority that claims to be primary for four KPIs is either the whole strategy or an overstated one. One or two is typical.
- Every partial names its component. If the owner can’t say which part of the KPI it touches, it isn’t partial. It’s decoration.
A worked example
Take a KPI that needs to move five points, services share of sales from 31% to 36%. Before grading discipline, four priorities each claim a strong correlation. After applying the questions and the contribution budget, the picture is different.
| Priority | Before | After | Claimed contribution |
|---|---|---|---|
| Regional service center expansion | Strong | Primary | 3 points |
| Service contract pricing redesign | Strong | Primary | 1.5 points |
| Connected equipment telemetry | Strong | Secondary | About 0.5 points |
| Site maturity program | Strong | Partial | Two sites’ local service revenue only |
Illustrative regrading of one KPI’s correlations.
The KPI now has two primaries that fit inside the five-point gap, one secondary that helps, and one partial with a named scope. The monthly review knows exactly which two links to examine, and the telemetry priority is reviewed quarterly against a realistic claim rather than monthly against an inflated one.
Who decides the grade
Grading disputes follow a predictable shape. Priority owners push grades up, because a primary link signals that their work matters. KPI owners sometimes push them up too, for a different reason: more primaries spread accountability for a hard target across more people.
A simple convention resolves most of it. The KPI owner proposes the grade, because they carry the result and have the clearest view of what drives it. The priority owner accepts or contests it through catchball. If they can’t agree, the disagreement itself is useful: it usually means the two have different mechanisms in mind, and that should be settled before the year starts rather than discovered in September.
Record the outcome with the correlation, including any dissent. A grade that was contested and settled is more trustworthy than one that nobody discussed.
Regrading as the evidence comes in
Grades are estimates, so they should change when the evidence does. Two patterns from the year’s data should trigger a regrade.
A primary correlation in outcome lag, where the diagnosis points to a broken mechanism rather than a long lag window or weak adoption, should be downgraded or removed. Leaving it graded primary keeps it consuming monthly attention for a link the evidence no longer supports. (The diagnosis itself is covered in Outcome lag: when priorities are green and KPIs don’t move.)
A KPI that moves well ahead of its graded priorities, an unexpected win, suggests a primary driver that isn’t on the matrix. Find it and add it, even mid-year, or next year’s plan will give credit to the wrong work.
What shouldn’t happen is quiet regrading to make a review look better. Every change to a grade should carry a date and a reason, so that at the annual reset the team can see not only which dots changed but why. That record is the most valuable planning input the next offsite will have.
What grading needs from the software
The rules above survive a planning offsite on their own. What they do not survive is eleven months, because a grade is only useful if it keeps changing the review, and a dated reason only exists if something stored it. Five things decide whether the discipline holds past March.
| What grading needs | Why it matters | How to test it in a demo |
|---|---|---|
| Three grades, not two | Strong and weak collapses partial into secondary and loses the named component | Ask what grades exist, and whether partial can carry a scope |
| The grade drives the review | If primary and partial land in the same list, grading changed nothing | Ask to see a review filtered to primary correlations only |
| A contribution field | Rule two is arithmetic; it needs numbers in a field, not a sentence | Ask where the claimed contribution is entered and totalled |
| A dated strength history | Quiet regrading is the failure mode, and history is the only defense | Ask to see why a grade changed nine months ago |
| Dissent recorded with the grade | A contested grade is more trustworthy, but only if the contest survives | Ask what happens to an objection the KPI owner overruled |
Row two is the one that makes grading worth doing at all.
Row two is the test worth insisting on. A product can let you tag correlations primary, secondary and partial, display them in three colors, and then show every one of them in the same monthly view. Nothing has been allocated. The grade has become decoration, which is where the inflation started. Ask to see what a monthly review actually opens, and whether a partial correlation can be absent from it.
Row four is the one buyers skip and regret. Hoshin kanri software that stores only the current grade makes the annual reset a memory exercise, and the single most valuable planning input, the list of bets that were downgraded and why, does not exist. The seven pass/fail tests cover the broader evaluation; these five are specific to keeping grades honest.
Grading in Profit.co
Profit.co’s X-matrix grades every correlation as primary, secondary or partial contribution, and stores the expected lag on each one. On each objective, the matrix correlations panel lists the KPIs mapped to it with their grade, owner and current value, so a review can go straight to the primary links.
Regrading leaves a trail. Every change to a correlation’s strength is written to its strength history as a PDCA Act entry, and when the change came out of a countermeasure A3, the entry links to it. That is exactly the dated, reasoned record this article asks for, kept without anyone maintaining it. More on the hoshin kanri software overview.
Hoshin Kanri sits in the Essentials plan with OKRs and Goals, Balanced Scorecard and Strategy Roadmap, which matters for rule two: the leading signal behind a contribution estimate is often a team key result, and on one data model its owner’s check-in updates the correlation rather than a number copied from OKR software every month, or from balanced scorecard software when the KPI is reported to the board. The 30-day free trial is long enough to regrade one KPI’s correlations and see whether the monthly view actually changes.
Free Product Demo
How many of your strong correlations are primary?
A 30-minute walkthrough, tailored to your team.
Book a free demo
What happens in the call
How your teams set goals, track progress and review results.
Strategy, performance, projects and engagement on one platform, with AI.
A practical path to align teams, execute faster and drive results.
Questions this article answers
Primary: cancelling the priority would likely mean missing the KPI target. Secondary: it helps, but the target would survive without it. Partial: it affects only one component of the KPI, such as one region or product line.
At least one, so someone is directly working on it, and few enough that their claimed contributions fit inside the movement the target requires.
The KPI owner proposes it, because they carry the result; the priority owner accepts or contests it through catchball. Record the outcome, including any dissent.
Offer three grades rather than two, let the grade drive which correlations the review opens, provide a contribution field that totals, keep a dated strength history, and record dissent alongside the grade.
Because that is what makes inflation self-correcting. If primary means writing a hypothesis and being reviewed monthly, many owners will choose secondary, and the ones who insist on primary believe it.
Yes. If the cancellation, help and scope tests are all no, leave it blank. A blank says this priority serves other results and this KPI depends on other work, which is information.
