The number one question I get when talking to people about Marketing Mix Modeling (as the Founder of Vexpower.com), is “what does the actual output look like?”.
I never like to say “it depends”, so I’ve taken to giving people a whirlwind tour of what the output looks like at each budget level – I’m publishing it now, so I can just point people here when I next get that question.
In modeling the currency is ‘accuracy’: a simple model done with Excel in an afternoon could get you 60% or 70% of the way there, but who can risk being 40% off target? In the Fortune 500 you might employ a team of 4 people for 6 months once a year, to build a more accurate model that you can use for annual budget planning.
Putting more hours (and ultimately money) into MMM can get you closer, as well as give you unique insights into the mechanics of your business, but is it worth the extra investment? What is appropriate to spend for your business? And how do you know it’ll work?
So how much does a Marketing Mix Model cost? I usually break it down into 3 price points, $500, $5,000 and $50,000+. Let’s see what each model might look like, and along the way I’ll tell you what you get for your money.
Watch the Video
I put together a short video that explains how this works, with additional information not covered in the blog post below. I recommend watching the video first then continuing with the post.
The $500 Model
If you’ve got someone on your team who has a spare afternoon and is decent with Excel / GSheets, you can build a reasonable Marketing Mix Model to help you make a decision.
Of course if you say that to a statistician they’ll recoil in horror, but you don’t go to a chemist for cooking advice – we’re running a business not a laboratory and there’s a big difference between theory and practice.
You can do a surprising amount in spreadsheets, from measuring the halo effect of ads, to determining the impact of word of mouth and forecasting your diminishing returns at scale, and it’s not uncommon for me to build simple ~10 variable models that predict sales to 60-80% accuracy.
Given the simplicity, speed and low cost, even if you are planning to work with a more sophisticated solution, I recommend starting with an Excel model so you can play around and understand the mechanics of what’s going on. It can only help.
Use this template and read my post on econometrics in gsheets to get started.
The $5,000 Model
If you’re not building a model yourself, you’re going to be hard-pressed to find anyone who will build you a model for less than a few thousand dollars. For a serious practitioner it might be closer to $10k, depending on the complexity.
Modeling experts are in high demand, because they can usually earn a lot more in the Finance industry (some of us do it for the love of marketing, or because we graduated in 2009 at the peak of the Great Recession, and there were no jobs in finance so we stumbled into a marketing career…)
You need someone who can gather data from multiple stakeholders, get it into the right format (collecting and cleaning data is 80% of the job!), build complex statistical models and then present it back in a way that business executives can understand… know many people like that? Didn’t think so.
It’s usually not economical to hire someone to do MMM full time, but it’s perfectly viable as a task to give to your data science team. They might not be fully up on all the older statistical techniques, so it might take some light reading to round out their knowledge.
Code takes longer up front, but once you have your data cleaned and scripts written you can run it again and again as much as you like. In fact, it’s common practice to run thousands of potential model combinations (which can take anything from a few minutes to a couple hours) and automatically return the most accurate result.
If you’re going down this path, check out my Python scripts, or Facebook’s open source ‘Robyn’ scripts in R, or contact me as I take on the occasional project.
The $50,000 Model
What’s the difference between a $5k model and a $50k model? Well to be honest the output usually looks about the same! Sure the model will be more accurate (or at least less likely to be wrong). But the general rule is that companies that can afford it, tend to get charged more – the joys of enterprise pricing.
If you’re spending $100m a year on marketing, $100k is just 0.1% of your spend, and if it helps you optimize your spend even a little it can easily be worth it. However for someone spending $1m a year you can’t afford to blow a whole month’s budget on attribution! Even if your model is amazing and took $100k+ of work, only big companies can justify that cost.
Bigger businesses are often more complex, so there definitely is an increased workload between $5k and $50k projects, but it’s not 10x more work. The majority of the increase in cost is explained by admin – it’s just far more expensive to sell to and work with the Fortune 500, so every vendor that works with them bumps up their price to accommodate.
For example I once worked with a Fortune 500 client who took 8 months to sign off on a 3 month project, and then took 18 months to pay us. I counted 57 meetings total. You can bet I charged more the next time we landed an enterprise client.
There is one interesting difference with the work: marketing tends to be a much weaker effect for larger, more established businesses. Startups that aren’t well known yet can just switch on Facebook ads and immediately see a spike: you don’t even need a model to tell you it’s working. When you’re Coca Cola or Nike people buy you for their long term exposure to your brand, not because of anything you did in marketing this year.
If you want to see an example of how MMM presentations usually look, you can check out module 4 of the vexpower.com MMM training simulator (free).