How long does marketing mix modelling take to set up?

By the Odins team · Updated October 2026 · About 6 minutes

Ask five MMM vendors how long it takes to get a first model and you get five numbers that do not compare, because they start the clock in different places. Some count from the day your data is already collected and prepared. Some count from kickoff. Some leave the data work to you and never count it. This guide sets out the whole timeline, from kickoff to a model you can plan with, and names the step that decides whether it takes four weeks or four months.

The short answer

Plan for four to six weeks from kickoff to a first verified model, with data collection and structuring included. About two weeks goes to getting the data in and structured, more for complex setups, and two to three weeks to the model itself: a few short sessions on what your team already knows about each channel, then training and a round of iterations. Vendors who quote shorter timelines usually count from the day your data is already collected and prepared, and often do not offer data collection at all.

The four stages, and what sets the pace

Stage	What happens	Typical time	What slows it down
1. Access	Read access to the ad accounts, a feed of sales data, and the offline history (TV, radio, out-of-home, print) in a usable format.	Days to two weeks	Waiting for agency logins; offline spend that exists only in invoices.
2. Structuring	Every channel mapped into one hierarchy, online and offline together, down to each placement; two to three years of history loaded; sales data checked for breaks.	About two weeks, more when complex	Campaign naming that changed mid-history; tracking that broke silently; several markets at once.
3. The model	Two or three one-hour sessions to document what your team believes about each channel, then training, verification against actuals and a round of iterations.	Two to three weeks	Channels with no variation in spend; an outcome definition that changes halfway through.
4. In use	Monthly retrain and recommendations; data flows in daily; scenarios and forecasts any time.	Ongoing	Nothing on your side beyond approving access.

Stages one and two are where the timeline is won or lost, and they are the stages many vendors do not count. If your TV and radio spend sits with a media agency, request the history on day one. With a cooperative agency the whole history is up in a day; with a slow one it is the long pole in the plan.

Reporting arrives before the model does

The structured dataset is useful weeks before the model is ready. Once spend and sales sit in one hierarchy, cross-channel reporting on it works immediately, so the first weeks produce something your team can use while the model trains.

What you actually have to do

Very little. The data structuring happens almost without you. What the vendor needs from your side:

Approve read access to the ad platforms, or ask your agency to.
Point to the sales data: a database, a warehouse, an export, or a finance system.
Send the offline history, or ask your agency to send it in the format the vendor specifies.
Join two or three one-hour sessions on what you already know about each channel. The budget you run today reflects beliefs about what works; the sessions write them down so the model can start from them.

After the first model, the recurring work on your side is zero beyond the monthly review of recommendations. Data flows in daily, the model is retrained monthly, and scenarios and forecasts are yours to run whenever a question comes up.

After the first model

Further models are faster. Once the data is in, a second model on the same data, for another brand or another outcome, takes about two weeks.
Multi-market setups can start with one market. One model per market means the first market can be live in the normal window and the others added afterwards, each on its own data.
The first months are a ramp. Early forecasts are checked against actual sales as they arrive. Steady-state accuracy comes after the model has seen a few months of its own forecasts verified; that is when scenarios carry the most weight.

How to read a vendor's number

Three questions turn any quoted timeline into a comparable one.

Does the clock start at kickoff or when our data is prepared?
Who collects the offline spend, and is that inside the number?
What is on our side of the plan, week by week?

A vendor who answers all three with specifics is giving you a plan. A vendor who answers with a smaller number is giving you a sales line. The full checklist for vendor meetings is in how to choose a marketing mix modelling vendor.

Where Odins fits

At Odins the first model is typically live four to six weeks after kickoff, data collection and structuring included, digital and offline. Our team does the collection, the structuring and the modelling; your team joins the sessions on what you already know and reviews the result. A concrete timeline is planned together with our customer success team once you have signed; before that, the range above is what it has taken so far, never a promised date.

Frequently asked questions

How long does marketing mix modelling take to implement?

Four to six weeks from kickoff to a first verified model when data collection and structuring are included. Roughly two weeks of that is data, two to three weeks is the model. Setups with many markets, many offline channels or a complicated sales feed sit at the long end.

Why do some vendors say two weeks?

Usually because they count from the day your data is already collected, cleaned and prepared, and often because they do not offer data collection at all. Ask where their clock starts. Two weeks of modelling on prepared data is consistent with four to six weeks from kickoff once the data work is counted.

Can we speed it up?

The fastest lever is the offline data: ask your agency for the history on the day you decide, in the vendor's format. Giving platform access the same week removes the other common delay. The modelling itself does not compress much; the sessions and the iterations are where the model gets good.

How much historical data do you load?

Two to three years where it exists. The model needs to see each season at least twice to separate seasonal lifts from marketing lifts. With less history the model can still be built, with wider ranges on the seasonal effects.

For what the data stage involves in detail, read data for marketing mix modelling. To see how the four steps run every month after the first model, see how Odins works. Whether your budget is large enough for a model to pay off is in is our company big enough for marketing mix modelling.