What are the obstacles to GenAI activation?

Is GenAI proving to be a transformative instrument for entrepreneurs working with information and analytics in 2026 – or is it one other hurdle to surmount? On this piece, our knowledgeable commentators break down the challenges going through entrepreneurs and organisations who’ve got down to apply GenAI in data-driven decision-making, from why AI can’t save poor-quality information to why AI tasks aren’t making the leap from experiment to rollout.

Further ‘elephants within the room’ for information and measurement going into 2026 embody fragmenting identification alerts, reliance on legacy metrics and fashions, and the necessity to repair underlying information foundations somewhat than investing in fancy instruments.

But it surely’s not all doom and gloom. Our consultants predict that ‘T-shaped expertise’ and entrepreneurs who can straddle the worlds of information engineering and advertising observe will thrive within the months and years forward. Different forecasts embody information collaborations which are seamless and personal by default – and perhaps even a extra superior, energy-efficient iteration of AI across the nook. Learn on for extra insights, with due to:

What’s the elephant within the room in information and analytics proper now?

AI can’t save unhealthy information

Andrew Hood, CEO at Lynchpin:

AI doesn’t mechanically repair information high quality, and numerical hallucinations are rather a lot tougher to discern (and repair) than verbal ones.

Proprietary context – e.g. how a selected enterprise operates and has operated – is essentially the most helpful factor to enhance foundational mannequin coaching with for information interpretation, but in addition usually the toughest bit to efficiently combine.

There may be additionally usually nonetheless a gulf between the general capabilities of GenAI, and the method and plumbing required to use it safely and compliantly to non-public information, particularly throughout a number of platforms.

For instance, MCP (Mannequin Context Protocol) concurrently makes it very straightforward to plug a LLM into all your information however very onerous to organisationally management what would possibly in the end occur to it.

There may be … nonetheless a gulf between the general capabilities of GenAI and the method and plumbing required to use it safely and compliantly…”

Graham Subject, CRO at Outra:

Some corporations are making progress with generative AI, however many should not seeing the complete profit. The core situation is usually ignored: AI performs solely in addition to the identification and alerts underlying it. When information is fragmented, inconsistent or siloed, AI finally ends up amplifying noise somewhat than producing dependable intelligence. Because of this manufacturers confined to closed ecosystems or depending on unstable identifiers battle to understand worth.

Outdated working fashions… and metrics

Hugh Stevens, Managing Director EMEA at AUDIENCES:

The elephant within the room is that, for all of the speak of transformation, most organisations are nonetheless operating on working fashions designed 10–15 years in the past.

I’ve spent my profession in roles the place the promise of “data-driven advertising” was enormous, however groups had been nonetheless attempting to work from disconnected techniques and inconsistent IDs. Their shopper information might have been centralised, however this didn’t translate to it being democratised and accessible for all elements of the enterprise.

Manufacturers don’t have a knowledge downside; they’ve an operational one. Your information technique isn’t tangible until it runs inside your personal cloud the place your groups can derive worth from it with out leaping by means of hoops. Till manufacturers shut this hole, every little thing else, from AI to personalisation and measurement, will at all times be compromised.

Sarah Robson, International Head of Promoting Effectiveness, On Machine:

There may be an pressing want for manufacturers to reassess how they strategy model carry measurement. Many are nonetheless trusting legacy metrics that rely closely on advert recall, however this methodology fails to account for immediately’s fragmented media atmosphere and the complexities of cross-platform publicity. In consequence, it usually results in an incomplete or deceptive view of true marketing campaign efficiency.

…The business must prioritise metrics that higher seize actual modifications in shopper perspective and intent, similar to passive measurement, making certain that model carry measurement displays real influence somewhat than memory-based guesswork.

Deteriorating alerts and fragmenting identification

Graham Subject, Outra:

The core situation going through advertising groups is that the alerts they depend on are deteriorating. As identification fragments, each focusing on and measurement lose accuracy, making it a lot tougher for manufacturers to attribute outcomes or exhibit industrial influence.

