A bottle of wine poses an awkward question for an advertising model. When should you make it? At Josh Cellars, that question reaches years into the future. A campaign can stimulate demand long before every resulting case leaves a shop. If marketing, finance and production disagree about when those sales will arrive, a successful advertisement can still leave a business badly prepared.
- The job: estimate which investments generate additional sales.
- The machinery: marketing mix models, attribution, testing and Activate software.
- The customer: large companies coordinating brands, markets and budgets.
- The catch: a forecast becomes useful when people can act on it.
The bottle before the billboard
Ipsos MMA’s account of its Josh Cellars engagement starts with a familiar constraint: distribution expansion was reaching its limits. Growth increasingly required stronger consumer demand. Beginning in 2018, the partnership connected marketing measurement to planning across the business. Annual scenario sessions included finance, sales and agency teams; weekly media schedules followed. The company reports a 167% improvement in above-the-line profit ROI over more than seven years.
The telling detail is the meeting. People responsible for different parts of the business had to agree on what the forecast meant. A predicted sale has consequences for cash, inventory and execution. Measurement gave those consequences a common starting point.
“Quantifying the long-term impact of marketing with confidence enabled us to build a growth strategy”
Dan Kleinman, Josh Cellars / company-published account
Who gets credit for the sale?
MMA, short for Marketing Management Analytics, has been working on that larger problem since 1989, the founding year given in its official history. It was an early commercial practitioner of marketing mix modeling. Ipsos completed its acquisition in 2018, bringing the analytics business into a research group with consumer and brand expertise.
A marketing mix model asks how sales respond to advertising alongside prices, promotions, competitors and economic conditions. Econometric analysis estimates the contribution of those drivers. The distinction matters: a customer who clicks an advertisement may already have intended to buy. Counting that purchase tells you something different from estimating the additional business the advertisement produced.
There is a useful discipline hidden in that definition. A sales increase is an observation; its cause is a separate question. If a price cut and a television campaign happen together, assigning all the improvement to either one would be a consequential shortcut. A planning model has to make those relationships explicit enough to question.
MMA connects this strategic view with Agile Attribution, its more detailed campaign measurement. The company says the latter does not depend on individual-level tracking. The ambition is to connect a budget decision to choices about audiences, creative and placements, without letting the two levels of analysis drift apart.
Where should the budget go?
Marketing mix modeling considers media alongside pricing, operations, competition and market conditions.
A car sale is a long conversation
Volvo supplies a useful example. A car purchase unfolds through research, comparisons and dealer visits. A channel-level answer cannot settle every question about the advertisements encountered along the way. In MMA’s published case study, standard modeling lacked the detail Volvo needed. MMA and Meta’s Marketing Science team examined campaign choices against several stages of the buying journey.
The comparisons were specific. Purchase-objective campaigns showed five times the return on advertising spend of campaigns optimized toward alternative targets. Particular brand-campaign settings and platform-adapted creative also performed better against their respective comparison groups. These are reported findings within Volvo’s analysis. Treating the headline multiple as a prediction for another advertiser would miss the point: the useful output is a decision about a defined campaign lever.
Software with people attached
Activate is the product that turns analysis into planning. Teams can test spending scenarios, monitor performance and inspect model assumptions. MMA also provides data processing, harmonization and reporting infrastructure. There is a practical appeal to rehearsing a budget: you can compare possible allocations while the money is still available to move.
The business combines SaaS with consulting and ongoing analytical services. Its positioning is enterprise: complicated portfolios, multiple countries, several teams sharing a revenue target. Public customer examples include Nationwide, McCormick and Regeneron. Alternatives in the broader measurement market include Analytic Partners, Nielsen, Ekimetrics and Gain Theory. MMA’s pitch combines software, research context and help putting recommendations into practice.
That last element deserves attention. Its consulting work includes executive workshops, implementation and change management. A mathematically persuasive result still needs someone to revise the plan. The organization must agree who can make that change, when to make it and what evidence would justify reversing it.
Its internal training program offers a small clue to that mix of skills. MMA University covers analytics, business, industry and leadership. The selection makes sense: an analyst explaining a forecast needs to understand both the statistical reasoning and the decision facing the person across the table.
Make the forecast earn its place
At Nationwide, the measurement overhaul involved data challenges and internal adoption as well as analytical detail. MMA’s in-market testing offering addresses another requirement: checking an intervention before scaling it. Matched controls and impact measurement give a business a way to compare expectations with observed results.
For a buyer, that suggests a useful sequence. Establish what success would look like before reviewing a dashboard. Ask which assumptions could change the recommendation. Then identify a test with consequences small enough to learn from. Those questions give a model a job description, and give its users a way to judge the work.
The company is investing in that machinery. In June 2025, MMA reported 31% year-over-year business growth in the first quarter and described parent-funded NextGen work on automated data intake, AI-assisted analytics and planning. In February 2026, Susmita Chaudhury returned to lead its Bangalore Center of Excellence.
The lesson a reader can borrow is procedural. Pick a business outcome. Put the people who control the budget and the people who must deliver the forecast in the same conversation. Decide how to test the recommendation. Without usable data, validation or authority to change the plan, even an elaborate model has little practical room to work. The final question belongs to the business: what will you do differently on Monday?