Skip to content

Tracking Plan

beta  ·  Family governance-docs  ·  Phase undefined  ·  Sizes lean, full  ·  ~4,250 tokens

The standing catalog of the analytics events a product records, the properties and traits each one carries, and why each is collected, where it comes from, who owns it, and whether it is still live. Distinct from a single feature’s own PRD instrumentation slice, the data dictionary, the broader measurement plan, and the KPI dashboard it stands upstream of.

Fast reference for using the tracking-plan bundle. For the full reasoning, origins, and sources, read tracking-plan_companion.md.

  • The product already records, or is about to record, more than a handful of events, and nobody has one place that lists what those events are, why each one is collected, and who answers for it.
  • More than one person will add, change, or simply go looking for an event over the product’s life, not just during one feature’s build. A tracking plan is the standing catalog that outlives any single launch.
  • A feature’s own PRD has already proposed events in its “Analytics and instrumentation” section, and that feature is shipping. This plan is what every feature’s own slice merges into once it ships, a reading this library states as its own position rather than any source’s claim (see tracking-plan_companion.md, section 8).
  • Someone, eventually, will ask what an event’s properties are, where it comes from, or whether it is still live, and the honest answer today is “ask the person who built it,” which is the owner-departure failure this bundle’s research names as its most fully sourced failure mode (section 7 of the companion).
  • No event has been proposed yet, and the team is choosing between autocapture and a hand-built plan. Exactly one source in this bundle’s research argues from the autocapture side, and even it does not call a plan unnecessary: it names the choice itself “a false choice” and still advises a team to manually capture a small, named set of core events to start. Route a team with nothing yet to that short, hand-picked start, not to a full catalog it cannot yet fill in honestly.
  • A single feature is proposing its own launch metrics. That belongs in that feature’s own PRD, in its “Analytics and instrumentation” section, first. This plan is what the feature’s slice merges into once it ships, not the place a feature’s launch is first drafted.
  • The question is what the data means once it lands in the warehouse. That is the data dictionary’s job, stated in the opposite tense from this plan’s forward-looking “what you intend to track.”
  • The question is why the business tracks a number at all, or what the reporting cadence is. That is the broader measurement plan, which contains a tracking plan but also covers business objectives, KPI definitions, and reporting logic this plan does not carry.
  • The question is a metric’s own formula or target. That is the KPI dashboard’s Metric Definitions section. This plan defines the events a metric is computed from; it does not lock the formula itself.
  • The question is a risk, an open item, or a problem that has already happened. Those belong in the risk register, the RAID log, or the issue log. A tracking-plan failure, such as an event that silently stops firing, becomes a row in one of those once someone notices it; it is not a second home for the problem inside this plan’s own catalog.

Tracking plan, or one of its three neighbors?

Section titled “Tracking plan, or one of its three neighbors?”
Tracking plan Data dictionary Measurement plan KPI dashboard
Tense Forward-looking: what you intend to track Backward-looking: what the data means once implemented Forward-looking, broader Defines a metric’s own formula and target
Scope Events, properties, traits Implemented warehouse structure Business objectives, KPIs, reporting logic, and tracking One metric at a time
Relationship to this plan N/A Fed by this plan once implemented Contains this plan as one part Consumes the events this plan defines

A single feature’s own launch metrics are not a fourth neighbor on this table: they are this plan’s own future content, first drafted in that feature’s PRD and merged here once it ships.

  • Lean (default): Purpose and Boundary, Naming Conventions, Event and Property Catalog, User and Account Properties, Review and Ownership. A complete, working plan for a product starting out or a small event set, following the research’s own framing that a team can start with a document when the event set is small, and the advice not to try to track everything at once.
  • Full: adds Privacy, Sensitivity and Retention (which event or property might carry personal data, how long it is kept, and what happens to it before storage) and Change Process and Version History (how a change to the standing plan is proposed and approved, how an event is formally removed, and the running record of what changed and who approved it). Move to full once more than one team changes this plan, once it holds personal or sensitive data, or once someone will later need to ask what changed and who approved it.

