Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Analytics › Measurement Planning

KPI Trees

How to build a KPI tree from a business outcome down to drivers teams can move: the arithmetic between levels, inputs versus outputs, and finding the lever.
Guideintermediate11 min readUpdated

What a KPI tree is

A KPI tree is a hierarchy with one outcome at the top and a chain of metrics beneath it, where every parent is a stated function of its children. The function is usually multiplication (orders equal sessions multiplied by conversion rate), sometimes addition (revenue equals new-customer revenue plus returning revenue), occasionally subtraction (net revenue equals gross revenue minus refunds). The rule that makes it a tree rather than a list is that the children fully explain the parent: if sessions and conversion rate both hold steady, orders cannot have moved.

That property is what distinguishes a tree from a scorecard. A scorecard lists metrics the team cares about; a tree says how they combine. When the outcome moves, the tree says which branch to look in and, by following the arithmetic down, which leaf. When a team proposes an initiative, the tree says which leaf it touches and, by following the arithmetic up, how much the outcome could move if the initiative works. Both directions — diagnosis down, planning up — depend on the arithmetic being explicit.

Choosing which outcome sits at the top is the job of the objectives and KPIs guide; the north-star-and-OKRs framing makes the same choice at organisation level. This page assumes the outcome has been chosen and shows how to decompose it. The KPI framework reference holds the standard metric definitions the leaves should align to.

The arithmetic between levels

Each level of the tree is connected to the next by a formula, and the formula must balance. Write it beside the parent node. The commonest forms in marketing:

RelationshipFormTypical use
Rate decompositionCount equals a larger count multiplied by a rateOrders from sessions and conversion rate; leads from form views and completion rate
Value decompositionValue equals a count multiplied by a value per unitRevenue from orders and average order value; pipeline from leads and average deal size
Additive splitTotal equals the sum of segmentsRevenue by channel, by market, by new versus returning customers
Funnel chainFinal count equals the first count multiplied by each stage rate in turnVisits to product views to add-to-basket to checkout to purchase
Cost relationshipCost per result equals spend divided by resultsCost per lead; acquisition cost (CAC) from spend and new customers
Lifetime relationshipValue per customer equals value per period multiplied by retention periodsLifetime value (LTV) from average margin and expected tenure

Two rules keep the arithmetic honest. First, every child of a multiplicative node must be measured on the same population and period as its siblings: a conversion rate computed on all sessions cannot sit beside a session count that excludes bots. Second, an additive split must be exhaustive and non-overlapping, or the parts will not sum to the whole and the tree loses its explanatory power. Where a segment is partially measured — offline orders, say — add an explicit "unattributed" child rather than letting the gap hide in the arithmetic. In a funnel chain, each stage rate, such as add-to-cart rate, must be computed on the count of the stage before it, or the product of the rates will not reproduce the final count.

Illustrative example — not a real result

An online retailer's revenue tree, with the numbers that balance

Revenue = Orders × Average order value

Orders = Sessions × Conversion rate

In March the retailer records 420,000 sessions, a conversion rate of 2.1% and an average order value of £58, giving 8,820 orders and £511,560 in revenue. In April, revenue falls to £469,800. Sessions are 435,000 (up), average order value is £60 (up), so the orders node must explain the fall: 7,830 orders implies a conversion rate of 1.8%. One level further down, the split by device, which is 40% desktop and 60% mobile in both months, shows desktop conversion rate unchanged at 3.0% and mobile down from 1.5% to 1.0%. The blended rate follows from the segments and their shares: 0.4 × 3.0% + 0.6 × 1.5% gives March's 2.1%, and 0.4 × 3.0% + 0.6 × 1.0% gives April's 1.8%. A funnel chain beneath mobile shows the add-to-basket rate flat and the checkout completion rate down from 48% to 32%. Because the device mix did not move, there is no second leaf to explain, and the tree has taken a £41,760 revenue shortfall to a single leaf — mobile checkout completion — in four steps, each a formula that balances. Had the mix shifted towards mobile, that shift would be a second leaf to report beside the mobile rate. Running the mobile checkout numbers through the conversion rate calculator confirms the arithmetic before anyone opens a ticket.

