Tattva Docs
Use casesPerformance marketing

Channel budget scenarios

"What if I move 10% from Meta to Google?" — modelled revenue, CM3 and 6-month repeat-retention impact, in seconds.

The question, verbatim: "What if I increased 10% budget on meta? … 5% goes here and there, max I can take 2%, and hence the impact will be 8% on Revenue and 3% on CM3. This answer he wants."

"How do I become 30% less Meta-dependent?"

"If I scale by 30%, how is my M3 looking after six months?"

These are the forecasting questions — the ones the perf leader can't answer manually because they require modelling thousands of scenarios. The platform's insight agent does this with a multimedia mix model (MMM) layered on top of the team's BigQuery data.

Who asks

  • Perf-marketing leader, in monthly planning
  • Leadership during quarterly strategy
  • Founder during board-prep season

Frequency

Monthly during planning. Spike during quarterly strategy and budget revisions.

Data you need

  • Daily spend by channel × campaign type × product
  • Daily revenue (new + repeat) by channel × product
  • Customer table with first-purchase date (drives cohort retention)
  • Historical CM3 by acquisition channel and product
  • Time-series of incrementality experiments where available (Meta CBO tests, geo-holdout tests)

Honest disclaimer: without explicit incrementality experiments, MMM is directional, not surgical. The platform will say so. Use these scenarios as decision aides, not as oracle truth.

How to ask it

Level 1 — generic scenario:

  • "What if I increase Meta spend by 10%? Forecast revenue, CAC, and CM3 impact at 1m, 3m and 6m horizons."

Level 2 — specific reallocation:

  • "What if I move ₹50L/month from Meta acquisition to Google brand? Forecast impact and confidence intervals."

Level 3 — strategic / Meta-dependence:

  • "We're at X% Meta dependence. Show me three pathways to reduce that to (X-30)% over 6 months, with revenue and CM3 trade-offs for each."

Level 4 — cohort retention scenarios:

  • "If I scale acquisition by 30% this quarter, what's the expected M3 and M6 repeat-retention curve based on current cohort behaviour?"

What you'll get back

For each scenario:

  1. Headline — projected revenue delta, CAC delta, CM3 delta with confidence band.
  2. By time horizon — point estimates at 1, 3, 6 months.
  3. By cohort — how new-customer M1/M3/M6 retention plays through to LTV.
  4. Assumptions called out — channel saturation curves used, retention shape assumed, what's modelled vs guessed.
  5. Where the model is least confident — usually any move > 20% from current spend on a channel, or any reallocation that crosses incrementality-untested territory.

The strategic question — "less Meta-dependent"

This question deserves its own treatment because it's not a number — it's a strategy. The platform produces:

  • Three to four pathways (e.g. "scale Google brand", "stand up Amazon DSP", "build content-led organic", "test Snap / influencer-led").
  • For each pathway: required investment, time to break even, sensitivity to Meta's response (does cost rise as supply tightens?), and the assumption set.
  • A side-by-side trade-off matrix so you can read the choice in one screen.

This is where Tattva earns its keep — answering a question that's genuinely impossible to answer manually because it requires modelling thousands of scenarios.

The logic tree

Channel reallocation question?
├── Pure budget delta on one channel? → MMM-driven point forecast
├── Reallocation across channels? → MMM + retention-cohort overlay
├── Strategic dependence reduction? → multi-pathway scenario set with trade-off matrix
└── Compound (e.g. "scale Google AND reduce Meta")? → run both legs, compose, surface conflicts

How to make it recurring

Most scenario questions are one-off. But these are worth scheduling:

  • Monthly: A "scenarios I considered last month — how did the actual data compare to the forecast?" run, so you calibrate trust in the forecasts over time.
  • Quarterly: A regenerated "current strategy options" run during planning.

Save the recurring run as a report, not a tracker — these aren't pass/fail alerts; they're material for decision meetings.

Pitfalls

  • Treating point forecasts as exact. Look at the confidence band. A 10% reallocation might be "+5% to +15% revenue, 80% confidence" — that's the answer, not "+10%."
  • Asking impossible questions. "What if I 5x Meta spend?" exceeds anything in the historical data. The model will warn you; respect that.
  • Skipping incrementality. MMM without periodic incrementality experiments drifts. Plan one geo-holdout or CBO test per quarter and feed the result back in.
  • Confusing forecast with promise. This is decision-support, not commitment.

Audience overlap diagnosis

Are two ad sets eating each other's new-customer reach? The most common deep-dive within an NRoAS drop.

Product-lag RCA

"Shilajit is -30% behind target. Why?" — the most-asked weekly RCA, with the 70%-coverage logic tree.

On this page

Who asksFrequencyData you needHow to ask itWhat you'll get backThe strategic question — "less Meta-dependent"The logic treeHow to make it recurringPitfallsRelated