Build your marketing mix model in minutes.
One file of weekly sales, spend and prices. TEA reads it, fits the model and shows what every channel, promotion and price move contributed. No supplier, no six weeks, no black box.
Invitation only while in beta. Tell us who you are and we get in touch.
Weekly sales, explained by driver
52 weeks, sample marketAn illustration on one of our sample markets. Your file draws its own, and the numbers come with their confidence.
What the model hands back
A decomposition is only useful if every number in it can be questioned. So every number can.
Every driver gets its share
Media, promotions, price and seasonality, each with its contribution to the KPI week by week, and the baseline underneath: what you would have sold anyway.
A year as a waterfall
From baseline to actual sales, one bar per driver. The number a planning meeting asks for, with the drill-down behind every bar.
The model shows its work
R², MAPE, Durbin-Watson, residuals, and every coefficient with its p-value and VIF. The fit quality is declared next to the result, never assumed.
An export you can defend
The contribution table as CSV, the narrative as PDF or slides, in the same palette as the screen. What you present is what the model said.
The year, split by what drove it
Start from the baseline, add what each driver put on top, land on the sales you actually booked. Marketing added a fifth of the year here, and the waterfall says which fifth.
Hover a bar and the contribution comes with its share. In the product, a click opens the week-by-week detail behind it.
A year of sales, from baseline to actual
sample marketWhat your file needs
Three kinds of column. You almost certainly have them already, in the sheet the agency sends and the one finance keeps.
A date
One row per week or per month. Two years is good, one year works.
week date
The number you care about
Sales, units, sign-ups, store visits. One column, the one the meeting is about.
sales units
What might have driven it
Spend by channel, price, a promo flag, and anything you suspect: weather, distribution,
a competitor. tv_spend price promo
Around a hundred rows. The Data Scout reads the file before any run and tells you what it can answer, what it would answer badly and why, and which one more column would open up.
How it works
-
Drop the file
CSV or Excel. The Data Scout maps the columns to their roles and says which analyses the file can carry.
-
Pick the model
Linear, log-linear, or with adstock and saturation built in. TEA explains the trade-off of each before you choose, not after.
-
Read the answer
Contribution by driver, the waterfall, the diagnostics. Export it, or take the next step: the saturation curves and the budget.
What marketing mix modeling is, in a paragraph
A marketing mix model, MMM for short, is a regression of a business outcome on the things that might have moved it: media spend by channel, promotions, price, seasonality and whatever controls you can name. It gives every driver its share of the outcome, period by period, and it gives you the baseline, the volume you would have had with nothing running at all. It works on aggregate data, needs no cookies and no user-level tracking, which is why it is the method that survived the end of third-party data.
It used to take a supplier six weeks because the data had to be collected, cleaned and argued over, and the model lived in a spreadsheet only one analyst could read. The maths was never the slow part. TEA keeps the maths and removes the rest: your file, read by the Data Scout, fitted in minutes, with the diagnostics on the same page as the result.
What is marketing mix modeling?
A regression of a business outcome on the things that might have moved it: media spend by channel, promotions, price, seasonality and controls. The result is a contribution per driver per period and a baseline, the volume you would have had anyway. It needs no cookies and no user-level data, which is why it survived the end of third-party tracking.
How much data do I need?
Weekly data for at least a year, two is better, with the spend by channel on the same rows as the sales. About a hundred rows is where seasonal claims start to hold. TEA reads the file before any run and tells you if it is too thin for the question.
How long does it take?
Minutes. The upload takes seconds, the Data Scout reads the file in under a minute, and a decomposition on a two-year weekly file fits in well under five. The six weeks were never the maths. They were the supplier.
Is it a black box?
No. You see the specification, the coefficients with their p-values, the fit quality and the residuals, and you can change the model and run it again. The export carries the same numbers as the screen, in the same colours.
What does it cost?
During the beta, nothing: every run is measured and no balance moves. After the beta, a plan with a monthly amount of compute. The plan never limits which questions you can ask.
Your sales, explained. This afternoon.
Drop the file, read the answer, take it to the meeting. Free while in beta, by invitation.
Join the beta