Grow lean into full by inserting Privacy, Sensitivity and Retention as the fifth section, before Review and Ownership, and appending Change Process and Version History last; the first four keep their name, order, and tables. The scaling signal is whether the plan holds sensitive data or needs a defined change procedure, not how many events it lists.

Score each row 0, 1, or 2. Below 11 out of 16, the first person who tries to add an event, trace why one exists, or find out who to ask about it will fail to do so from the plan alone, and will end up asking around instead of reading the document. A lean plan carries no privacy or change-process section to trace, so it is scored, and cleared, against a smaller table; see the scope table below.

# Criterion 0 1 2
1 Boundary rules out neighbors No line says what this plan does not cover A boundary line exists, but it names no specific neighboring document You can point at the sentence naming at least one neighbor by name (a feature’s own PRD slice, the data dictionary, the measurement plan, or the KPI dashboard) and at what it hands that neighbor instead of carrying it here
2 Naming convention stated No tense, case, or cross-platform answer is given anywhere A tense and case are named, but nothing says what happens when a platform’s own automatic tracking would otherwise produce a different name for the same action You can point at the stated tense and case, and at the specific fallback for a platform whose automatic tracking would otherwise diverge from it
3 Every row answers “why” Rows exist with the reason cell blank, or filled with the event name restated A reason is given for most rows, but at least one event’s business question cannot be stated back by a non-analyst You can point at any row and repeat its business question back in one sentence, because the reason cell answers it rather than restating the event name
4 Owner and status named An event has no owner, or “owner” is a team name rather than a person Owners and statuses are present, but the status values are invented per row rather than drawn from one stated vocabulary You can point at any row’s owner (a person) and its status, and the status comes from the vocabulary stated once at the top of the table
5 Traits kept apart A persistent fact about a user or account is listed as though it were a property of one event The User and Account Properties table exists, but at least one row leaves the subject (user or account) or the source blank Every trait names its subject, its type and allowed values, and where it is set, and no persistent fact has leaked into the event catalog above
6 Sensitivity at property level (full) No property is flagged, or a whole event is flagged with no property named A property is flagged, but the retention period or the handling step before storage is missing You can point at a specific flagged property, its retention period, and its handling step, and nothing here is presented as settled legal advice
7 Change roles named (full) No one is named as able to propose or approve a change, or “the team decides” is the only statement A proposer and an approver are named, but nothing says what a reviewer checks before a change ships You can point at the named role that proposes a change, the named role that approves it, and what a reviewer checks before it ships
8 Removal is a procedure (full) An event’s status flips to removed with no record of when or why A version history exists, but at least one row names no event, no approver, or no date You can point at a removed event’s own row in the version history, naming the event, who approved the removal, and when, rather than a quiet status flip with nothing recorded
Variant Rows scored Maximum Threshold
lean 1-5 10 7
full 1-8 16 11

Lean ships no Privacy, Sensitivity and Retention or Change Process and Version History section, so rows 6 through 8 grade content it does not carry; score a lean plan against rows 1 through 5 only. A lean plan at 7 out of 10 clears a comparable bar to a full plan at 11 out of 16.

