What an analytics provider is hired to do
Analytics help comes as an agency with a measurement team, a freelance Tag Manager specialist, a consultant, or the analytics arm of the agency that already buys the media. The work is reviewing a set-up and saying whether its figures deserve trust; building, from the measurement and tracking plans through the property, the container, consent behaviour and the BigQuery export to dashboards; and looking after the result with weekly checks, change requests and container audits.
Some decisions stay with the business whoever does the work: which questions the measurement answers, what counts as a sale or a qualified lead, which actions become key events and which conversions steer bidding. The measurement plan tutorial starts from objectives the business has already agreed and says plainly that it cannot invent them. A supplier can help put them into words, but they belong to the people who will act on the numbers.
Tracking work also carries a conflict of interest. The comparison of in-house teams and agencies argues that measurement belongs with the side that is not judged by it. When the agency spending the advertising budget also builds the tracking that reports its results, the business holds the definitions and checks the figures itself.
Shortlists, pitches, fee models, contracts and reviews for any agency sit in the site's agency selection and review guide, and where tracking feeds an advertising account, the paid search version of this guide explains how conversions are defined there and which are primary.
What the business keeps
Tracking work happens inside five objects, and who holds each is settled before any proposal is compared. The guide to hiring a CRO agency already lists the analytics property and the tag container among what the business keeps; for an analytics supplier, building inside all five is the whole job.
| Object | How the business holds it | What the supplier receives |
|---|---|---|
| The GA4 property | Under the business's own account, with an administrator who works for the business | A user role that matches the work |
| The Tag Manager container | In the business's own account, with publishing under written review rules | Container rights for the people doing the work |
| The BigQuery export project | A Cloud project the business owns and pays for, with a named owner | Access to the datasets it queries |
| The dashboards' data sources | Connected through a service account or a role the business controls | Edit rights on its reports, not credentials tied to its staff |
| The tracking plan | A versioned document the business can open without the supplier | The right to propose changes through the plan's change control |
The GA4 setup tutorial sets the property up in an account the business itself controls, keeping it out of agency accounts and personal logins on the grounds that a property's ownership is difficult to shift afterwards, and grants access outward from there. The guide to containers and workspaces applies the same rule to Tag Manager: agencies and contractors join the business's account, rights go to named Google accounts, and a container that starts life in an agency's account can be exported yet cannot be moved, so its version history and permissions stay behind.
The BigQuery export basics guide lands the daily tables in the business's own Cloud project, where they stay for as long as the business keeps them, and lists a project with no named owner among its mistakes. Each dashboard data source connects with somebody's credentials, and the dashboard governance guide describes every copy relying on a departed owner's credential going blank, which is why it puts credentials on a service account or a role.
The agency onboarding checklist keeps analytics and tag containers with the business among its first checks, and each grant follows the access request procedure for marketing tools: a named account with the least access the work requires, an end date from the requesting manager, and nobody given the business's own password.
Scoping: audit, implementation or ongoing care
Each kind of tracking work leaves something different behind, so scope them separately.
| Scope | What the business receives | When it fits |
|---|---|---|
| Audit | A scored review of plan, tags, consent, events, campaign tagging, integrations and data quality | A set-up exists and nobody can say whether to trust it |
| Implementation | Agreed plans, the property and container configured to them, developer specifications and a verification record per row | Nothing trustworthy exists, or an audit found gaps too wide to patch |
| Ongoing care | Weekly checks, change requests run through the tracking plan, container audits and current access | Once an implementation has been accepted |
Begin with an audit wherever something is already live: the analytics implementation audit scores what exists, from the written plan to whether its figures match the business's own records, and ranks the gaps it finds by how many decisions they distort.
For an implementation, the measurement plan is the brief and the tracking plan's rows are the acceptance criteria. Write into the scope who changes the site itself: an event that depends on values only the site knows, such as an order's value or a repair reference, needs the site's developers to push them into the Data LayerA structured object on a web page that passes values to a tag manager for tags to use., and a supplier with access to the container alone cannot finish that part. Ongoing care is continuing work with a steady scope, the kind the selection guide matches to a RetainerA standing fee paid regularly to an agency or supplier for continuous work with a stable scope, as distinct from project fees, time and materials or commission.; an audit or a build has a finish line and is scoped as a deliverable.
The request for proposal template for agencies has every supplier answer the same numbered questions in one fixed order, with the scoring published before responses arrive; put the questions below among them. After the choice, the scope of work template for agencies holds what was agreed: list the five objects in its ownership field, and write whether tags for media the supplier does not run, the consent configuration and the dashboards are in scope. That scope becomes the Statement of WorkA supplier's written confirmation of what it will deliver, by when, at what price and how it will invoice, against which the order is raised. the work and invoices are read against.
Questions to ask before hiring
These are the questions particular to tracking; those about the team, cover and fees belong to the selection guide.
| Question | A careful answer includes |
|---|---|
| Whose property, container and export project will the work run in? | Ours, with named access that we grant and can remove |
| What will you work from, and what will you leave behind? | Measurement and tracking plans that we hold, updated with every change |
| How will we see each event working without reading code? | A shared preview session, DebugView, negative tests and a dated record per row |
| Who publishes container changes, and how is each reviewed? | Named people, one change per workspace, version notes naming the plan version |
| How will tags behave when a visitor declines consent? | They follow the choice, with no promise to measure people who said no |
| How do you keep personal data out? | No addresses or names in page paths or parameters; hashed emails treated as personal data |
| How will figures be checked against our own records? | A weekly comparison of key events with the order system or CRM |
| What happens when the arrangement ends? | A current plan, a record of everything built and access removed in an agreed order |
On publishing, the container governance guide sets who builds, who publishes and under which review rules, including when an agency's people may publish.
