The attribution model debate at growth stage often becomes a distraction. Last-click, first-click, linear, time-decay, data-driven, Meridian, Robyn, Lightweight MMM — the options generate more confusion than clarity. Most of the time, the model is not the problem. The data is the problem. The right question is not which model to adopt but what question you are actually trying to answer and whether you have the underlying data quality to trust the output of any model.

What multi-touch attribution actually measures — and where it breaks

Multi-touch attribution assigns fractional credit for a conversion across the touchpoints a user interacted with before converting. A user who clicked a Meta ad, received an email, then searched your brand name might get credit split proportionally across all three interactions. MTA requires individual-level tracking: a persistent user identifier that connects sessions across touchpoints and time. GA4's client_id serves this purpose for web sessions, and Segment or similar CDPs can extend it across channels. The fundamental limitation of MTA is that it only measures what it can track. iOS ATT opt-outs, Safari ITP restrictions on JavaScript cookies, and ad blockers all create gaps in the touchpoint sequence. MTA with incomplete tracking paths produces attribution that looks precise but is systematically biased toward the channels that happen to survive the tracking gaps — typically last-click channels like branded search and direct.

What marketing mix modelling actually measures — and where it breaks

Marketing mix modelling takes an aggregate statistical approach. It uses regression analysis to model the historical relationship between spend across channels and aggregate outcomes (revenue, leads, conversions) over time. Individual-level tracking is not required, which is why MMM has seen renewed interest as user-level tracking degrades. MMM can incorporate channels that have no pixel — TV, out-of-home, events, PR, word-of-mouth — which MTA cannot. The limitations are different: MMM requires substantial historical data (typically 2–3 years of weekly spend and outcome data to produce stable estimates), produces channel-level estimates rather than campaign-level insights, and cannot tell you which specific creative or audience drove performance within a channel. It is a strategic budget allocation tool, not a campaign optimisation tool.

Which one growth-stage companies actually need

At growth stage — call it ₹5 crore to ₹100 crore ARR — you are running three to five digital channels at most. Your measurement problem is not strategic allocation across fifteen channels including TV and out-of-home. Your problem is: are my Meta campaigns generating leads that turn into revenue, and which campaigns within Meta are doing the work? This is an MTA question, not an MMM question. The correct sequence: build a clean GTM → GA4 → CRM → Looker attribution chain first. Get to a state where you can answer 'which campaign sourced which closed deal' with reasonable confidence. Run incrementality holdout tests on channels where budget is large enough to matter. MMM becomes the right tool when you have offline channels with meaningful spend, offline events, brand campaigns without trackable clicks, or channels where individual pixels simply do not exist.

The geo holdout test that works at any stage

The method that bridges MTA rigour and MMM accessibility is geo holdout testing. Divide your target market into test and control regions — Indian states work well for this, or tier-1 versus tier-2 cities if the volume justifies the split. Run your campaigns normally in test regions. Withhold the specific channel you are testing from control regions for 4–6 weeks. Measure whether conversion rates differ between exposed and unexposed regions. The gap is your incremental lift. This does not require individual-level tracking and does not require historical data for statistical modelling. It gives you channel-level incrementality without the complexity of either full MTA or full MMM. For most growth-stage companies, a well-run geo holdout test on each major channel provides more actionable information than either attribution model deployed on top of incomplete tracking data.