  1. Owner departure. The most fully sourced failure mode in this bundle’s research: an event with no accountable person decays once its author leaves, and nobody left can explain what it means or whether it still fires correctly. Fix: a named person, not a team, behind every event, recorded in the catalog’s owner column from the day the event ships.
  2. Governance treated as a one-time setup. Writing the plan once and never reopening it is named directly in this bundle’s research as a primary reason governance programs fail; a plan that is set up, documented, and left alone decays the same way an unowned event does. Fix: a stated review interval in the Review and Ownership section, actually kept, not written once and forgotten.
  3. Unclear, team-level ownership. “The analytics team owns it” is named in the research as functionally the same as no owner at all, because no single person can be asked when something breaks. Fix: one named person per event, full stop, not a team name in the owner column.
  4. No gate before an event ships. Where no one checks a new or changed event against what the plan says before it reaches production, the plan and the live data drift apart silently. Fix: a named validation step in the Change Process section, run before an event ships, not only after someone notices it is wrong.
  5. A status flip standing in for a removal procedure. Marking an event “removed” with no record of when or who approved it erases exactly the information a later reader needs to know whether the removal was deliberate or accidental. Fix: a removal procedure with its own steps, and a version history row naming the event, the approver, and the date, every time.
  6. A naming convention with no cross-platform answer. A convention that states a tense and a case but says nothing about what happens when a platform’s own automatic screen-tracking produces a different name for the same user action lets the same action quietly become two different events. Fix: name the fallback in the Naming Conventions section itself, not as a decision made silently per platform later.
  7. The plan outgrowing a spreadsheet, or outgrowing its own boundary with a neighbor. A plan that has scaled past what a spreadsheet can hold reliably is one named, sourced failure shape in this bundle’s research. A plan that has quietly started carrying a feature’s own launch metrics, a metric’s formula, or a risk is a second, in this library’s own reading of where those contents belong; no source names it. Fix: move to a validated tool once the spreadsheet itself becomes the risk, and route each neighbor’s own content back to its own document rather than letting this plan absorb it.

This bundle’s family contract states plainly that no governance skill exists in pm-skills today for any of its six members, so every member adopts pairs_with: [] until one does. Until a governance-side skill exists, this template is filled by hand.

