What the framework covers
The measurement model the plan is built on starts from the business, not the tool. Objectives come from the people who run the business and say what it exists to do; goals are the specific things that must happen for each objective to be met; a KPIA metric tied directly to an objective. is the metric that proves each goal; a target is the value the KPI must reach; and segments are the cuts of the data that explain why a KPI moved. Kaushik's model also asks that objectives be doable, understandable, manageable and beneficial, which in practice rules out the vague ambitions that produce unmeasurable plans.
The framework on this page adds three components the model implies but does not spell out. The tracking specification translates each KPI and segment into events, parameters and dimensions that a developer can implement. The owners name who is accountable for each definition, its implementation and its reporting. The cadence says how often each number is read and when the plan itself is reviewed. Without those three, a plan is a good argument with nobody to carry it out.
What the framework makes visible is orphaned measurement: numbers on a dashboard that trace to no objective, and objectives with no number beneath them. Both are common, and both are found by reading the plan from either end. What it hides is data quality. A plan can be complete and the data beneath it can still be partial, mis-attributed or broken, and the plan will not say so.
This page is the framework. The measurement plan tutorial is the step-by-step method for producing one, the measurement plan strategy guide makes the case for the plan as a strategy document, and the measurement planning hub holds the sequence of artefacts the plan sits among.
The components
| Component | What it records | Test of a good entry | Where it usually goes wrong |
|---|---|---|---|
| Objective | What the business is trying to achieve, in its own words | A leader would recognise it as their goal | Written as a marketing activity rather than a business outcome |
| Goal | The specific thing that must happen for the objective to be met | Could be observed happening or not happening | Several goals folded into one sentence |
| KPI | The metric that proves the goal | One per goal; changes when the goal is met and not otherwise | A metric chosen because it is already in the tool |
| Target | The value the KPI must reach, by when | A baseline, a figure and a date | A target with no baseline, or a baseline nobody can reproduce |
| Segment | The cut that explains movement | Each segment answers a question someone will actually ask | Every available dimension listed |
| Tracking specification | The events, parameters and dimensions that produce each KPI and segment | A developer can implement it without asking a question | Written after launch from whatever was tracked |
| Owner | Who is accountable for the definition, the implementation and the reading | A name, not a team | Shared ownership |
| Cadence | How often each KPI is read, by whom, and when the plan is reviewed | A rhythm tied to the decisions each number informs | Everything reported weekly, nothing decided |
The table reads top to bottom as a chain, and the chain is the point. A tracking specification that does not trace up to a KPI is tracking for its own sake; a KPI that does not trace up to an objective is a report nobody asked for.
From objectives to tracking spec
Write the objectives as the business states them
Take the objectives from the people accountable for the business, in their words, and resist translating them into marketing language. "Make the repair service a profitable second line" is an objective; "grow repair-page traffic" is not. Keep the list short, because each objective generates several goals and each goal a KPI.
Break each objective into goals
For each objective, list the specific things that must happen for it to be met. A goal is observable: a booking is made, a customer returns, a support request is avoided. If a goal cannot be observed, it is either an objective in disguise or a hope.
Attach one KPI and one target to each goal
Choose the metric that moves when the goal is met and not otherwise, record its baseline and the method used to get it, and set the target with a date. The KPI tree framework is the tool for choosing a KPI that can be decomposed later; a KPI with no drivers beneath it cannot be acted on.
Name the segments that will explain movement
For each KPI, write the questions that will be asked when it moves (which channel, which region, new or returning, which device) and keep only the segments that answer them. Segments are where plans bloat.
Translate KPIs and segments into events and dimensions
Each KPI becomes an event or a query over events; each segment becomes a dimension that must be present on those events. This is the hand-off to the tracking plan, which specifies names, parameters, triggers and platforms in the detail a developer needs.
Fix the conventions the data depends on
Record the campaign-tagging convention (UTMURL tags identifying traffic source. parameters and their allowed values), the source of truth for each number when tools disagree, and the AttributionAssigning conversion credit to touchpoints. setting the reports will use. Which attribution models an analytics property offers changes over time, so the plan records the setting chosen and the date rather than assuming a default. In GA4Google's event-based analytics platform. the chosen model and the lookback window are property settings; the plan names them so a later change is a recorded decision.
