Is GenAI proving to be a transformative tool for marketers working with data and analytics in 2026 - or is it another hurdle to surmount? In this piece, our expert commentators break down the challenges facing marketers and organisations who have set out to apply GenAI in data-driven decision-making, from why AI can't save poor-quality data to why AI projects aren't making the leap from experiment to rollout.
Additional 'elephants in the room' for data and measurement going into 2026 include fragmenting identity signals, reliance on legacy metrics and models, and the need to fix underlying data foundations rather than investing in fancy tools.
But it's not all doom and gloom. Our experts predict that 'T-shaped talent' and marketers who can straddle the worlds of data engineering and marketing practice will thrive in the months and years ahead. Other forecasts include data collaborations that are seamless and private by default - and maybe even a more advanced, energy-efficient iteration of AI around the corner. Read on for more insights, with thanks to:
AI doesn't automatically fix data quality, and numerical hallucinations are a lot harder to discern (and fix) than verbal ones.
Proprietary context - e.g. how a specific business operates and has operated - is the most beneficial thing to augment foundational model training with for data interpretation, but also often the hardest bit to successfully integrate.
There is also often still a gulf between the overall capabilities of GenAI, and the process and plumbing required to apply it safely and compliantly to personal data, especially across multiple platforms.
For example, MCP (Model Context Protocol) simultaneously makes it very easy to plug a LLM into all your data but very hard to organisationally control what might ultimately happen to it.
Some companies are making progress with generative AI, but many are not seeing the full benefit. The core issue is often overlooked: AI performs only as well as the identity and signals underlying it. When data is fragmented, inconsistent or siloed, AI ends up amplifying noise rather than generating reliable intelligence. This is why brands confined to closed ecosystems or dependent on unstable identifiers struggle to realise value.
The elephant in the room is that, for all the talk of transformation, most organisations are still running on operating models designed 10-15 years ago.
I've spent my career in roles where the promise of "data-driven marketing" was huge, but teams were still trying to work from disconnected systems and inconsistent IDs. Their consumer data may have been centralised, but this didn't translate to it being democratised and accessible for all parts of the business.
Brands don't have a data problem; they have an operational one. Your data strategy isn't tangible unless it runs inside your own cloud where your teams can derive value from it without jumping through hoops. Until brands close this gap, everything else, from AI to personalisation and measurement, will always be compromised.
There is an urgent need for brands to reassess how they approach brand lift measurement. Many are still trusting legacy metrics that rely heavily on ad recall, but this method fails to account for today's fragmented media environment and the complexities of cross-platform exposure. As a result, it often leads to an incomplete or misleading view of true campaign performance.
…The industry needs to prioritise metrics that better capture real changes in consumer attitude and intent, such as passive measurement, ensuring that brand lift measurement reflects genuine impact rather than memory-based guesswork.
The core issue facing marketing teams is that the signals they rely on are deteriorating. As identity fragments, both targeting and measurement lose accuracy, making it much harder for brands to attribute outcomes or demonstrate commercial impact.
Industry surveys show that most marketers and agencies are encountering significant roadblocks in their data strategies, and the dominant theme is weakened identity and inconsistent signals.
Most companies are still struggling to reach a consensus on how to turn GenAI experimentation into real business impact. Yet one thing is clear: the barrier isn't the models, it's data and governance. In 2025, organisations increasingly came to the realisation that AI is only as good as the underlying data quality, but those lessons aren't much use without being operationalised.
The biggest blockers remain fragmented data pipelines, low trust in automated outputs, and a lack of traceability when things go wrong.
As we head into 2026, the companies making real progress will be the ones pairing AI with strong data foundations, clear use-case focus, and cross-functional processes built for transparency and control.
Enthusiasm doesn't seem to be matching up with reality. … Despite showing glimmers of their potential, it seems that many [AI projects] just aren't gaining the traction needed for full business-wide rollouts, with much of this arrested development likely down to internal disorder and pressure.
The majority of CMOs and their teams are juggling at least seven channels as well as multiple agencies, leaving them with little time to focus on running, optimising, and fine-tuning AI experiments. Many also aren't using AI to help ease workloads by driving efficiencies, such as content repurposing, even though demand for content has now tripled.
GenAI is great at summarising, and that's certainly helping deal with less structured data - e.g. survey responses - and shortening the time to insight and action.
For structured data (metrics and dimensions) a key challenge is context: a lot of GenAI is embedded into platforms that are already data siloes, which then simply turn into context siloes. The challenges of data fragmentation then get multiplied from a decision-making standpoint.