tracking-plan_template-lean.md · ~4,250 tokens

---
title: "{{product_name}} Tracking Plan"
product: "{{product_name}}"
plan_owner: "{{plan_owner}}"
data_steward: "{{data_steward}}"
sources_and_destinations: "{{sources_and_destinations}}"
plan_version: "{{plan_version}}"
last_reviewed: "{{date}}"
status: "{{status}}"
doc_type: tracking-plan
size: lean
source_template: tracking-plan
source_template_version: 0.1.0
---
<!--
LEAN TRACKING PLAN. The smallest tracking plan that is still a real one: what it covers and the four
neighboring documents it rules out, the naming convention its catalog follows, the load-bearing event and
property catalog itself, the user and account traits kept apart from any single event, and the plan's own
named owner with a stated review interval. Use it for a product starting out, or for a small event set,
following the practitioner framing that "Teams can start with a spreadsheet or document when the event set
is small" and the advice to "Don't try to track everything at once. Build your tracking in phases." To grow
it into the full form (see tracking-plan_template-full.md), ADD sections; never rename or reorder the ones
below, because the full variant is a strict superset of this one.
WHAT A TRACKING PLAN IS. Amplitude puts the job in one sentence: "A tracking plan defines every event and
property you collect, why you collect each one, and which source emits it." Countly adds what the document
does across teams: "It translates product questions into an implementation contract for product,
engineering, analytics, privacy, and marketing teams." The sources that define the type are mostly vendors,
and most of them sell the tool the plan is later enforced in, so this template treats their shared
definition as the convergent core and does not present any one vendor's case for its own product as a
finding. See tracking-plan_companion.md section 1.
NO SOURCE READ FOR THIS BUNDLE MEASURED WHETHER KEEPING ONE IMPROVES DATA QUALITY OR DECISIONS. What exists
is practitioner testimony about what goes wrong without one: a named failure taxonomy and one
practitioner's own claim that an ungoverned plan is "a document that's wrong within 60 days of shipping."
This template's reason to exist rests on that testimony and on the failure modes it names, not on a
measurement. See tracking-plan_companion.md section 1 and section 7.
IT IS NOT A SINGLE FEATURE'S INSTRUMENTATION SLICE, A DATA DICTIONARY, A MEASUREMENT PLAN, OR THE ENFORCED
SCHEMA IN A TOOL. A single feature's own events are first proposed in that feature's own PRD, in its
"Analytics and instrumentation" section; this plan is what every feature's slice merges into over the
product's life, a reading this library states as its own position rather than as any source's claim. A
data dictionary "documents what the data means and how it is structured once implemented," the opposite
tense from this plan's forward-looking "what you intend to track." A measurement plan contains this plan
but also covers business objectives, KPI definitions, and reporting logic. And several vendors keep a live
copy of this plan inside their own product and check incoming data against it; this template is the
document, not that enforced copy. See tracking-plan_companion.md section 1, section 3 (Purpose and Boundary), and section 8.
HOW TO FILL THIS IN
1. Read the comment under each heading: WHAT, WHY (with a companion pointer), ASK, GOOD, WEAK, TRAP; the
two table sections also carry PRIORITY and ROW HINT.
2. Replace each {{placeholder}}. The Event and Property Catalog is the heart; no event is listed without a
named owner and a stated reason, and the naming convention, the status vocabulary, and the catalog's
columns stay consistent with each other.
3. If a section does not apply, write "N/A" and one line of why, rather than deleting it.
4. This document is never finished, but before you share it: self-grade against tracking-plan_guide.md,
then DELETE every HTML comment. They are guidance, not content.
-->
# {{product_name}} Tracking Plan
## Purpose and Boundary
<!-- WHAT A short statement of what this plan covers, product-wide, and what does not belong in it.
WHY The name sits beside four neighbors a reader will meet under similar words, and the first job
of this section is to rule them out before the catalog starts. Wizbrand's own account draws the
plan forward-looking: "A tracking plan describes what you intend to track (events, triggers,
parameters) and why," against the opposite tense for a data dictionary, "A Data Dictionary
documents what the data means and how it is structured once implemented." Against a broader
measurement plan, the same source states containment directly: "the measurement plan explains
why you track; the tracking plan explains how." This library reads this plan as what every
feature's own PRD instrumentation slice merges into over the product's life, and reads the KPI
dashboard's own metric formula as a separate concern this plan does not lock, each stated here
as this library's own position rather than any source's claim. Deep dive:
tracking-plan_companion.md section 3 (Purpose and Boundary) and section 8.
ASK What product, or which part of it, does this plan cover? What four neighbors does it rule out,
by name: a single feature's own instrumentation slice, a data dictionary, a broader measurement
plan, and a tool's own enforced schema? Where does a single feature's own instrumentation
