Analytics engagements usually begin with a request for better reporting and should usually begin somewhere else. Dashboards built on a broken collection layer produce confident answers that are wrong, and the most common finding of a competent first engagement is that several months of data cannot be trusted. Buying the dashboard first is how businesses end up making decisions on numbers nobody validated.
Collection before reporting, always
Tags firing twice, conversions counted on page reload, filters excluding a market, consent handling that drops a chunk of traffic silently, and offline sales that never make it back. Any one of these makes every downstream number wrong in a way no dashboard will reveal. A first engagement should be an audit of what is collected and whether it is right, delivered as a list of defects with severity. If an agency proposes dashboards before that, it is selling the visible half.
The agency that spends the money should not grade its own work
Where the same supplier buys the media and reports on its performance, the attribution choices are theirs. That is not necessarily dishonest and it is a structural problem. Either buy the measurement from somebody who does not buy the media, or agree the attribution model, the window and the source of truth in writing at the start so the choices are not made after the results are known.
Attribution models are assumptions, and they should be stated
Last click, data-driven, position-based and media mix modeling each answer a different question and produce different winners. None is correct in the abstract. What matters is that everybody knows which model is in use, what it assumes, and that comparisons are made within it rather than across it. A report that changes model between quarters is not a report, it is an argument.
Consent handling is now part of the measurement
Privacy controls and consent requirements mean a portion of activity is not observed, and how a measurement setup handles that, modeling it, ignoring it, or silently under-counting, changes every number in the report. Ask any prospective agency to explain what happens to your data when a visitor declines, and treat an unclear answer as a finding about the agency.
Buy the first engagement as a fixed project
An audit of collection, a documented measurement plan, a definition list for every metric the business uses, and a small number of reports somebody will actually read. That is a finite piece of work with a deliverable you keep, and it tells you whether the supplier is any good. Analytics retainers bought before that tend to produce dashboards nobody opens and a monthly invoice nobody questions.
Questions people ask about marketing analytics agency
What should an analytics agency deliver first?
An audit of what is being collected and whether it is correct, with a defect list ranked by how much it distorts decisions. Dashboards built on unvalidated collection are the most expensive kind of wrong, because they are confident. Every competent analytics engagement starts here.
Can our media agency do the measurement?
It can, and it means the party spending the money is also choosing how success is judged. If you do it that way, fix the attribution model, window and source of truth in writing before the first campaign. Otherwise buy the measurement separately, which is cheaper than it sounds because it is periodic rather than continuous.
Do we need media mix modelling?
It is useful at meaningful spend levels and across several channels, and it needs a long history of clean data to be worth anything. For most mid-sized advertisers the cheaper and more decisive method is incrementality testing: turn something off deliberately and see what changes. Any agency that proposes modeling before you have clean collection is sequencing it wrongly.
How much should this cost?
The audit and measurement plan is usually a fixed project and should be priced as one. Ongoing work should be scoped to what is actually maintained and read, which is a smaller number of reports than most engagements produce. Pay for the definition work; be sceptical of paying monthly for dashboards nobody opens.