I've seen a predictable pattern from working across both client-side and agency-side: when budgets tighten, organisations stop experimenting and start rationalising.
…[B]udgets must shift from throwing yet more money at tech that professes to solve all the problems at great cost to fixing the foundations.
Right now, that means moving investment away from decorative spend tools that sound impressive but don't materially improve outcomes and towards governance and identity; data quality; secure and measurable activation; and composable technology. The brands moving fastest in this journey are the ones creating hybrid teams with shared KPIs, breaking down the old separation of "data over here, marketing over there".
Budgets are shifting as CMOs face stronger pressure to demonstrate value. … Organisations are simplifying their technology stacks, strengthening first-party data strategies and building more direct relationships with publishers to regain clarity and control.
Many marketers are also reconsidering how to make their programmatic investment work harder, as declining addressability and less certain ROI increase the pressure for greater efficiency and transparency. These pressures are accelerating a move towards more accountable and transparent approaches.
We're in an era of rapid innovation which has also meant unprecedented levels of competition. The marketers who stand out won't be the ones with the fanciest tools - those are accessible to most. It's the people who actually know how to use them.
The real gap is T-shaped talent: someone with a clear superpower who also understands the data, AI and measurement systems around them. Too many teams still lack that translator who can connect strategy with analytics and automation or, simply put, lack data-literate, AI-fluent marketers. And with AI platforms evolving weekly (if not daily), self-directed learning isn't a nice-to-have, it's survival. The divide is growing fast: those who invested early in AI skills are pulling away, and others are scrambling to catch up.
The most in-demand skill is the ability to translate between marketing outcomes and data realities. You can hire great engineers and great marketers, but people who understand how to translate between both worlds deeply enough to build something useful are exceptionally rare.
The ideal practitioners are those who can spin many plates, from designing data models with a commercial lens through to managing cloud infrastructure and juggling privacy constraints with activation. The future belongs to such hybrid talent and teams; people who can operate, understand and deliver returns across data, product, and marketing while challenging "this is how we've always done it" mindsets.
Teams increasingly require commercial data literacy to interpret signals at a more granular level. Identity expertise is increasingly important, especially the ability to link data from offline, CRM and device-level sources in a privacy-safe way.
Strong data governance is becoming critical because inconsistent and fragmented datasets no longer support effective activation. Without these capabilities, data remains disjointed and difficult to activate, which directly restricts improvements in targeting, measurement and overall performance.
For as long as there are more conflicting definitions of agentic AI than a large language model can accurately count, it would be great if there was more focus on the business problems and processes that we're trying to fix/optimise/automate as opposed to constantly repackaging ever-increasing lumps of compute to throw at everything.
On a more positive note, well organised, structured and documented datasets are great news for both humans and machines. Organisations that focus on their pipelines and sort out their semantic layers (a unified business-friendly representation of their data) are well placed to take advantage of advances in technology alongside human experience and wisdom.
The industry needs to move past the expectation that legacy identifiers or platform-only targeting can continue to deliver strong performance. Those approaches no longer provide the clarity or reliability required.
The focus should shift to rebuilding and owning identity, centred on enriched first-party data that is unified across touchpoints and activated consistently. This identity-first approach is essential for improving targeting accuracy, strengthening attribution and restoring confidence in marketing investment.
Interoperability will be the watchword of the coming year. The data collaborations of 2026 will be enabled by technologies that allow organisations to derive insight and utility without ever having to expose any sensitive information or surrender their competitive edge.
And in the age of AI, protecting not only consumer privacy, but also commercially sensitive, high-value proprietary data takes on greater importance than ever: marketers want to work with richer data sources and introduce AI into their workflows without moving data out of their own environment or losing control. They will be able to make this a reality without technical challenges or the need for in-house data analysis expertise. Everything will be seamless, secure, and private by default.
The current "AI bubble" narrative is rather dominated by broad questions of perception - i.e. will a productivity gain materialise that is of sufficient scale to justify the scale of company valuations?
But another perspective on it would be a power and infrastructure bubble: it would only take one significant development in AI that trounces LLMs in terms of (power) efficiency on certain tasks to upend that sunk (or indeed sinking) investment. LLMs are already terribly inefficient for a lot of analytical tasks compared to other AI approaches; maybe they will ironically help invent the next wave of energy efficient algorithms that then immediately render them obsolete.