Look-alikes are obsolete
By Alain Briançon, Vice President of Research and Data Science, Dynata
For years, the insights industry has relied on demographic proxies as if they were predictive truth. Look-alike audiences were efficient, inexpensive and comfortable. They powered media plans, digital targeting and even parts of measurement. But they now restrain more than they enable.
The market has outgrown them. Consumers move across screens, platforms and contexts at a pace that no static profile can track. Their motivations evolve quickly. Their behaviors adapt instantly. And the data signals we depend on are too dynamic – and too complex – to be compressed into demographic buckets built for another era.
The conclusion is unavoidable: Look-alikes are no longer aligned with the world we measure. It is time to move on.
The problem with look-alikes
Demographics are not predictive models. They are shortcuts. And today those shortcuts create unnecessary inaccuracies at the exact moment the industry is under pressure to deliver sharper guidance, tighter confidence intervals and more accountable insights.
Look-alikes flatten nuance. Millions of individuals reduced to broad bins inevitably lead to overgeneralization and underperformance.
Look-alikes age poorly. They operate at the speed of census updates while the real world shifts by the hour.
Look-alikes erase context. People are treated as isolated entries instead of connected participants in a larger structure of influence, exposure and attitude.
In information-theory terms, look-alikes introduce avoidable uncertainty. We knowingly discard structure and then attempt to rebuild it later through modeling, weighting or heroic assumptions. It is an inefficient, backward workflow.
Moving to behave-alike: progress with limits
Behavioral modeling was the industry’s first attempt to break out of demographic gravity. It gave us trajectories, patterns and richer signals. But behavior alone cannot answer the most important question: Why?
Two consumers may take the same action with entirely different intentions behind it. Without attitudinal context, behavioral similarity still leaves prediction incomplete. Behave-alike is necessary but it is not the finish line.
The attitude-alike advantage
Attitudes – beliefs, motivations, values, expectations – are the most stable, most interpretable and most predictive components of decision-making. They give meaning to behavior and allow us to anticipate future actions with higher confidence.
The challenge is scale. Attitudes cannot be passively scraped or continuously observed. They must be measured, validated and responsibly extended. A conventional dataset cannot do this. A graph can.
A graph is the only path forward. A graph allows us to understand individuals the way they actually exist: as connected entities, not isolated rows.
A graph captures structure. Sequence, similarity, influence and context are preserved as first-class features.
A graph enables responsible inference. Attitudinal signals can be propagated to structurally similar nodes without violating privacy or precision.
A graph reduces uncertainty. Instead of throwing away information early, we retain it – and use it intentionally.
A graph aligns with the ecosystem we actually operate in. Identity is fragmented. Exposure paths are multichannel. Behavior is continuous. A graph architecture is built for this complexity.
This is not an incremental improvement. This is a structural shift in how insight systems function.
A practical roadmap for the industry
Moving toward attitude-alike systems follows a clear progression:
Identity becomes relational, not stitched. We stop forcing deterministic matches and start understanding structural similarity.
Exposure becomes sequence-aware. CTV, social, in-feed video, display and retail media become connected pathways, not siloed impressions.
Behavior becomes encoded. Embeddings capture behavioral patterns cleanly and efficiently.
Attitudes become anchors. They provide stability and meaning and the graph carries them where direct measurement is impossible.
Synthetic augmentation adopts structural guardrails. Quality is measured through divergence, fidelity and attitudinal consistency – not just point estimates.
Measurement becomes more realistic. Brand lift, reach and incrementality reflect how influence actually moves across attitudinal communities.
This is the blueprint for the next decade of insights leadership.
The industry wake-up call
Organizations that cling to look-alikes will continue to misread audiences, misinterpret campaign outcomes and misallocate budgets. Their models will produce inconsistent signals. Their predictions will drift. Their measurement will lack the structural fidelity needed for modern media.
The industry is outgrowing its old tools. Precision requires structure. Prediction requires attitude. And both require a graph.
The final word
The future of insights will belong to companies willing to model people the way they actually live: connected, contextual and dynamic. Look-alikes were a starting point. Behave-alike moved us closer. Attitude-alike is where real prediction begins.
And if you’re ready for that future, you can belong to my graph on LinkedIn.