Inputs, outputs and guardrails

Not every node in a tree is the same kind of thing, and labelling the kinds prevents the most common misuse.

Outputs are the metrics that compose the outcome: orders, average order value, leads, pipeline. They are what the business wants to move, and nobody can move them directly — they are the result of the drivers beneath them. Reporting an output without its drivers tells the team what happened and nothing about why.

Inputs, or drivers, are the leaves: the metrics a named team can change through its own work. Paid media can change clicks and the cost per click; the content team can change organic sessions to a set of pages; the product team can change checkout completion. A leaf that no team owns is not a driver; it is an output that needs decomposing further, or an external factor to note and leave alone.

Guardrails are metrics that are not in the arithmetic but constrain it. Refund rate, unsubscribe rate, lead quality as judged by sales, customer acquisition cost against lifetime value, page speed. A tree that optimises orders without a refund guardrail will happily report growth that the finance team later reverses. Guardrails sit beside the tree with a threshold and an owner, and a driver initiative that breaches one is not a success however far it moved the outcome. The same discipline applies when a team adopts AI tools: the guide to measuring AI's impact on team output connects tasks such as product description drafts to the outcome drivers they feed, with defect and rework rates as guardrails.

Diagnostics are the measurements beneath a driver that explain its movement without being levers in themselves: the share of sessions on each device, the error rate on a form, the position of a keyword. They are read when a driver moves, and the interactive builder further down this page shows them as the fourth level.

A practical check on each leaf: can you name the person who would be asked to move it, and the action they would take? If not, it is not yet an input.

Building a tree for a revenue goal and a lead goal

The two commonest marketing outcomes decompose differently. Both trees below stop at the point where a named team owns the leaf; a diagnostic level beneath each leaf is implied.

A revenue goal for a transactional site:

LevelNodeFormula or definitionOwner of the leaf
OutcomeRevenueOrders multiplied by average order value (AOV)—
OutputOrdersSessions multiplied by conversion rate (CVR)—
DriverSessions by channelAdditive split: paid search, paid social, organic, email, direct, referralChannel managers
DriverConversion rate by device and landing page groupRate per segment; additive weighting gives the blended rateConversion and product teams
OutputAverage order valueRevenue divided by orders—
DriverItems per orderBasket composition, cross-sell placementMerchandising
DriverAverage item priceProduct mix, promotion depthMerchandising and pricing
GuardrailRefund rate; acquisition cost against lifetime valueMonitored, not decomposedFinance with marketing

A lead goal for a service or business-to-business company:

LevelNodeFormula or definitionOwner of the leaf
OutcomeSales-qualified leadsMarketing-qualified leads multiplied by qualification rate—
OutputMarketing-qualified leadsForm submissions multiplied by the share that meet the qualification rule—
DriverSessions to lead pages by channelAdditive split by channelChannel managers
DriverLead page conversion rateSubmissions divided by sessions to lead pagesConversion and content teams
DriverQualification rule pass rateShare of submissions matching firmographic or intent criteriaMarketing operations with sales
OutputQualification rateSales-qualified divided by marketing-qualified—
DriverSpeed to first contact; lead routing accuracyOperational process metricsSales operations
GuardrailCost per sales-qualified lead; sales-reported lead qualityMonitored, not decomposedMarketing and sales leads

The lead tree has a feature the revenue tree lacks: its outcome depends on another department's judgement. The qualification rate is owned jointly, and the tree should say so, because a marketing team judged on sales-qualified leads without influence over qualification will optimise the wrong leaf. The SEO KPIs and reporting guide shows the same decomposition applied to one channel, where the leaves become impressions, click-through and indexed pages. A business with recurring revenue needs a third shape, with first payments and churn beneath the revenue a cohort retains, which the guide to measurement for subscription businesses lays out with the systems that feed each level.

A lead goal raises its own questions about ownership and timing, which the KPI tree for lead generation answers with a pipeline root, source-split stage rates and named sales-side leaves.