request go before it merges into this standing plan?
GOOD "This plan covers every event Shiftwell's worker app and manager console record. A new
feature's events are proposed in that feature's own PRD and land here when it ships. What the
warehouse tables mean once these events reach them is the data dictionary's job. The schema
Shiftwell's analytics tool enforces is generated from this plan and is never edited ahead of
it."
WEAK "This plan tracks what we track." (names no neighbor it rules out, and a reader cannot tell a
single feature's own slice from the standing plan)
TRAP Treating this plan as the place to write a single feature's launch metrics, or trying to lock a
KPI's own formula here instead of in the KPI dashboard's own Metric Definitions section. -->
{{purpose_and_boundary}}
## Naming Conventions
<!-- WHAT A short statement of the convention this catalog's event names follow: tense, case, and whether
the same name holds up across every platform this product ships on.
WHY Countly names the obligation generally: "A strong plan defines naming conventions," among other fields. The
object-then-past-tense-action form is attributed to the practitioners who state it, not to any
vendor's own specification: a consultancy's engineering handbook reports Segment's own
object-action framework and states the rule, "Action: defines how users can interact with the
objects in the application. An action must be in the past." An independent newsletter states
the same preference in its own voice, "I recommend using the object+action framework because
it's easy to visualise what object your user is interacting with and what action has taken
place," and explicitly declines to settle casing, "Choose one case and stick with it," while the
same engineering handbook requires snake_case outright. One source names the exact failure a
plan without a cross-platform answer invites: "the screen names differ per platform, which
makes it hard to actually use the data." This section asks for a convention and picks none of
these forms for you. Deep dive: tracking-plan_companion.md section 3 (Naming Conventions) and
section 6, item 3.
ASK What tense do event names follow? What capitalization and casing? Does the same event name hold
up across every platform this product ships on, and if not, what is the fallback when a
platform's own automatic tracking would otherwise diverge?
GOOD "Shiftwell event names are object-then-past-tense-action, snake_case: shift_claimed,
shift_released, worker_onboarded. The same name is used on iOS, Android, and web; where a
platform's own automatic screen-tracking would otherwise produce a different name for the same
action, that automatic event is excluded from this catalog and the matching manual event is
used instead."
WEAK "Name events clearly." (no tense, no case, and no answer for what happens when two platforms
disagree)
TRAP Naming a convention without saying what happens when a platform's own automatic tracking, such
as a default screen-view event, would otherwise produce a different name for the same user
action. -->
{{naming_convention}}
## Event and Property Catalog
<!-- WHAT The load-bearing table, one row per event: its name, when it fires, why it is collected, its
properties with type and allowed values, where it comes from and where it goes, its owner, and
its status.
WHY Amplitude's own one-sentence definition names every column this table carries: "A tracking plan
defines every event and property you collect, why you collect each one, and which source emits
it." Countly's field list adds triggering conditions and sensitivity to the same spine. When it
fires is a worked example's own LOCATION column, "the location and time at which the event is
being tracked." Why it is collected is the same worked example's WHY column, "a short
description of the business purpose that justifies tracking the event," and Avo's own
instruction to "Summarize the reason for tracking each event and property." Properties, with
type and allowed values, follow Avo's "Property value types" and Twilio Segment's "Permitted
Values" fields. The owner column is convergent, not single-sourced: Amplitude's blog ties the
source field directly to an owner, "the person responsible for the implementation of that
event," and Amplitude's own event-taxonomy page asks for "central documentation listing every
event, property, data type, and owner." Status is likewise convergent: Amplitude's taxonomy page
asks teams to "Track all taxonomy changes with version labels like 'proposed,' 'active,'
'deprecated,' and 'removed.'" An event's own owner is a separate question from the plan's own
owner and from the approval path; this template keeps the three apart rather than collapsing
them into one column. Deep dive: tracking-plan_companion.md section 3 (Event and Property
Catalog) and section 6, item 5.
ASK For each event: what is its name, following the convention above? When does it fire? What
business question does it answer? What properties does it carry, with type and allowed values?
What source emits it, and what destination receives it? Who is its owner? What is its current
status?
PRIORITY No event ships without a named owner and a one-sentence reason. Status uses a stated
vocabulary, proposed, active, deprecated, and removed, or the team's own list named once rather
than invented per row.