Worked plan
A measurement plan for an outdoor-gear retailer launching a repair service
An outdoor-equipment retailer with 14 stores and a web shop launches a repair service for jackets and tents. The plan, condensed:
| Objective | Goal | KPI | Target | Segments | Event or source |
|---|---|---|---|---|---|
| Make repairs a profitable second line within 12 months | Customers book repairs online | Repair bookings per week | 150 a week by month 6, from 0 | New vs existing customer; store region; device | repair_booking_complete with item_category and store_id |
| Same | Repair customers buy again | Share of repair customers making a retail purchase within 90 days | 35%, against a 22% baseline for all customers | Repair type; first vs repeat repair | CRM match on email, read monthly |
| Same | The service does not swamp support | Repair enquiry emails per week | Down from 400 at launch to 150 by month 4 | Enquiry topic | Help-desk tag repair-enquiry |
Owners: the head of e-commerce owns the definitions, the analytics lead owns implementation and the weekly reading, a named developer owns the events. Cadence: bookings read weekly, the repeat-purchase share monthly because the 90-day window makes a weekly reading meaningless, support volume weekly, and the whole plan reviewed at the end of each quarter. The attribution setting for the booking KPI is recorded as data-driven with a 30-day lookback, dated at launch.
The condensed plan is already enough to catch the common failure. The second goal cannot be read from the analytics tool at all; it needs a customer-record match, and discovering that at planning time rather than at the first monthly review is most of the value of writing the plan down.
Owners and cadence
Three kinds of ownership keep a plan alive, and they are rarely the same person. The definition owner decides what each KPI means and approves any change to it, because a silent change in definition breaks every trend line beneath it. The implementation owner makes sure the events exist, fire correctly and survive site releases, and runs the checks that prove it. The reading owner pulls the number at its cadence, compares it to target, and brings the segments that explain any movement to whoever decides. When one person holds all three the plan works until that person leaves; when nobody holds one of them the plan drifts within a quarter.
Cadence is set by the decision each number informs, not by habit. A KPI that drives a weekly budget decision is read weekly; one that depends on a long window, such as repeat purchase or retention, is read when the window has closed; and the plan itself is reviewed when an objective changes, a tool changes, or a planning cycle ends. Every change to a definition, target or setting is dated in the plan, because the most frequent question a year later is "when did this number start meaning something different".
Limits
The plan does not fix the data. Consent choices, blocked scripts, cross-device journeys and sampling mean that in practice a share of what happens is never recorded, and a plan that treats the analytics figure as the whole truth will set targets against a partial count. The plan can name the source of truth; it cannot make the source complete.
Attribution is a stance, not a fact. The plan records which model the reports use, and that is the right thing to do, but the choice shapes every channel KPI beneath it. Two plans with different settings will report different results from the same campaigns, and the framework has no way to say which is right.
Objectives handed down without a measurable form stall the chain at the first step. The framework asks the planner to push back until an objective can be broken into observable goals, and in practice that conversation is the hardest part of the work.
A complete plan can also be too long to use. Every objective generates goals, every goal a KPI, every KPI segments and events; a plan for a large business runs to many pages, and a plan nobody reads is not working. The remedy is a one-page summary of objectives, KPIs and targets with the detail held beneath it.
Finally, the plan stops at measurement. It says what moved and, through segments, where; it does not say why or what to do. That is the work of the KPI tree beneath each KPI and of experiments, and a team that expects the plan to supply answers will be disappointed by a document that only supplies good questions.
Related frameworks
| Framework | Reach for it when |
|---|---|
| KPI tree | A KPI needs decomposing into the drivers a team can act on |
| North star metric and OKRs | The objective layer needs a single shared metric and quarterly commitments |
| AARRR | A product business needs its lifecycle stages defined before KPIs can be assigned |
| Attribution models | The attribution stance the plan records needs choosing deliberately |
When the objective is enquiries that close later in a sales system, the measurement plan for lead generation adds conversion depths, a shared lead identifier and offline imports. For mobile products, the measurement plan for app installs covers SDK release timing, ownership across tools, conversion value design and deep-link checks. Campaigns bought to build memory rather than clicks need the measurement plan for awareness, which runs to reach, frequency, lift studies and branded search. Where the outcome happens in a shop, the measurement plan for store visits swaps website conversions for intent proxies, modelled visits, regional sales matching and geo holdouts.
Common mistakes
- Starting from the tool. Listing what the analytics platform already reports and working upwards to objectives produces a plan for the tool, not the business.
- Several KPIs per goal. When three metrics prove one goal, nobody knows which one the goal is judged on.
- Targets without baselines. A target with no recorded starting point and method cannot be graded.
- Every dimension as a segment. Segments that answer no question make the plan unreadable and the reports slow.
- Undated changes. A definition altered quietly breaks every trend and nobody remembers when.
- No implementation owner. Events decay with every release unless someone is named to keep them alive.
- Treating the plan as finished at launch. Sites, apps and tools change; a plan with no review cadence stops describing what is actually measured.
