For each stage: what goes in, what gets done, what comes out. Preparation activities sit before the eight-week clock; the build stages all run within it; the presentation cycle follows.
Preparation · before the build
Variable timeline. Ends with Design ✓ and Data ✓, which together start the eight-week build clock.
·
Data AssemblyPreparation · Local Team · MathCo support
Local teams consolidate the data they hold across their internal systems and any third-party sources in scope for the brand × country × category. The consolidated package arrives with MathCo either through the data lake or as flat-file handovers. Volume, formats and refresh cadence vary by market.
In brand × country × category scope · sponsor
Out consolidated data package landed with MathCo
·
Summaries & DiagnosticsPreparation · MathCo
MathCo prepares data summaries on what's been handed over — coverage, granularity, variable presence, time-period continuity. First-pass diagnostics surface what's missing, what's too coarse, and where time-period gaps will constrain the model. These summaries become the substrate for the design conversation and for the data-side decisions that need to be made before the build can start.
In consolidated data package
Out data summaries · diagnostics note · open-questions list
·
Joint Design CallsPreparation · Joint · Cell Lead facilitates
The model-design conversation, run jointly with the local team. Topics include: number of models the cell needs, the level of data (national / regional / SKU), how brand vs. sub-brand vs. variant gets treated, and the granularity at which media campaigns enter the model. Permutations and combinations get walked through; trade-offs around data availability and analytical value get debated. Iteration is normal — the summaries refresh, the design firms up, the local team confirms.
In data summaries · diagnostics · local team priorities
Out agreed design · variable list · model framework · PnP scope flag
·
Design ✓ + Data ✓ Sign-offsPreparation gate · Joint
The phase closes with two sign-offs. Data ✓ means the local team has signed off on the dataset that will go into the build — its scope, its granularity, the explicitly-accepted gaps. Design ✓ means the modelling approach is locked. Both gates need to be in before the eight-week build window opens. Once they are, the clock starts.
In agreed design · agreed dataset scope
Out two formal sign-offs · build window opens
8-Week Build · committed delivery window
From sign-offs to "ready for presentation" in eight weeks. Every stage below runs inside this window.
02
EDA & ADS prepBuild · Stage Lead — Gopinath V
The first stage inside the build window. Exploratory data analysis on the signed-off dataset, followed by construction of the analytical dataset (ADS) that the model will actually consume — joining, normalising, deriving features, handling the agreed-on gaps. This is not the data-acquisition stage (that lives in Preparation) — it's the analytical preparation that turns the agreed dataset into something model-ready.
In signed-off dataset · agreed design
Out EDA findings · model-ready ADS
03
MMMBuild · Cell Lead
Fitting the marketing-mix model on the prepared dataset. The technical core: decomposing volume / value into base, media, price, promotion and external drivers; calibrating ad-stock and saturation; back-testing against holdout periods. Iterative — a poor fit may send the cell back to EDA & ADS prep for additional or cleaner inputs.
In model-ready ADS · agreed design
Out fitted MMM · model card · response curves · contribution decomposition
04
PnPBuild · where scoped · Cell Lead
Price-and-promotion modelling. Calibrates own-price elasticity, cross-elasticity (within and across categories), promotional response curves and base-vs-incremental decomposition for promotions. Runs in parallel with MMM for cells where it was scoped at design — not every cell needs it (some markets are media-only, some don't have the trade data to support it). Where present, its outputs feed straight into the optimizer alongside the MMM.
In model-ready trade ADS · agreed design (PnP scope flag)
Out elasticity matrix · promo-response curves · combined model spec
05
OptimizationBuild · Stage Lead — Gopinath V
The "what-if" engine. Takes the fitted model(s) — MMM, plus PnP where present — and produces optimal allocations under client-set constraints: budget envelopes, channel minimums and maximums, SKU-mix rules, regional splits. The numbers that end up in the recommendation come from here.
In fitted MMM (+ PnP) · constraint set
Out optimal media plan · scenario set · sensitivity analysis
06
ValidationBuild · Stage Lead — Sibasis Dash
The independent sense-check on both the fitted model and the optimization output. Decompositions checked against priors and prior-wave findings; elasticities and ROIs sanity-checked against business intuition; optimizer scenarios pressure-tested for plausibility. Not a rubber stamp — Validation can send the cell back upstream for a re-fit. It catches the implausible before it ever gets in front of a client.
In fitted model · optimizer outputs · prior-wave benchmarks
Out validation note · go / no-go · re-fit asks where applicable
07
Insights & StoryboardingBuild · Cell Lead · includes deck making
Where validated outputs become a deliverable. Three activities, run in sequence inside one stage — the Cell Lead is back in the driver's seat for all three. The build window closes here — "ready for presentation."
Insights — translating the validated stats into business meaning. ROIs, elasticities and decay parameters become "TV's ad-stock decay is short on this brand — frontload bursts rather than sustain." Cross-cell pattern-spotting surfaces here.
Storyboarding — the deck spine. The narrative arc: what the recommendation is, why it's defensible, the three lead messages, what gets called out vs. relegated to appendix, what objections to anticipate. Done on paper or whiteboard before any slide is built.
Deck making — the artefact. PowerPoint construction following the storyboard — house style, exhibits pulled from the optimizer, executive summary, defendable charts, heavy QC pass on numbers and message coherence. Two cuts usually exist by the end: a fuller business deck and a tighter market read-out, sharing the same spine.
In validated outputs · prior-wave findings · audience profile
Out insight register · storyboard · business deck · market deck
Presentation Cycle · after the build
Timing externally driven — slots are scheduled at market convenience. The build artefacts are ready; sign-off lands when the cycle completes.
08
Business PresentationPresentation · Cell Lead · internal-to-Unilever
The first read-out — a call or meeting with the Unilever business team, typically global Marketing Effectiveness and category leads. Stress-test ahead of taking the recommendation to market: catches framing issues, calibrates which messages will land vs. need defending, surfaces objections we'll face from the local team. Outcome: green-light to proceed to the market call, or a re-cut request on Insights & Storyboarding (most often a tighter narrative, occasionally a re-fit ask back to MMM / Optimization).
In business deck
Out business team's go-ahead · re-cut asks (where applicable)
09
Market PresentationPresentation · Cell Lead · local market call
The second read-out — a call or meeting with the local market team, the operating client who will act on the recommendation. Recommendations land here. Objections, change requests and "we already tried that" feedback flow back upstream — most often resolved live, occasionally triggering a re-cut on Insights & Storyboarding or, rarely, a re-fit on MMM / Optimization before sign-off can be given.
In market deck
Out market team's intent to accept (or re-cut / re-fit ask)
10
Final Sign-offPresentation · terminal state · Cell Lead → Sponsor
Formal acceptance. The market team signals (in writing) that they are taking the recommendation. Cell closes if it's a one-shot, or moves into support if there's a refresh cadence in scope. The protocol ends here — anything that follows (refresh runs, post-implementation tracking) is governed by a fresh cycle of the same protocol on the next wave.
In market team's go-ahead · meeting notes
Out signed acceptance · cell closes or moves to support