Buying principleThe valuable deliverable is not a dashboard. It is a repeatable route from a product question to trustworthy evidence, a named decision and a recorded next action.
A product analytics agency should start with one decision
Ask the agency to write the decision before it writes a tracking plan. “Understand activation” is a theme. “Should the team require workspace setup before inviting a colleague?” is a decision with two visible product options.
That wording determines the evidence. The team needs to know who reached each step, which path they took, where they stopped and whether the behavior persisted. It does not need every possible click in the application.
The GOV.UK Service Manual recommends defining objectives, deciding what data is needed and considering measurement from the start. Its performance data guidance also ties metrics to decisions about whether a service meets user needs. A supplier that starts with dashboard layouts has skipped the question that gives the data meaning.
Turn the critical journey into an event contract
An event contract states what happened, who or what performed it, when it happened and which properties make the record interpretable. It also states who owns the definition and which conditions must be true before the event fires.
For the workspace example, `workspace_created` should fire after the backend accepts the workspace, not when someone merely clicks the submit button. Useful properties might include plan type, creation source and whether a template was used. Free-form names, email addresses and unbounded text do not belong in the payload by default.
PostHog’s event documentation describes an event as a name, distinct identifier, timestamp and properties. Google Analytics likewise documents case-sensitive naming rules and reserved names. The exact vendor matters less than a contract engineers and product managers can read before implementation.
| Product question | Required record | Quality check |
|---|---|---|
| Do invited teams activate more often? | Workspace created, invite sent, core task completed | One stable account and user identity |
| Where does setup fail? | Each completed setup step plus error code | Server acceptance, not button clicks alone |
| Does a new feature retain use? | Feature used with release cohort and date | Test accounts and staff excluded |
Set identity and privacy boundaries before implementation
Funnels become misleading when the same person appears as three anonymous visitors or when a shared business account is treated as one individual. The agency should define anonymous, user and account identifiers, the moment identities merge and the fields that must never leave the product.
Imagine a buyer who explores anonymously, signs up on a phone and later joins a company workspace on a laptop. Without explicit merge rules, the activation funnel can count three people and miss the relationship between acquisition and product use.
Privacy is not a checkbox supplied by an analytics vendor. The UK’s ICO updated its storage and access technologies guidance on April 29, 2026, covering cookies, pixels, fingerprinting and related technologies. Ask the agency to document data purpose, collection location, retention, access and consent behavior, then have the appropriate privacy owner or counsel approve the design for your markets.
Test raw payloads before trusting the dashboard
A chart can render perfectly while its underlying events are duplicated, late or attached to the wrong identity. Quality assurance should inspect raw payloads, not only whether a line moves after a test click.
Use controlled accounts for the critical journey. Complete it once, abandon it once and trigger each expected error. Then compare the database state, event stream and report. The workspace example should produce one accepted creation event, no creation event for a rejected name and a clear invite sequence for the same account.
PostHog’s data management documentation exposes raw names, properties, examples, verification status, history and ingestion warnings. Segment also documents versioned tracking plans and schema validation. Whatever stack is chosen, require a test matrix, observed payloads and an owner for violations after launch.
- Happy path, abandoned path and each material error state are tested.
- Client and server events cannot double-count the same outcome.
- Test, staff and automated traffic can be filtered consistently.
- A renamed event has a migration rule and an effective date.
- Release changes are annotated so metric movement has context.
Dashboards need an owner and a decision cadence
A report earns its place when a named person reviews it on a stated schedule and knows what action follows a change. Without that operating rule, the dashboard becomes office wallpaper with excellent filters.
For the workspace decision, the product lead might review the activation cohort every Tuesday for four releases. If invitations improve core task completion without increasing setup abandonment, the team keeps the sequence. If not, it revisits the invite timing and qualitative evidence.
PostHog documents that cohorts can group people by shared properties or completed behavior and be reused across funnels, retention and other analyses. The agency should deliver the cohort definition, time window, exclusions, owner and decision threshold together, not leave those assumptions hidden inside a chart.
Product analytics shows behavior, not motive or causality
Behavioral data can reveal that people stop at invitation, but it cannot tell you whether they fear spamming colleagues, lack permission or simply do not have a teammate yet. A correlation between invitations and retention also does not prove that forcing invitations will create retention.
Pair the signal with a small number of session replays where appropriate, support conversations or a focused UX research engagement. Then use an experiment or staged release when the product risk justifies causal evidence.
This is also where scope matters. Choose website analytics implementation for acquisition channels, pages and marketing conversions. Choose a product discovery workshop when the team has not yet agreed which problem deserves investment. Product analytics is the right purchase when a live product has a decision, observable behavior and someone able to act.
Use this checklist before hiring a product analytics agency
A credible proposal shows how a decision becomes a tested event, a trusted analysis and an owned action. Check the following points before comparing day rates or tool logos.
- The first product decision, decision owner and review date are explicit.
- Every critical event has a definition, trigger, properties and technical owner.
- Identity merge rules, privacy boundaries, retention and access are documented.
- The proposal includes raw payload QA and failure cases, not only dashboards.
- Cohorts, time windows, exclusions and release annotations are visible.
- The handoff includes a data dictionary, test evidence and a maintenance owner.
- The agency states what the data cannot prove and where research or experiments are needed.
Bring Seuil one product decision and the journey behind it
Send Seuil the decision your team needs to make, the user journey it affects and access to the existing event definitions. That is enough to identify the smallest useful evidence set, the data risks to audit and whether the engagement should begin with analytics, research or discovery.