Business surveys present that the majority entrepreneurs and companies are encountering important roadblocks of their information methods, and the dominant theme is weakened identification and inconsistent alerts.

Are you seeing corporations efficiently make use of GenAI to be extra data-driven of their decision-making and advertising?

If not, what do you discover is holding them again?

Firms are struggling to translate experimentation into enterprise influence

Harriet Durnford-Smith, COO at Adverity:

Most corporations are nonetheless struggling to achieve a consensus on learn how to flip GenAI experimentation into actual enterprise influence. But one factor is obvious: the barrier isn’t the fashions, it’s information and governance. In 2025, organisations more and more got here to the realisation that AI is barely pretty much as good because the underlying information high quality, however these classes aren’t a lot use with out being operationalised.

The largest blockers stay fragmented information pipelines, low belief in automated outputs, and an absence of traceability when issues go incorrect.

As we head into 2026, the businesses making actual progress would be the ones pairing AI with sturdy information foundations, clear use-case focus, and cross-functional processes constructed for transparency and management.

The largest limitations [to AI implementation] stay fragmented information pipelines, low belief in automated outputs, and an absence of traceability when issues go incorrect.

Mónica González, CMO at Making Science:

Enthusiasm doesn’t appear to be matching up with actuality. … Regardless of displaying glimmers of their potential, evidently many [AI projects] simply aren’t gaining the traction wanted for full business-wide rollouts, with a lot of this arrested growth doubtless right down to inside dysfunction and strain.

The vast majority of CMOs and their groups are juggling no less than seven channels in addition to a number of companies, leaving them with little time to concentrate on operating, optimising, and fine-tuning AI experiments. Many additionally aren’t utilizing AI to assist ease workloads by driving efficiencies, similar to content material repurposing, though demand for content material has now tripled.

Context is every little thing

Andrew Hood, Lynchpin:

GenAI is nice at summarising, and that’s actually serving to cope with much less structured information – e.g. survey responses – and shortening the time to perception and motion.

For structured information (metrics and dimensions) a key problem is context: numerous GenAI is embedded into platforms which are already information siloes, which then merely flip into context siloes. The challenges of information fragmentation then get multiplied from a decision-making standpoint.

How are budgets and methods of working altering?

Firms must put money into fixing the foundations

Hugh Stevens, AUDIENCES:

I’ve seen a predictable sample from working throughout each client-side and agency-side: when budgets tighten, organisations cease experimenting and begin rationalising.

…[B]udgets should shift from throwing but more cash at tech that professes to resolve all the issues at nice price to fixing the foundations.

Proper now, which means shifting funding away from ornamental spend instruments that sound spectacular however don’t materially enhance outcomes and in the direction of governance and identification; information high quality; safe and measurable activation; and composable know-how. The manufacturers shifting quickest on this journey are those creating hybrid groups with shared KPIs, breaking down the outdated separation of “information over right here, advertising over there”.

 A shift in the direction of worth, accountability and transparency

Graham Subject, Outra:

Budgets are shifting as CMOs face stronger strain to exhibit worth. … Organisations are simplifying their know-how stacks, strengthening first-party information methods and constructing extra direct relationships with publishers to regain readability and management.

Many entrepreneurs are additionally reconsidering learn how to make their programmatic funding work tougher, as declining addressability and fewer sure ROI improve the strain for larger effectivity and transparency. These pressures are accelerating a transfer in the direction of extra accountable and clear approaches.

What expertise are in demand, and that are almost certainly to be lacking?

T-shaped expertise

Harriet Durnford-Smith, Adverity:

We’re in an period of fast innovation which has additionally meant unprecedented ranges of competitors. The entrepreneurs who stand out received’t be those with the fanciest instruments – these are accessible to most. It’s the individuals who really know learn how to use them.

The actual hole is T-shaped expertise: somebody with a transparent superpower who additionally understands the info, AI and measurement techniques round them. Too many groups nonetheless lack that translator who can join technique with analytics and automation or, merely put, lack data-literate, AI-fluent entrepreneurs. And with AI platforms evolving weekly (if not each day), self-directed studying isn’t a nice-to-have, it’s survival. The divide is rising quick: those that invested early in AI expertise are pulling away, and others are scrambling to catch up.

