Measuring offline and digital together runs in three steps. First, assemble clean historical spend and outcome data. Second, build a marketing mix model for the top-down view across every channel. Third, settle disputes between the model and your dashboards with geo-lift and holdout experiments, then reallocate on a cadence.
The article showed why the trackable channels quietly win the budget they did not earn. This is how to build a measurement view that sees every channel at once, so the money follows results rather than the ease of counting them. Treat it as a strategic bet, not a weekend project. The payoff is real, and so is the effort.
Step 1: assemble the historical data
Nothing works until the data does. Marketing mix modelling reads history, so it is only as good as the history you feed it.
Pull weekly spend for every channel, offline included, and the outcome you care about, usually sales or revenue, over two to three years. Add the known drivers that also move sales: price changes, major promotions, distribution shifts and seasonality. A model that does not know a price cut happened will hand the credit for that cut to whatever channel ran at the same time.
Owner: your analytics or data engineering lead. What to measure at this stage: coverage and completeness. What share of total spend do you have in clean weekly form, and how many gaps or definition changes sit in the series? This is the step that loops back to data quality. If the history is patchy, fix that first, because the model inherits every flaw in the input and presents it back as confident output.
Do not wait for perfect. Aim for consistent. A complete-enough series with honest gaps beats a heroic reconstruction nobody trusts.
Step 2: build the marketing mix model
Now build the top-down view. Marketing mix modelling relates all that spend to sales over time and gives you each channel’s contribution, together, without needing a cookie or a login.
The entry cost has collapsed. Open-source tooling such as Meta Robyn, Google Meridian and PyMC-Marketing does the heavy statistics for you, so you are no longer paying six figures to a specialist agency for a black box. Start with one of these, on the data from Step 1.
Owner: a capable analyst, in-house or contracted. What to measure: model fit against actual sales, each channel’s estimated contribution, and the diminishing-returns curve per channel, which tells you where the next pound stops working. Read the results as direction, not gospel. The model will tell you brand and offline are doing more than your click dashboards ever showed. That is the point. It is seeing the channels the pixel cannot.
Two guardrails. Do not over-tune the model until it fits the past perfectly, because a model that explains every wiggle of history usually predicts the future badly. And do not present a single number as truth. Present a range, and be honest about what the model cannot see.
Step 3: settle disputes with experiments, then reallocate
The model will disagree with your attribution dashboards. Good. That disagreement is the most useful signal you have, because it marks exactly where your budget decisions are riskiest.
When the two fight, do not argue in a meeting. Run a test. A geo-lift experiment holds a channel back in some regions and runs it in others, then compares outcomes. A holdout does the same across a matched audience. Either gives a causal read, what the channel actually caused, which neither a model nor a dashboard can prove on its own. This is the tie-breaker.
Owner: marketing operations, working with the analyst. What to measure: the incremental lift the experiment reveals, then the budget you shift because of it, then the sales that follow. Feed every result back into the model so the next read is sharper.
Then set a cadence. Re-run the model quarterly. Keep one or two experiments live at all times on your biggest or most contested channels. Measurement is not a project you finish. It is a loop you run, and each turn of it puts the next pound somewhere better.
How to train your team to hold the fix
The maths is the easy part now. The hard part is a team that trusts a model over a click.
Start with the mindset. A digital-native team is trained to believe the pixel. Show them, with one experiment, a channel the pixel missed. A single clear geo-lift result does more to change belief than any slide. Let the evidence do the persuading.
Then split the roles cleanly. One analyst or small team owns the model and the experiments. Marketing operations owns acting on the output. Keep them close, because a model whose owner sits far from the budget becomes a report that gathers dust. Write down who reads the model, who calls the reallocation, and how often.
Finally, protect the data habit. The measurement is only as durable as the weekly data feeding it. Someone has to keep the spend and outcome series clean every week, not scramble to rebuild it before each quarterly run. Make that a standing job, not a fire drill. The team that keeps the inputs clean is the team whose model still works a year from now.
Where Morphy helps
We build a first cross-channel measurement read in four to eight weeks, using open-source tooling so you own the model and the method, not a vendor contract.
The engagement is scoped to a metric. We assemble your historical spend and outcome data, build a working marketing mix model on Robyn, Meridian or PyMC-Marketing, and design one geo-lift or holdout experiment on your most contested channel. You leave with a model you control, an experiment running, and a named owner trained to run the loop after we go. No black box, no permanent retainer, no platform to buy.
If your historical data is not ready, we say so on day one and fix that first, because a model on broken data is worse than no model. Honest measurement beats flattering measurement every time.
Go deeper on customer data maximization
Three ways forward. Pick the one that fits where you are.
- Get the playbook. Practical notes on turning the customer data you already own into revenue, straight to your inbox. Join the newsletter at the foot of this page.
- Take the assessment. Score your customer data maximization in four minutes and see your top revenue blockers. Start the assessment →
- Book a meeting. Bring your data problem. Leave with a prioritised fix, not a platform pitch. Book a call →
The playbook companion to How do you measure marketing across offline and digital channels?. Post 9 of 25 in the Customer Data Maximization series.
Frequently asked questions
How long does it take to build a first marketing mix model?
With clean weekly data and open-source tooling such as Meta Robyn or Google Meridian, a first usable model takes a few weeks, not a quarter. The slow part is almost never the modelling. It is assembling two or three years of consistent spend and outcome data to feed it.
Do marketing mix modelling and attribution replace each other?
No. They answer different questions. Marketing mix modelling gives the top-down strategic view across every channel, including offline. Attribution gives the tactical view inside digital. Run both, and when they disagree, use a geo-lift or holdout experiment as the tie-breaker rather than picking a favourite.
What data do I need before I start?
Weekly spend by channel, including offline, plus the outcome you care about, usually sales or revenue, over two to three years. You also want known external factors such as price changes, promotions and seasonality. Patchy history is the most common reason a measurement project stalls before it produces a number.
Who should own cross-channel measurement?
A single analyst or small analytics team owns the model and the experiments. Marketing operations owns acting on the output. Split those and the model becomes a report nobody uses. The person who builds the read and the person who moves the budget need to sit close together.