ROW HINT A good row answers "why it is collected" in one sentence a non-analyst could repeat back. A
weak row names the event and leaves the reason blank because the name alone looks
self-explanatory.
GOOD | shift_claimed | A worker taps Claim on an open shift | Measures how fast open shifts fill,
feeding the fill-rate metric | shift_id (string), worker_id (string), claimed_at (datetime),
source_channel (enum: push, email, in_app) | Worker app -> event pipeline -> warehouse | Jordan
Ruiz | active |
WEAK | shift_claimed | | | | | | active |
TRAP Treating the event name as self-explanatory and leaving the "why it is collected" cell blank; a
plan nobody can re-justify is the owner-departure failure mode this bundle's research names. -->
**Status vocabulary used below:** {{status_vocabulary}}
| Event | Fires when | Why (business question) | Properties | Source -> Destination | Owner | Status |
|---|---|---|---|---|---|---|
| {{event_name}} | {{event_trigger}} | {{event_reason}} | {{event_properties}} | {{event_source_destination}} | {{event_owner}} | {{event_status}} |
## User and Account Properties
<!-- WHAT A table of the traits kept about a user or an account, apart from any event.
WHY Five independent sources keep this list separate from the event catalog rather than folding it
in. Twilio Segment defines traits by what they are not: events, but persistent facts, "Objects
persist over time, so traits can change over time." Amplitude's documentation splits the idea in
two: "User properties capture traits about the user", and, as a parallel field for a group,
"Group properties make it easy to associate a user with a particular account". Mixpanel's own
best-practice checklist names "user profile properties" alongside event properties as a distinct
line item, and mParticle's platform guide groups "an event, user attributes, or user identities"
as the three kinds of thing a single data point can be. Deep dive: tracking-plan_companion.md
section 3 (User and Account Properties).
ASK What traits are kept about a user? About an account or group? What is each trait's type and
expected values? Where is each one set, or updated?
PRIORITY A trait is a persistent fact, not a property of one event; if it changes independently of
any single action, it belongs here, not in the catalog above.
ROW HINT A good trait row names where the value is set, such as a signup form or an admin console,
so a reader can tell a fact kept about a person from an event that merely mentions them. A weak
row leaves the subject (user, or account) and the source blank.
GOOD | plan_tier | user | string: free, pro, team | Set at signup in the billing service, updated on
upgrade or downgrade |
WEAK | plan_tier | | string | |
TRAP Folding a trait into the event catalog as though it were a property of one event, when it is
really a fact that persists across many events and many sessions. -->
| Trait | Subject (user or account) | Type and allowed values | Source (where it is set) |
|---|---|---|---|
| {{trait_name}} | {{trait_subject}} | {{trait_type_values}} | {{trait_source}} |
## Review and Ownership
<!-- WHAT The plan's own named owner, apart from any event's owner, how often the whole plan is reviewed,
and what else, beyond the calendar, should trigger an earlier review.
WHY Snowplow states the plan-level role directly: "An explicit owner; that is responsible for
maintaining and updating the data over time". How often the whole plan is reviewed is contested
outright: one practitioner states a fixed ninety-day cadence and credits it to a named third
party, a monitoring vendor rejects any fixed interval, and Countly instead ties review to an
event, "It should be versioned and reviewed whenever the product or measurement strategy
changes." No interval is a convention here; this section asks the author to name one and to say
plainly whether it is this team's own choice or a borrowed figure. Deep dive:
tracking-plan_companion.md section 3 (Review and Ownership) and section 6, item 2.
ASK Who is the plan's own owner, apart from any event's owner? How often is the whole plan reviewed?
What, besides the calendar, should trigger an earlier review? Is the stated interval this team's
own choice, or a figure borrowed from somewhere else?
GOOD "Owned by Jordan Ruiz, Analytics Lead, apart from any individual event's own owner. Reviewed
every ninety days; this interval is this team's own choice, not a sourced norm. A major
navigation redesign also triggers an off-cycle review."
WEAK "Reviewed periodically." (no owner named, no interval, and nothing a reader could check the plan
against)
TRAP Confusing the plan's own owner with an individual event's owner, or borrowing a cadence figure
and presenting it as an industry norm rather than naming it as a borrowed figure or a team's own
choice. -->
**Plan owner (apart from any event's owner):** {{plan_owner}}
**Review interval:** {{review_interval}}
**Other review triggers:** {{review_triggers}}

The reasoning, the history and every source, in the repository:

  • Companion - the long-form argument: why these sections, where the sources disagree, and what the bundle refuses to claim
  • History - what changed in this bundle, and when
  • Research log - every source consulted, with what each one actually supports
  • Catalog metadata - the machine-readable record this page is generated from

Catalog record: 7 sections across 1 format(s), methodology growth/data, typically owned by PM / Data.