The iOS 19 privacy changes that landed late last year have forced another rebuild cycle for marketing mix modelling at most serious advertisers. The signal degradation in digital channels has been substantial enough that MMM models that worked acceptably through 2025 are producing increasingly unreliable channel attribution in 2026.

This is a working view of what’s actually required to maintain MMM reliability in the current environment, drawn from work with several mid-sized to large Australian advertisers through the first half of 2026.

What iOS 19 Actually Changed

The iOS 19 changes built on the trajectory of previous Apple privacy releases but extended several restrictions in ways that further degraded digital marketing signal:

Cross-context tracking that had survived previous restrictions through workarounds has become substantially harder. The technical mechanisms that some advertisers relied on have either been blocked or significantly degraded.

Conversion API implementations that were working effectively in late 2025 have seen meaningful drops in match rates and conversion fidelity through early 2026.

The default privacy settings have been tightened further, with the effect that the share of users with reduced tracking has grown.

The third-party measurement provider workflows that the major ad platforms had developed have been affected, with various platforms experiencing meaningful disruption to their measurement workflows.

The cumulative effect is that digital channel measurement that worked reasonably in 2025 has become noticeably less reliable in 2026. The signal degradation isn’t dramatic in any single dimension but the compound effect is material.

What This Means for MMM

The implications for marketing mix modelling have been significant:

The historical data used for model training contains transitions that affect model accuracy. Models trained on data spanning the iOS 19 transition period need careful handling to avoid distortions.

The channel-level coefficients that had stabilised in 2025 models are shifting again as the underlying signal quality changes. Models that aren’t being refreshed are producing increasingly unreliable channel attribution.

The relative attribution between digital and non-digital channels has shifted, often in ways that don’t reflect actual marketing effectiveness but rather measurement artefacts.

The interaction effects between channels — particularly the digital-traditional interactions — need re-validation in the current data environment.

The forward projections from MMM models for budget planning need explicit attention to which assumptions hold and which need to be relaxed.

The Rebuild Activities

The MMM rebuilds happening across Australian advertisers in 2026 share several common activities:

Data quality assessment and gap analysis. Understanding which data inputs have been affected by iOS 19 changes, where signal quality has degraded, and what compensating data sources can fill the gaps.

Model architecture revisions. The model structures that handled the pre-iOS 19 data environment may need modification to handle the current data quality patterns reliably.

Incorporation of non-digital signals. The relative reliability of non-digital signals — sales data, distribution data, brand surveys, market research — has improved compared to digital signals. Models that lean more on these signals are producing more reliable outputs.

Integration with first-party data more deeply. The first-party data assets that advertisers control directly have become more central to measurement. The integration depth required for this is substantial.

Validation against ground truth where possible. The use of randomised controlled experiments, geo-experiments, and other ground truth validation has become more important for model calibration.

Stakeholder education on uncertainty. The reasonable confidence levels for MMM outputs have shifted. Stakeholders making decisions based on MMM outputs need to understand the current uncertainty environment.

What’s Working Well

Several adaptations are producing genuine improvements in measurement reliability:

Geo-experiments for ground truth. The geographic randomisation experiments that some advertisers can run produce reliable channel effect measurements that calibrate MMM outputs effectively.

Bayesian MMM frameworks that explicitly model uncertainty. The models that produce explicit uncertainty estimates alongside point estimates are more useful in the current environment than point-estimate-only models.

Increased frequency of model refresh. Models refreshed monthly or quarterly are tracking the current environment better than annually-refreshed models.

Integration of third-party data partnerships for cross-validation. The use of data partnerships to validate model outputs against external benchmarks has improved confidence.

First-party measurement infrastructure investment. Advertisers with strong first-party measurement infrastructure are less affected by the third-party signal degradation.

What’s Not Working Well

A few approaches that aren’t producing the results that advertisers might hope for:

Trying to compensate for digital signal loss through more aggressive digital tracking workarounds. The risk of regulatory or platform action against workarounds increasingly exceeds the benefit.

Relying on platform-provided attribution metrics as substitutes for proper MMM. The platform attribution is increasingly self-serving in ways that distort budget decisions.

Treating MMM as a settled methodology rather than an evolving practice. Advertisers who built MMM capability several years ago and treat it as stable are seeing degrading reliability they may not fully recognise.

Avoiding investment in MMM capability because the digital attribution feels easier. The illusion of precise digital attribution has been more damaging than the acknowledged uncertainty of properly built MMM.

The Talent and Capability Side

The MMM rebuild work requires capability that many marketing analytics teams don’t have in-house. The combination of statistical modelling, marketing domain knowledge, data engineering, and business communication needed to do this work well is uncommon.

Some advertisers have built internal capability through hiring or upskilling. Others have engaged specialist agencies or consultancies. The right model depends on the scale of the marketing investment and the centrality of measurement to the business.

For advertisers building internal capability, the investment is substantial but pays off across multiple capabilities beyond MMM specifically. The data engineering, statistical analysis, and measurement design skills support broader analytics maturity.

For advertisers engaging external partners, the choice of partner matters substantially. The market has consultancies with deep MMM specialisation, agencies with measurement teams of varying capability, and specialist boutiques with technical depth. The right choice depends on the scope of work and the integration with broader marketing analytics activity. Some of the more complex builds involve partners like Team400 on the data engineering and platform side combined with specialist measurement consultancies on the statistical modelling.

The Stakeholder Communication Challenge

A persistent challenge in MMM work is communicating with stakeholders whose preferences favour precise attribution over honest uncertainty estimates. The stakeholder culture that grew up around digital marketing’s apparent precision struggles with MMM’s explicit uncertainty.

The marketing analytics teams handling this well are doing structured education work — explaining what MMM actually measures, what the uncertainty estimates mean, and how to make decisions in the presence of uncertainty. This is patient work that has to be repeated.

The marketing analytics teams handling this badly are sometimes providing point estimates without uncertainty context, which reproduces the illusion of precision that misled previous decision cycles. The short-term comfort produces longer-term problems.

What’s Coming

Several factors will continue to shape MMM practice over the coming year:

Continued regulatory and platform changes affecting digital measurement. The iOS 19 changes won’t be the last. Models and capability that can adapt to ongoing change have strategic value.

The further maturation of privacy-preserving measurement techniques. Various technical approaches to measurement that preserve user privacy while supporting business measurement are progressing.

The evolution of platform-provided measurement and attribution tools. The major ad platforms continue to invest in their measurement offerings, with varying degrees of usefulness for independent decision-making.

Continued development of MMM-specific tooling and platforms that reduce the heavy custom development that current implementations often require.

The integration of MMM outputs with operational marketing technology stacks for more direct decision support.

The Mid-2026 Position

Marketing mix modelling in 2026 is harder to do well than it was a year ago. The signal environment is more degraded. The model accuracy is more uncertain. The skill required to do this work effectively is real and not always available.

What hasn’t changed is that MMM remains the most reliable available approach to channel attribution and budget optimisation for advertisers with meaningful marketing investment. The alternatives — pure platform attribution, last-click metrics, marketing intuition — have all become more visibly unreliable than they were.

For marketing leaders making decisions about MMM investment in the current environment, the practical position is that the investment case has strengthened despite the work being harder. The marketing decisions being made — increasingly significant budget allocations across complex channel mixes — require the best measurement methodology available, and MMM remains that methodology when done properly.

The next year will probably bring continued challenges as the measurement environment continues to evolve. The advertisers who maintain serious investment in measurement capability will be making better marketing decisions than those who don’t. The discipline required is real but the alternative — making large marketing decisions on increasingly unreliable signal — is worse.