Build in-house if you have a data scientist with Bayesian modelling experience, clean historical data, and the bandwidth to maintain and interpret the model quarterly. Hire a specialist if you need reliable results quickly or if you lack that capability internally.
Open-source MMM frameworks like Meridian and Robyn have made DIY modelling more accessible. But accessible does not mean easy. Building a reliable MMM requires specific statistical skills, significant time, and an ability to catch and correct modelling errors that are not obvious to non-specialists. Before deciding, be honest about what you actually have available.
What DIY MMM actually requires
- A data scientist with experience in Bayesian inference, time series analysis, and marketing domain knowledge.
- Clean, consistent historical data across all channels going back at least 2 years.
- Time to build and validate the model: typically 12 to 20 weeks for a first model.
- A process for quarterly re-runs as new data becomes available.
- The ability to interpret outputs critically and identify when the model is giving implausible results.
- Stakeholder management skills to translate model outputs into budget recommendations leadership will act on.
The hidden costs of DIY
The up-front appeal of DIY is cost saving. But the true cost includes your data scientist's time (a senior data scientist costs 70,000 to 120,000 per year), the opportunity cost of that person not working on other projects, the cost of fixing errors that make it into decisions before they are caught, and the potential cost of a model that is technically built but so complex that marketing leadership does not trust or use it.
Many in-house MMM projects also stall. The first model gets built, results are presented, and then the quarterly re-run never happens because the data scientist has moved to other priorities. A model that is not regularly refreshed quickly becomes irrelevant.
When DIY makes sense
DIY is the right choice when you have an experienced data science team that is already working on measurement, when the budget for an external specialist is genuinely not available, and when you can commit to treating MMM as an ongoing programme rather than a one-off project. Large technology companies and sophisticated retail businesses often build excellent in-house capabilities over several years. It is a long-term investment, not a quick win.
The most common failure mode for in-house MMM is not technical. It is organisational. A model that marketing leadership does not understand, trust, or know how to apply does not change decisions. Whatever approach you choose, invest as much in communication and adoption as in the technical build.
When to hire a specialist
Hire a specialist when you need results within a quarter, when you do not have in-house Bayesian modelling expertise, or when the budget decision being informed (often hundreds of thousands or millions of pounds) is large enough that model accuracy is critical. A specialist brings accumulated experience across dozens of similar businesses, validated methodology, and the ability to spot and fix data issues that would trip up a first-time model builder.
A hybrid approach
Some businesses start with a specialist to build and validate an initial model, then transition model management in-house once the methodology is established and the team is trained. This hybrid approach captures the speed and rigour of specialist delivery while building internal capability for the long term. If this is your plan, make sure the specialist provides full documentation, model code, and training as part of the engagement.
Can we use a junior data analyst to build an MMM rather than a senior data scientist?
Open-source frameworks lower the barrier to entry, but MMM requires enough statistical understanding to know when results are implausible, how to set priors correctly in a Bayesian model, and how to handle the specific challenges of marketing data (multicollinearity, short time series, adstock). A junior analyst following a tutorial is likely to build something that produces outputs but cannot be trusted for major decisions. Invest in training or bring in senior support.
Our marketing team is growing fast. At what size should we consider building an in-house capability?
When your annual marketing spend exceeds 5 to 10 million, the cost of a specialist measurement capability (whether in-house or retained) is justified by the budget optimisation it enables. Below that level, external specialists on a project basis are usually more cost-effective. The decision is less about team size and more about total spend under management.
We built an MMM in-house two years ago. Should we get an external validation?
Yes, particularly if the model has not been significantly updated since it was built, or if it has been used to make major budget decisions. An external validation reviews the model specification, checks for common errors (overfitting, incorrectly specified priors, poor handling of seasonality), and independently assesses whether the outputs are credible. Think of it as an audit of your measurement system.