Chapter 2

From Pilot to Proof – Operational Impact, Value, and the Barriers in Between

For organisations still running pilots, the clock is ticking to evolve beyond them. For those that already have, a demanding question arises: what has the investment actually returned?

And the answer?

The operational footprint of AI is expanding faster than its economic footprint.

Zero percent report enterprise-level impact from AI in CX – that is, material effects on revenue growth, retention, and margin at a scope the C-suite actually responds to. Eighteen percent describe having no quantifiable impact to date, and a further 39% describe gains that are real but isolated – pockets of efficiency that haven't yet found their way into anything a CFO would recognise on a profit and loss statement. Only 11% can point to functional successes with measurable commercial outcomes.
This is a direct consequence of the maturity gap.

Impact breeds greater impact; it must be built, deliberately and in sequence. Naturally, organisations concentrated in experimentation are unlikely to generate disproportionate returns because value accrues exponentially as adoption expands across workflows, functions, and customer journeys. In many respects, these numbers mirror the profile of the adoption curve itself.

Figure 6: What level of business impact has AI generated within your CX operations over the past 12 months?

So where, exactly, is ambition failing to translate into outcomes?

Forty-two percent of organisations identify their capacity to transition from pilots to scaled operations as the single largest gap between AI investment and realised value – comfortably the most-cited barrier and nearly double the next.

Indeed, a pilot can only do so much. It can demonstrate that a model works; it can’t demonstrate that an organisation can roll it out across the value chain. These are distinct problems requiring distinct capabilities – one technical, one operational – and most organisations have solved for the first while remaining structurally underpowered for the second as it stands.

The pattern continues across adjacent hurdles. Twenty-four percent struggle to turn customer insight into real-time action, meaning the intelligence exists, but the organisational reflex to act on it is lacking. Twenty-one percent cite converting efficiency gains into tangible financial outcomes – a signal that should give any CFO pause in the context of current AI investment narratives.

All said, pilots are performing their intended function. The constraint lies in the connective infrastructure – operating systems, governance frameworks, and organisational muscle memory – that converts validated models into live capability. Until that infrastructure is established, each new pilot merely adds to a growing inventory of proven concepts waiting for a route into wider production.

organisational reflex to act on it is lacking. Twenty-one percent cite converting efficiency gains into tangible financial outcomes – a signal that should give any CFO pause in the context of current AI investment narratives.

Figure 7: Where is the gap between AI investment and realised business value currently most evident within your CX operations? (Select up to 2)

The barriers to scale are consistent across organisations. Technology fragmentation (34%), data quality issues (34%), and organisational silos (30%) dominate, with governance (22%) and talent gaps (20%) rounding out the top five.

Viewed in aggregate, these discrete challenges reveal that implementing AI at scale is rarely constrained by any single bottleneck. Instead, it is the cumulative interaction between them. Technology fragmentation degrades data quality. Data fragmentation limits automation. Limited automation reinforces organisational silos. The cycle feeds itself. This is why the primary obstacle of AI adoption increasingly resembles one of enterprise coordination over and above technological prowess. Access to advanced models is becoming ubiquitous. Access to high-quality, accessible, context-rich, proprietary data, on the other hand, is not – that is one of the principal variables now.

More fundamentally, these constraints are a tax on speed. Every silo adds friction. Every disconnected workflow introduces delay. Every fragmented channel increases the likelihood that an insight arrives after the moment it could have influenced an outcome. An organisation can possess strong data, capable models, and clear governance, and still be outmaneuvered by a competitor that simply moves faster. In today’s AI-enabled environments, outperformance accrues to those minimising the distance between insight and action.

Figure 8: What are the biggest barriers preventing your organisation from scaling AI more rapidly across CX? (Select up to 2)

If pilot-to-scale is the defining operational challenge of AI adoption, personalisation is its clearest test case.

Eighty-six percent of organisations remain at "limited" or "developing" personalisation capability. Genuinely predictive, individualised experiences are, for the moment, the exception.

Personalisation has been the CX industry's longest-running aspiration, defined in the public imagination by Amazon, Netflix, and Spotify, and chased, at varying distances, by almost every organisation since – it predates the current AI cycle entirely. That 43% of organisations are operating on broad segments or static rules – the same paradigm that defined personalisation back in 2016 – is a sharp finding. A decade of investment in customer data, digital platforms, and now AI, and still nearly half of organisations are exactly where they started. It reinforces the case for rewiring around AI against accommodating it.

The 14% delivering advanced degrees of personalisation are distinguished by the strength of their foundations – the architecture that determines which signals matter, how they connect across disparate data sources, and routes the right recommendation to the right moment in the customer interaction. Once that is in place, these organisations then showcase the confidence to let AI-driven recommendations operate in real time, without a human override at every step. And that confidence is a cultural threshold as much as a technical one, and it takes longer to build than any algorithm.

Figure 9: How would you describe your organisation’s current capability to deliver AI-driven personalisation at scale across customer interactions?