Unused, even the best model is worthless. Used, a good model becomes the best.

Your marketing mix model is finished and presented, the team says the numbers are interesting, then goes back to the plan it already had. Ask why and someone says the model isn’t good enough yet.

That explanation is half right. The model isn’t good enough because nobody uses it, and nobody uses it because it isn’t good enough, so someone has to break the loop, and waiting for better data won’t do it.

The gap is between the output and the decision, not in the maths

A Harvard Business Review survey of 547 marketers, sponsored by Google, found that 87% say marketing mix modelling is important, but only 28% say their organisation is very effective at turning those insights into timely action.

That points to a problem with how the results are presented and how the business uses them, not with the statistics.

Use is also what makes a model better. A finance lead challenges a number, or a planner asks why a channel looks flat, and each question sends the analyst back to the data. The answer doesn’t always change the model, but it often builds confidence in it, and sometimes it exposes a gap worth fixing. A model nobody questions never gets that treatment.

Four reasons a sound model still goes unused

Strong econometrics doesn’t protect a model from any of these:

It can be slow. By the time results land, the numbers look out of date and the team has moved on. It can be seen as biased. If the media agency owns it, or it sits only in marketing, finance and sales will doubt the output because it looks like marking your own homework. It can live with only one person. If the business can’t act without that person in the room, it won’t act. It can end at a deck. A long presentation in a one-hour meeting, with no recommended action, no owner and no date for a decision, gets filed.

• It can be slow. By the time results land, the numbers look out of date and the team has moved on.
• It can be seen as biased. If the media agency owns it, or it sits only in marketing, finance and sales will doubt the output because it looks like marking your own homework.
• It can live with only one person. If the business can’t act without that person in the room, it won’t act.
• It can end at a deck. A long presentation in a one-hour meeting, with no recommended action, no owner and no date for a decision, gets filed.

Each of these costs the model use, and each loss of use costs it improvement.

Waiting for perfect data keeps the loop spinning

First projects rarely have all the data they need, and that is normal, not a failure. You might get average price only, impressions and clicks from agencies by channel with no campaign splits or viewability overlay, or CRM data with little segmentation.

Perfect data is not a precondition for a useful model, and waiting for it delays both the model and the improvements that use would bring. Without use nobody has a reason to collect better inputs, so the better inputs never arrive.

Start with what you have, because a strong first version is still possible as long as you stay open to onboarding more data afterwards. Say plainly what the model can’t tell you yet and treat that list as your data plan.

Stakeholders find hidden data when you give them a reason to look

Ask stakeholders for better data and you get a polite “let me see” and little else, because nobody has a reason to go looking, and often nobody knows what they are looking for.

A model that stays fixed to the first set of data it receives gives people nothing new to look for, and it never evolves to answer more. So give them something to react to instead, with recommendations, teasers and what if scenarios, and push them beyond marketing:

What happens if we change the price? Is there room to send more CRM? How did that channel or event perform?

• What happens if we change the price?
• Is there room to send more CRM?
• How did that channel or event perform?

Once a recommendation touches a new area, people start finding data they didn’t realise you could use. It was already in the business, it just needed someone to ask.

That is how new data has arrived on our projects in the past, and each new source lets you answer more strategic questions:

relative price variables CRM split by campaign, with opens, reads and segments event tracking moved from recording that an event happened to capturing contacts and engagement

• relative price variables
• CRM split by campaign, with opens, reads and segments
• event tracking moved from recording that an event happened to capturing contacts and engagement

Stakeholder engagement matters, but it follows from giving people a reason to look. Each new answer makes the model more useful, use makes it trusted, and trust brings better data, so the two compound and each update gets faster and richer.

Three changes that break the loop

Deliver the first version on whatever data you have, be explicit about the gaps, and ask for the data that would close them. End every output with questions beyond marketing, because price, CRM and event questions give stakeholders a reason to provide even better data. Ask for the next data upgrade by name, give each new source an owner and a visible payoff, and show the result in the next update so people see their data being used. Giving each new data source an owner also stops the model living with one person.

• Deliver the first version on whatever data you have, be explicit about the gaps, and ask for the data that would close them.
• End every output with questions beyond marketing, because price, CRM and event questions give stakeholders a reason to provide even better data.
• Ask for the next data upgrade by name, give each new source an owner and a visible payoff, and show the result in the next update so people see their data being used. Giving each new data source an owner also stops the model living with one person.

Two conditions sit behind all three:

Ownership of the model has to be seen as independent and fully informed, with finance involved from the start. Updates need to be fast enough that the numbers stay relevant.

• Ownership of the model has to be seen as independent and fully informed, with finance involved from the start.
• Updates need to be fast enough that the numbers stay relevant.

Summary

The model isn’t the problem, a model nobody acts on can’t improve, and a model that can’t improve gives nobody a reason to act.

You don’t need perfect data to start, you need a model people use. Make recommendations, even where the data and the model appear thin, because that is what pulls out new data and improves the data you already have. Recommend a price change and the commercial team will share the relative price data they track when making those decisions, and the next version of the model is better for it. Judge your measurement programme on whether it changed a decision this quarter, not on how good the model looks.

Look at your last model output and name the decision it changed. If you can’t, write down three questions it didn’t answer and send them this week to the people who might hold the data.

Which decision will your next update change?