The science
The science powering our marketing model.
“This model is the result of a team effort that combines years in branding, growth marketing, and data scientists trained in econometrics modeling, plus more academic papers than I care to admit.”
The engine, up close
The five are one model, not five settings. All of them run for every question, on top of the marketing data you loaded.
What actually sold?
Reads every channel together and lands on the share of the sale each one actually caused
How long until it pays off?
Spreads one week's spend across the next ten
Does branding actually drive sales?
Traces brand into downstream demand
How far can I push a channel?
Finds where the curve flattens
Where does the next real go?
Points at the next real you spend
The base the engine runs on
Purple works on top of your own database, finding the patterns of what works: your spend, your sales, your calendar, your context.
How the model is built
Purple uses Bayesian time series to understand cause and effect relationships between marketing actions and results. This approach lets the model carry prior knowledge and handle the complex behaviour typical of branding efforts.
Purple is designed to capture relationships between variables, identifying not just the direct impact of an action but how it influences other channels before driving results. This is what reveals, for example, when OOH media drives organic searches that lead to conversions.
The model allows the inclusion of variables that generate no direct clicks: offline media, upper funnel digital channels, OOH, PR. These are modelled alongside other channels, making it possible to read their impact in an integrated way.
Purple was built to accept variables that do not have a direct investment value. This enables the model to include actions like your organic social strategy, reflecting a more holistic view of marketing.
Not all marketing impact happens instantly. The model captures effects that unfold over days or weeks, so you can allocate budget based on long term returns.
Purple's model does not rely on fixed assumptions about how long an action takes to deliver results. Instead it looks at your brand's actual data to determine the real timing of each variable's impact, unlike many traditional models.
Why it took years
It took us a few years to get to the Purple Metrics model.
Most marketing models are built for media buyers. They're designed to match money in with money out, usually media and sales. That kind of direct connection leaves brand efforts out of the picture.
Here's how we approached it differently.
The founder who runs product worked at top branding agencies, then built a branding company for startups. She saw up close how brands are built and how they drive growth. Purple Metrics started as brand research software, which gave us primary data on branding and consumer behavior.
When we pivoted to modeling, our data scientists dove into marketing measurement. But every early model leaned heavily toward performance. That's where the long discussions began, crossing branding experience with growth marketing logic. All of this alongside a cofounder who spent nearly a decade at Google.
The data science team went deeper. Not just into marketing, but into economics (their original field), benchmarking across industries, and even drawing ideas from how bacteria and viruses behave. Don't ask.
Eventually, we figured out the adjustments needed to add branding to a marketing model. And we started seeing it.
Then we picked 10 paying clients to test with, chosen for the quality of their data, their industry mix, and their business models. We weren't after scale. We wanted sharp feedback and strong datasets. All of them had CMOs who were data oriented and up for the challenge.
We trained the algorithm, relaunched the model a bunch of times, and kept refining it until we could trust what it showed: contribution and reallocation that finally includes branding.
Yes: it sees branding.
Then it had to be feasible. Pulling API data, organizing a clean data lake, orchestrating the model, and turning that into a simple dashboard. That's our cofounder and CTO making the magic happen.
Imagine having a top data science team inside your marketing department, focused entirely on building a model that reflects how marketing actually works. Built by people who come from branding and growth. Trained on real data from real teams. Designed so clearly that you actually know what to do with it.
It doesn't feel like it was designed only by engineers. Because it wasn't.
That's Purple Metrics.
Sources
Alchemy: The Dark Art and Curious Science of Creating Magic in Brands, Business, and Life
Rory Sutherland
Beyond the pair: media archetypes and complex channel synergies in advertising
J. Jason Bell, Felipe Thomaz and Andrew T. Stephen
How Brands Grow
Byron Sharp
The Long and the Short of It
Les Binet and Peter Field
How econometrics is shaping the future of measurement
Les Binet
The impact of brand equity on brand preference and purchase intentions in the service industries
Hsin Hsin Chang and Ya Ming Liu
The Chain of Effects from Brand Trust and Brand Affect to Brand Performance: The Role of Brand Loyalty
Arjun Chaudhuri and Morris B. Holbrook
Brand equity, risk and return in Latin America
Marta Olivia Rovedder de Oliveira, Aline Armanini Stefanan and Mauri Leodir Lobler
The Impact of Brand Equity on Customer Acquisition, Retention, and Profit Margin
Florian Stahl, Mark Heitmann, Donald R. Lehmann and Scott A. Neslin
The Financial Value Impact of Perceptual Brand Attributes
Natalie Mizik and Robert Jacobson
The financial performance of the most valuable brands: A global empirical investigation
Gregor Dorfleitner, Felix Roble and Kathrin Lesser
Double Jeopardy Revisited
Andrew S. C. Ehrenberg, Gerald J. Goodhardt and T. Patrick Barwise
Brand Awareness Effects on Consumer Decision Making for a Common, Repeat Purchase Product: A Replication
Emma K Macdonald and Byron Sharp
Perceptional components of brand equity: Configuring the Symmetrical and Asymmetrical Paths to brand loyalty and brand purchase intention
Pantea Foroudi, Zhongqi Jin, Suraksha Gupta, Mohammad M. Foroudi, Philip J. Kitchen
Bring the question you can't answer today and we'll go through it together: your goal, your data, and what it would take to prove it.