Hybrid expertise that may straddle the info and advertising divide

Hugh Stevens, AUDIENCES:

Essentially the most in-demand ability is the power to translate between advertising outcomes and information realities. You possibly can rent nice engineers and nice entrepreneurs, however individuals who perceive learn how to translate between each worlds deeply sufficient to construct one thing helpful are exceptionally uncommon.

The perfect practitioners are those that can spin many plates, from designing information fashions with a industrial lens by means of to managing cloud infrastructure and juggling privateness constraints with activation. The longer term belongs to such hybrid expertise and groups; individuals who can function, perceive and ship returns throughout information, product, and advertising whereas difficult “that is how we’ve at all times finished it” mindsets.

You possibly can rent nice engineers and nice entrepreneurs, however individuals who perceive learn how to translate between each worlds deeply sufficient to construct one thing helpful are exceptionally uncommon.

Industrial information literacy, identification experience, and information governance

Graham Subject, Outra:

Groups more and more require industrial information literacy to interpret alerts at a extra granular stage. Id experience is more and more necessary, particularly the power to hyperlink information from offline, CRM and device-level sources in a privacy-safe means.

Sturdy information governance is turning into vital as a result of inconsistent and fragmented datasets now not help efficient activation. With out these capabilities, information stays disjointed and troublesome to activate, which immediately restricts enhancements in focusing on, measurement and total efficiency.

What’s the one pattern you would like everybody would cease speaking about? And what must be a key space of focus for organisations subsequent yr?

Organisations ought to zero in on tangible enterprise issues

Andrew Hood, Lynchpin:

For so long as there are extra conflicting definitions of agentic AI than a big language mannequin can precisely depend, it will be nice if there was extra concentrate on the enterprise issues and processes that we’re attempting to repair/optimise/automate versus always repackaging ever-increasing lumps of compute to throw at every little thing.

On a extra constructive word, properly organised, structured and documented datasets are nice information for each people and machines. Organisations that target their pipelines and type out their semantic layers (a unified business-friendly illustration of their information) are properly positioned to make the most of advances in know-how alongside human expertise and knowledge.

The main target must be on proudly owning identification

Graham Subject, Outra:

The business wants to maneuver previous the expectation that legacy identifiers or platform-only focusing on can proceed to ship sturdy efficiency. These approaches now not present the readability or reliability required.

The main target ought to shift to rebuilding and proudly owning identification, centred on enriched first-party information that’s unified throughout touchpoints and activated constantly. This identity-first strategy is crucial for bettering focusing on accuracy, strengthening attribution and restoring confidence in advertising funding.

Do you’ve an outrageous prediction, or a protected wager, to share?

‘All the things will probably be seamless, safe, and personal by default’

Alistair Bastian, CTO at InfoSum:

Interoperability would be the watchword of the approaching yr. The information collaborations of 2026 will probably be enabled by applied sciences that enable organisations to derive perception and utility with out ever having to reveal any delicate info or give up their aggressive edge.

And within the age of AI, defending not solely shopper privateness, but in addition commercially delicate, high-value proprietary information takes on larger significance than ever: entrepreneurs wish to work with richer information sources and introduce AI into their workflows with out shifting information out of their very own atmosphere or dropping management. They may have the ability to make this a actuality with out technical challenges or the necessity for in-house information evaluation experience. All the things will probably be seamless, safe, and personal by default.

Will we see the subsequent evolution of AI?

Andrew Hood, Lynchpin:

The present “AI bubble” narrative is somewhat dominated by broad questions of notion – i.e. will a productiveness acquire materialise that’s of enough scale to justify the size of firm valuations?

However one other perspective on it will be an influence and infrastructure bubble: it will solely take one important growth in AI that trounces LLMs by way of (energy) effectivity on sure duties to upend that sunk (or certainly sinking) funding. LLMs are already terribly inefficient for lots of analytical duties in comparison with different AI approaches; perhaps they may paradoxically assist invent the subsequent wave of power environment friendly algorithms that then instantly render them out of date.