The persistent question in marketing analytics — multi-touch attribution or marketing mix modelling — has shifted ground meaningfully through 2024-2026. The privacy landscape has continued to erode the underlying data foundations of traditional MTA. The accessibility of MMM has improved through better tools and better-trained practitioners. The serious marketing teams have largely landed on hybrid approaches that use each methodology for what it’s actually good at.

A practical mid-2026 read on the state of attribution.

Where MTA actually still works

The deterministic, user-level identification that MTA relies on remains intact in specific contexts.

Authenticated environments. Within logged-in environments — owned platforms, paid customer relationships, authenticated mobile apps — the underlying attribution data is still rich. MTA in these contexts continues to deliver meaningful insight about user journey patterns, content effectiveness, and conversion path optimisation.

B2B sales contexts. For B2B sales motions with long buying journeys and identifiable account-level engagement, the attribution data flowing through CRM systems remains workable. The journey across paid media, content engagement, sales touches, and eventual deal closure can still be modelled with reasonable fidelity.

Direct-response digital channels where conversion happens on platform. Paid social conversions that complete on the same platform, paid search conversions through tracked landing pages, retargeting flows with completed conversion events — the closed-loop measurement here remains useful even with the privacy framework changes.

Where MTA has fundamentally broken

The areas where MTA was always strong are also where it’s been most affected by privacy changes.

Cross-platform consumer journeys. The traditional MTA promise — tracking a user across initial awareness on one platform, consideration on another, and conversion on a third — is fundamentally broken for unauthenticated consumer journeys. The signal loss is real, the modelled gap-filling that platforms apply introduces meaningful uncertainty, and the resulting attribution is generally less reliable than buyers want to admit.

Third-party measurement of paid media. The third-party MTA platforms that built businesses on tracking conversions across walled gardens have had to rebuild their methodologies. Some have done so credibly with conversion API integrations, server-side tagging, and identity graph approaches; others have effectively given up on the third-party measurement model.

View-through measurement. Attribution credit for ad impressions that didn’t trigger a click but plausibly contributed to conversion has become increasingly difficult to measure cleanly. The methodologies that remain are essentially modelled rather than measured.

Where MMM has stepped up

Marketing mix modelling has had a renaissance through 2024-2026, reflecting both the limitations of MTA and improvements in MMM tooling and practice.

Better tooling. The open-source releases (notably Meta’s Robyn and Google’s Meridian) and the proliferation of commercial MMM tools have made the methodology more accessible. Marketing teams that previously couldn’t afford traditional MMM engagements can now run credible models with appropriate methodology training.

Better data. The discipline of constructing clean, granular media-spend-and-outcome data sets has improved. The data engineering work that supports good MMM has become more standardised, and the data quality going into models has improved.

Better integration with planning. MMM outputs are increasingly being used not just for retrospective measurement but for prospective planning — informing budget allocation, channel mix decisions, and campaign-level investment levels.

Better treatment of uncertainty. The honest MMM practice acknowledges meaningful uncertainty in the estimates and uses that uncertainty in decision-making rather than treating the point estimates as ground truth. Bayesian methodologies have moved from research to operational practice in many teams.

The hybrid approach that works

The serious marketing teams in mid-2026 have largely converged on hybrid approaches. The shape that’s working:

MMM as the strategic frame. The medium-to-long-term view of channel effectiveness, budget allocation, and incrementality of paid media. MMM informs the broad strokes of how the marketing investment is structured.

MTA for tactical optimisation. Within the channels where deterministic measurement remains workable, MTA informs the tactical decisions — creative selection, audience segmentation, bidding optimisation. The role is tactical, not strategic.

Incrementality testing for validation. Geographic holdout tests, time-period holdouts, structured incrementality experiments. These provide ground-truth validation of both MMM and MTA outputs and inform calibration of the broader measurement framework.

First-party data infrastructure as the foundation. The marketing teams that have invested in clean first-party data — proper CDP implementations, server-side tagging, well-structured customer data architecture — have meaningfully better outcomes than those still relying on browser-side tracking infrastructure.

What’s overhyped

A few areas where the discourse is ahead of the reality.

“AI-powered attribution platforms.” The marketing for these platforms suggests that AI/ML methodologies can recover what privacy changes have taken away. The methodologies are real but the limitations of inference from insufficient data are also real. AI doesn’t manufacture signal that wasn’t captured.

Pure-MMM nirvana. Some commentary suggests MMM has effectively replaced MTA. It hasn’t, and won’t. MMM is excellent at certain questions and unsuited to others. The teams that have abandoned MTA entirely have given up useful capability.

Universal identity resolution. The various identity resolution initiatives have their place but the practical reality is that universal cross-platform identity is not coming back in the consumer space. Authenticated relationships will continue to provide deterministic measurement; the broader unauthenticated landscape will rely on probabilistic and modelled approaches.

What I’d build right now

For marketing teams designing or refreshing their measurement framework in mid-2026, the practical shape:

A clean first-party data foundation — CDP, server-side tagging, structured customer data architecture, proper consent management.

MMM capability — either in-house using the open-source tooling and appropriately-skilled practitioners, or through external partners with credible methodology.

MTA where it works — authenticated environments, specific deterministic channels, B2B sales journeys.

Incrementality testing capability — geo holdouts, time-period holdouts, structured experiments. This validates the broader framework.

Clear governance of how the different measurement signals inform different decisions — strategic budget allocation, tactical channel optimisation, creative testing, campaign-level investment.

The marketing attribution landscape in mid-2026 is more complex than it was five years ago but the serious practitioners have meaningfully better methodology and meaningfully better outcomes than the practitioners chasing platform-native single-source-of-truth metrics. The complexity is the cost of operating in a world with real privacy constraints; the practitioners who’ve absorbed that cost are getting more credible answers than they were getting in 2020.