Hiring a Google Analytics or web analytics agency, and the five deliverables worth paying for
What an analytics agency can genuinely fix, the five deliverables to put in the scope, and the published per property limits and model list, read on 22 October 2026, that bound the work.
Published by AI Knows Us · Updated 22 October 2026
An analytics agency is worth hiring when your numbers cannot be trusted enough to act on, and not worth hiring to produce more reports. The scope that pays for itself is short: a measurement plan, a clean implementation, a documented set of definitions, a working join to your sales outcomes, and a handover that leaves you able to run it. Everything else is a dashboard, and dashboards are cheap.
The five deliverables, and what each one has to contain
| Deliverable | What it has to contain | How you check it was done |
|---|---|---|
| Measurement plan | Every business question, the event that answers it, the denominator it is read against | You can point at a question and find its event |
| Implementation | Events firing once, named consistently, with the parameters the plan asked for | A tag debug session you watch, not a screenshot |
| Definitions document | One page per metric, with the model and the window named | Two people in your team give the same answer to "what is a conversion" |
| Outcome join | A path from a closed deal or a refund back to the session that started it | You can find one real deal in the data end to end |
| Handover | Admin access held by you, the container documented, the plan in your drive | You could change agency without losing anything |
The handover line is the one to settle in the contract rather than at the end. Property and container ownership should sit with you from day one, with the agency granted access. That is the same principle that applies on the advertising side, where a manager account structure keeps the client's account in the client's name.
The published limits the agency is working inside
Three documented constraints shape what any analytics agency can deliver, and a good one will raise them unprompted.
First, collection ceilings are per property. Google's event collection limits page, read on 22 October 2026, states that the limits "apply on a per-property basis", that certain limits can be raised "by upgrading your property to Google Analytics 360" but not the event and parameter character limits, and that "Google Analytics doesn't log events, event parameters, and user properties that exceed" them. An over enthusiastic tagging plan can therefore lose data quietly rather than error loudly.
Second, retention is a setting with consequences you cannot undo. The data retention documentation, read on 22 October 2026, describes controls that set "the amount of time before user-level and event-level data stored by Google Analytics is automatically deleted from Analytics servers". Any year on year analysis has to be checked against the setting that was in force when the older data was collected.
Third, the model list is short. The attribution documentation, read on 22 October 2026, says "there are 3 attribution models available in the Attribution reports in Google Analytics properties: Data-driven attribution, Paid and organic last click, Google paid channels last click", and that "all attribution models exclude direct visits from receiving attribution credit, unless the path to key event consists entirely of direct visits". On the advertising side, the Google Ads attribution page says "the first click, linear, time decay, and position-based attribution models are no longer supported by Google". An agency promising a custom multi touch split from these sources is promising its own model, which is a legitimate thing to buy and a different thing to buy.
One export quirk worth knowing before you argue about a number
Search Console exports are a common source of reconciliation arguments. Google's performance report documentation, read on 22 October 2026, notes that values shown in the report as a tilde or a dash, meaning not available or not a number, "will be zeros in the downloaded data". A zero in a spreadsheet that was a blank on screen has started more than one wrong conclusion.
How to brief and judge one
Write the five business questions you cannot currently answer, in plain words, and make them the first page of the brief. Ask each candidate to return a measurement plan for those five before any implementation work, and judge the proposals on that document alone. Ask who will do the work and what percentage of their week you are buying. Ask for one anonymised definitions document from a previous engagement, because that artefact is the hardest to fake and the most useful to inherit.
Two things to refuse: a dashboard as the first deliverable, and any arrangement where the property or tag container is created under the agency's own account. For the reporting layer that leadership actually reads, our guide to reporting to leadership covers what belongs on three slides, and the denominators behind each metric are in the metric dictionary.
A figure that does not exist: there is no published industry benchmark for how many analytics implementations contain duplicate conversion events. We looked and there is none to quote. Count your own in a debug session and write the number down with the date.
A disclosure, because we sell a measurement of the thing this page discusses. AI Knows Us reads five AI surfaces and reports two separate findings, whether a business is named and whether its own pages are cited, and they are never added together. Our own reading on 22 October 2026 was named 0 and cited 0 across 635 readings, published in full as our own null result after 507 new pages. Where a rival is stronger, plainly: Profound publishes SSO and SAML, SOC 2 and up to nine engines; our rendered pricing page carries none of those words and names five surfaces, not nine. The cheapest plan of ours that reads all five is Growth at 299 dollars a month in the US and the UK, and the 149 dollar SEO Start plan reads 0 questions, 0 credits and 0 engines, so it is not an AI visibility plan.
Questions buyers ask about analytics agencies
Is an analytics agency different from a web analytics consultant
Mostly in capacity rather than in kind. A consultant is the right purchase for a plan and an audit. An agency is the right purchase when the implementation work is large enough to need more than one pair of hands.
Who should own the Analytics property and the tag container
You should, with the agency granted admin access. It costs nothing on day one and it saves the entire handover later.
Should the same firm do analytics and media buying
Separating them gives you an unconflicted read on whether the media is working. Keeping them together is faster. If you keep them together, make somebody internal the owner of the definitions document.
What is the single most common fault they find
Conversions counted more than once, usually from a tag firing on both a thank you page and a form submit event. It inflates every rate downstream, so it is worth checking before you hire anybody.
When can we judge whether the engagement worked
Judge it on artefacts and agreement rather than elapsed time: the five questions answerable from the data, two people in your team giving the same definition of a conversion, one real deal traceable end to end, and the admin access sitting in your account.
The channel your analytics cannot see
When an assistant answers a buying question without sending a click, there is nothing for an analytics property to record. That reading has to be taken directly on a fixed question set, and the citation count and the naming count are two findings reported side by side rather than added.