Rank the answers using a weighted scoring matrix: criteria and weights agreed before anyone reads a proposal, each score backed by what the proposal says, and must-haves applied as gates rather than added in. Holding the five objects in the business's name is such a gate.
Warning signs
- A fresh property or container in the supplier's account. Ownership of a property is hard to move later, and a container made in someone else's account leaves its version history and permissions there.
- A request for the business's own login. The access request procedure above gives each supplier a named account instead.
- No written plan, or the container offered as the documentation. The tracking plan guide lists letting the container become the plan among its mistakes, because nobody can then tell what the build was meant to achieve.
- A promise to recover visitors who declined consent. The server-side tagging basics guide answers plainly that server-side tagging is no way around consent, and the guide to privacy, consent and measurement adds that neither first-party nor server-side techniques recover measurement from people who said no.
- Hashed email addresses proposed as identifiers. The guide on personal data in analytics tools treats a hashed address as the address it came from and wants a User-ID built from an internal reference instead.
- A rebuild offered before anyone has read the current set-up. The scoping above starts with an audit wherever something is already live.
Accepting the work
Acceptance checks behaviour against a written specification, and anyone can watch behaviour.
Accept row by row. Each row of the tracking plan is an acceptance criterion: the event name, the trigger and what must not trigger it, the parameters and their types, the consent category, and an owner and a verifier who are different people. A row reaches verified only when someone other than its builder has checked it, and the measurement plan holds launch until every row is signed.
Watch each event yourself. The tutorial on debugging with preview mode describes a share link that lets a client follow a session on the draft container without edit rights, and DebugView then shows what the property actually received. During the session, buy the test product, send the enquiry form, decline consent and reload the confirmation page, and confirm each event appears once with the promised parameters and stays silent where it should.
Keep the evidence. The tracking verification checklist makes each item a pass or a fail from a real test action and ends with a record saved with screenshots and the date, which is what the business points to later when someone asks whether the figures could be trusted.
Compare with the business's own records. The weekly routine in the marketing data quality guide compares key events over the latest full week against the order system or CRM, within the tolerance the measurement plan records, and confirms the live container version is the one the tracking plan names. Start it in the first week after acceptance, run by a named person at the business.
Collect the paperwork. The build ends with a Handover RecordOne document, kept by a coordinator, listing everything a departing marketer owns: tasks, open work, accounts, suppliers and know-how, each with a new owner.: a row for every object built or changed, where it is documented and who owns it next, alongside the live container version, the vendor register entries for any new third parties and the current tracking plan.
Changing or ending the arrangement
The offboarding checklist for agencies sets the order in which the arrangement is wound up. The GA4 property handover checklist is run on the property before the outgoing supplier's access goes, since its documentation group needs their help. Then remove the supplier's rows from the register the Access ReviewA periodic check of who can use each marketing tool: every account matched to a current person, supplier or integration, with owners, deputies and narrow roles. keeps, person by person.
When results fall once a new supplier takes over, the diagnosis of a drop after an agency change looks at whether accounts, tags or history stayed behind. When the supplier still holds what the business paid for, the diagnosis for an agency keeping the accounts works out why and what to do, for an analytics property as for an ad account. If the property sits in the supplier's account and no transfer is on offer, the new-property tutorial sets up a replacement under the business's own account and keeps old and new running together until links, conversions and reports have moved across. The old property's history stays behind, which is the price of settling ownership late.
A clockmaker compares tracking proposals (illustrative)
Klivodrenz Clocks chooses and accepts tracking work
Klivodrenz Clocks, an invented clockmaker that sells new and restored clocks online and books repairs through its site, asks two agencies to rebuild its tracking after an audit found duplicate purchase events and no written plan. The owner and the operations manager agree the gate and the weights before opening either proposal.
| Criterion (weight) | Proposal A | Proposal B |
|---|---|---|
| Property, container, export project, data sources and plan held by the business (gate) | Passes | Fails: a fresh property in the agency's account |
| Written tracking plan (30%) | 5: 14 rows drawn from the measurement plan | 2: events listed, no plan |
| Evidence per event (25%) | 5: preview link, DebugView, dated record | 2: a report after launch |
| Consent (20%) | 4: tags follow the banner's choices | 1: promises to recover declined visits through a server container |
| Handover (15%) | 5: handover record, plan and version notes | 1: not covered |
| People and cover (10%) | 4: a named lead and a deputy | 4: one named lead |
| Weighted score out of 5 | 4.70 | 1.85 |
Proposal B fails the gate, so its scores, shown for comparison only, decide nothing. Proposal A is hired; its named lead builds in its own workspace in the business's container, and the operations manager reviews each version in a shared preview and publishes it, with a version note naming the plan version it implements.
At the first acceptance session 12 of the 14 rows pass. The repair booking event fires again when the confirmation page is reloaded, and the clock type parameter arrives empty on the restored clocks template because the shop's developer has not yet added it to the data layer push. Both rows go back to implemented; a week later the operations manager repeats the negative test, sees the parameter arrive in DebugView and signs them. The first weekly check finds the week's purchases in GA4 inside the measurement plan's tolerance against the order system, and the handover record closes the build with all 14 rows, their verifiers and dates.
The measurement planning hub points to the guides on the plans this guide relies on and on the routines that keep the data trustworthy.
Common mistakes
- Comparing proposals before deciding who holds the property. A cheaper build in the supplier's account becomes the dearest option on the day the arrangement ends.
- Letting the supplier set the definitions. Key events, conversions and what counts as a lead are the business's decisions.
- Accepting a demonstration the supplier chose. Take the test actions yourself, including those that should fire nothing.
- Dashboards connected through a supplier's login. When that login goes, the reports go with it.
