Marketing data is almost never ready for measurement by default. Platform exports, agency reports, and CRM data all use different definitions, date ranges, and granularities. Getting them consistent is unglamorous work, and it is the difference between measurement you can trust and measurement that sounds plausible.
Every measurement project eventually hits the same wall: the data is messier than expected. Spend data that does not match agency invoices. Conversion counts that differ between GA4, your CRM, and your ad platforms. Campaign names that changed halfway through the year. Fixing these problems mid-project is expensive and slow. Fixing them beforehand is much cheaper and makes every measurement initiative faster and more reliable.
Define your metrics and stick to them
Before you collect any data, agree on a single definition for each key metric: what counts as a conversion, what counts as a sale, what date a transaction is attributed to, and how you handle returns or cancellations. Document these definitions in a data dictionary, a reference document that lists every metric, how it is calculated, and where the source data comes from. Without a dictionary, different people calculate the same metric differently and your data is inconsistent before you even look at it.
Standardise your campaign naming conventions
Campaign names in ad platforms are free-text fields, and without a convention, they become unusable for analysis. A campaign called Test August 2024 tells you nothing about the channel, product, audience, or objective. Implement a naming convention that includes: channel, campaign type, product or service, audience, and a date or version identifier. For example: paid-social_prospecting_product-X_uk-25-44_2026-Q3.
- Apply the naming convention across all platforms and require all agencies to follow it.
- Audit existing campaign names and retroactively classify them where possible.
- Use a taxonomy document that defines every allowed value for each naming segment.
- Build a QA check into your campaign setup process so naming errors are caught before campaigns go live.
- Treat UTM parameters with the same rigour as campaign names: broken UTMs are one of the most common causes of unattributed traffic in GA4.
Reconcile spend data across sources
Spend data typically comes from multiple sources: agency invoices, platform billing exports, and finance system records. These rarely agree. Agree with your finance team on which source is the authoritative record for each channel and use only that source for measurement. Where discrepancies exist, investigate and resolve them before they enter your measurement data. A 10 percent discrepancy in spend data will create a 10 percent distortion in any model that uses it.
The most common spend discrepancy is between net media spend (what you pay for the media) and gross spend (which includes agency fees and production costs). Make sure your model input uses the same definition of spend throughout the historical period. Mixing net and gross spend in the same dataset systematically distorts channel efficiency comparisons.
Handle tracking changes carefully
Tracking changes (moving from Universal Analytics to GA4, changing how you count conversions, implementing consent mode) create breaks in your data series. A break is not automatically a problem, but an undocumented break is. Keep a change log of every significant tracking or methodology change with the date it took effect. Your measurement partner needs this log to correctly interpret anomalies in your historical data.
Collect the context data that explains your sales
Marketing data alone is rarely enough to explain revenue patterns. Your model also needs data on factors that affect sales independently of marketing: pricing history, promotional events, distribution changes, product launches, major competitor activity, and external events like economic shocks or category disruptions. This context data is often held in different parts of the business and needs to be gathered deliberately. Start collecting it now, even if you are not yet running a model, because you cannot go back and create historical records after the fact.
Our data lives across multiple agencies and platforms. How do we centralise it without a large tech investment?
Start with a simple shared data store: a well-structured spreadsheet or Google Sheet that each agency populates weekly in a standard format. This is low-tech but effective enough to begin measurement work. As your volume of data and your measurement ambitions grow, move to a cloud data warehouse (BigQuery, Snowflake, or Redshift are the most common) with automated ingestion from your platforms. Build the habit of data collection now and improve the infrastructure later.
We use multiple attribution tools. Which one should we rely on for our measurement data?
Pick one as the authoritative source and use it consistently. Using multiple attribution tools and averaging or selecting between them based on which looks most flattering creates unmeasurable noise in your data. If you are not sure which tool to trust, that is a sign you need to evaluate them against each other using an independent method, such as an incrementality test, and retire the one that performs less well.
How long does it take to get data truly ready for an MMM project?
For a business with reasonable data infrastructure and active agency relationships, two to four weeks of focused effort typically produces data that is ready for modelling. For businesses with fragmented systems, multiple agencies, and historical tracking changes, allow eight to twelve weeks. The time investment upfront saves significantly more time during the modelling project, and it results in a model you can actually trust.