The builder below shows a tree of each kind as four levels — objective, outcome, driver and diagnostic — with a note on each node saying how it is measured. Select a node to read its measurement and the drivers beneath it; use it to check that your own tree reaches a leaf a team can act on.

KPI tree builder

Reading a tree: finding the lever

A tree is read in two directions. Downwards, for diagnosis: the outcome moved, so which child moved, and which of its children, until the leaf. Upwards, for planning: a team proposes to move a leaf, so how much does the outcome move if they succeed. Both readings follow the same discipline.

Hold the siblings constant. At each level, ask which child changed and which did not. A multiplicative node with one child moving and the rest flat has one explanation; several children moving at once means the change has several causes, and each is followed separately. Resist the temptation to stop at the first child that moved.

Weigh the branches. Not every leaf has equal pull. In a funnel chain, a stage with a low rate and a large population has more room than a stage already near its ceiling; in an additive split, the largest segment moves the total most for a given rate change. The planning reading of the tree is a comparison of how much the outcome moves per unit of effort on each leaf, and that comparison is what the tree exists to make possible. A tree reads one period at a time; the tutorial on building a growth model in a spreadsheet adds the time dimension, carrying each month's new customers through activation and retention to compare leaves by their effect on the active base a year out.

Check the guardrails before celebrating. A driver that moved the outcome and breached a guardrail — orders up, refunds up faster — has not delivered, and the tree should show both in the same view.

Distinguish a lever from a symptom. Conversion rate fell because the mix of traffic shifted towards a colder channel: the leaf that moved is conversion rate, but the lever is the channel mix one branch over. Trees expose this when the segments are defined consistently across branches, which is the reason the measurement plan tutorial asks for dimensions to be defined before metrics are mapped to platforms. The guide to explaining a change in a marketing metric works through that mix and rate split step by step, then follows the change into segments and events and tests the explanation.

The reading improves with practice and with a habit: every time the outcome is reported, report the branch that explains it. The measurement planning hub places the tree as the second artefact in the sequence precisely because every later document — events, dimensions, checks — exists to make the tree readable. A written tree that leadership recognises is also one of the two gates in the marketing measurement audit scorecard, which rates whether budget decisions rest on evidence that traces back to it.

Common mistakes

  • Leaves nobody owns. A tree whose bottom level is still outputs explains nothing; decompose until a team can act.
  • Arithmetic that does not balance. Children measured on different populations or periods; additive splits that overlap or leave gaps. Check the formulas with real numbers before publishing the tree.
  • Mixing rates and counts in an additive split. Segments of a count add; segments of a rate must be weighted by population.
  • No guardrails. Optimising a single outcome without constraints produces growth the business later reverses.
  • Too many levels. Decomposing past the point of action turns the tree into a glossary.
  • One tree for several outcomes. Each outcome gets its own tree; shared drivers are noted, not merged.
  • Confusing the leaf that moved with the lever. A rate can fall because of a mix change in a neighbouring branch; read sideways as well as down.

Frequently asked questions

How many levels should a KPI tree have?

Enough to reach a metric that a named team can move directly, and no more. Most marketing trees reach that point in three or four levels: the outcome, the outputs that compose it, the drivers behind each output, and sometimes a diagnostic level beneath a driver. A tree that keeps decomposing past the point of action becomes a glossary rather than a tool.

What is the difference between a KPI tree and a dashboard?

A dashboard shows metrics side by side; a tree shows how they relate. A dashboard can tell you that conversion rate fell and traffic rose in the same week; the tree tells you which of the two explains the change in orders, because the arithmetic between levels is explicit. Good dashboards are often laid out to follow a tree, which is the strongest argument for building the tree first.

Can a KPI tree have more than one outcome at the top?

It can, but each outcome should have its own tree, because the branches beneath a revenue goal and a lead goal are different even when they share drivers like traffic. Where an organisation has a single north-star metric, that sits at the top and the business objectives become the first level beneath it; where it has several objectives, build one tree per objective and note the drivers they share.

Terms used on this page

Tools to try

Frameworks

Where this page is used

Part of these learning paths

Also related

Related