AI has become one of the dominant topics in customer marketing. Yet for many brands, the challenge is not access to tools, but clarity. Clarity on where AI should be applied, what problems it should solve, and what must be true inside the business before decisioning can deliver genuine value.
Those questions shaped two Plinc breakfast briefings held in London and Manchester earlier this year, centred on AI-assisted decisioning in CRM and what it will take to move from ambition to operational reality.
To frame the discussion, Stuart Russell, Chief Strategy Officer at Plinc, outlined a practical model for AI decisioning in customer marketing. He explored the shift from prediction to decisioning, the data foundations required to support it, and the role of causal measurement in enabling systems to learn from impact rather than correlation.
This was followed by panel discussions with senior CRM leaders including Giulia Diomampo, Head of CRM at Space NK; Grant Baillie, Head of CRM at Secret Escapes; Liam Savage, CRM Lead, Data & Martech at Currys; Nicola Travis, VP of CRM at Secret Sales; and Steve Moriarty, Group Head of CRM & Insight at Crew Clothing. The conversation moved from theory to practice, examining how AI is being adopted inside organisations, how teams are navigating governance and culture change, and which skills will matter most in an AI-assisted future.
This debrief distils the most important themes from Stuart’s framework and the subsequent panel discussions into grounded principles for CRM leaders navigating AI today.
1. From prediction to decisioning
Stuart opened both sessions by reframing the AI conversation in CRM. For years, customer marketing has focused on prediction. Who is likely to purchase. Which category they may move into next. What their future value could be. Those capabilities are now established.
The shift, he argued, is from prediction to decisioning.
Rather than using models to inform static rules, AI-assisted decisioning determines what should happen next for each individual customer. Which message should be prioritised. Which channel should be used. Whether a communication should be suppressed altogether. Not through fixed journeys or predefined hierarchies, but through continuous feedback from outcomes.
This represents a shift in the CRM operating model. Customers frequently qualify for multiple campaigns at once. Manual prioritisation rules attempt to resolve those overlaps, but they are rigid by design. Decisioning systems aim to resolve them dynamically, learning over time which choices drive incremental value.
This is not about increasing volume. It is about improving choice, closer to the moment of interaction, and allowing those choices to strengthen as evidence accumulates.
2. The foundations AI depends on
That shift only works if the underlying data is fit for purpose.
Stuart emphasised that AI decisioning is rarely constrained by algorithms. It is constrained by the quality and structure of the data it learns from.
Three foundations stood out.
The first is identity resolution. Are decisions being made at a channel level or a person level. Without confidently linking interactions across touchpoints, AI will optimise fragments rather than customers.
The second is data latency. Not every signal needs to be real time, but some do. Understanding which decisions require immediacy and which can operate on daily or weekly cycles is critical. Over-engineering real-time infrastructure consumes resource. Under-investing where speed matters limits impact.
The third is event and exposure history. Decisioning systems require memory. Not just what customers have done, but what they have seen. Without a consistent record of both actions and exposures, systems cannot avoid repetition or learn properly from cause and effect.
Stuart also outlined the strategic choices businesses face. Some will rely on platform-native decisioning. Others will co-create capabilities with partners. A small number may build internally. Each path has implications for data control, transparency and flexibility. AI amplifies whatever foundation it sits on. Weak data produces confident but unreliable decisions. Strong foundations create the conditions for meaningful learning.
3. Adoption in practice: how organisations are progressing
If Stuart’s framework set out what should be true for AI-assisted decisioning to work, the panel discussions added practical context from inside large CRM teams.
Across both events, leaders shared how AI adoption is unfolding in their organisations.
In many cases, progress is being driven by expectation at senior level. Platforms are embedding AI features as standard. Yet objectives are not always tightly defined. What exactly is being improved. Which decision is being made better. What success looks like beyond simply “using AI”.
Nicola Travis, VP of CRM at Secret Sales, reflected that “there are no AI experts.” Adoption, she explained, is iterative and exploratory. Teams are building experience as they go, rather than following a fixed blueprint.
Several panellists emphasised the importance of keeping AI anchored to business strategy. Liam Savage, CRM Lead, Data & Martech at Currys, described the risk of becoming technology-led rather than strategy-led:
“We risk being slightly disjointed from the business strategy, led by how we embed a cool feature or tool into a problem that might not actually be a priority.”
He also highlighted the importance of commercial clarity when navigating governance and approval processes:
“If you present the business problem and the challenge, and a clear solution, it’s much easier to get through governance and review processes.”
The theme of clarity also surfaced in London. Grant Baillie, Head of CRM at Secret Escapes, emphasised that AI must be anchored in a clear strategic use case rather than deployed as a blanket improvement. Without defined objectives and a culture of test and learn, he noted, decisioning quickly becomes superficial.
In practice, many teams are beginning with operational efficiency. Giulia Diomampo, Head of CRM at Space NK, suggested starting with the kinds of tasks you would typically assign to an intern as a practical entry point for AI agents within CRM workflows. These use cases are lower risk and immediately valuable. They reduce cognitive load and free capacity, even if they do not yet represent full decisioning.
The overall tone was pragmatic. Technology is available. The determining factors are clarity of objective, confidence in data and alignment across teams.
4. Measurement and judgement: what makes AI credible
A central theme in Stuart’s presentation was the distinction between correlation and impact.
AI systems are adept at identifying patterns. They can learn that sending more emails drives more clicks. They can observe that customers who browse a category are more likely to purchase from it. Without a causal learning loop, however, they will optimise what is easiest to measure rather than what truly matters.
Stuart argued that incrementality is foundational infrastructure in an AI-assisted world. Decisioning engines must learn from cause and effect. That requires consistent target and control groups, continuous measurement of incremental revenue or margin, and a feedback loop that informs future decisions.
Without that layer, AI optimises noise.
The panels reinforced this from a practical standpoint. Several leaders spoke about the importance of owning measurement logic rather than relying solely on platform-reported uplift. When multiple campaign variations are deployed simultaneously, understanding true incremental impact becomes more complex and more important.
There was also strong emphasis on judgement. Steve Moriarty, Group Head of CRM & Insight at Crew Clothing, emphasised that while AI can generate options, human judgement is still required to interpret uncertainty and commercial context. Measurement ensures the system learns the right lessons. Judgement ensures those lessons are interpreted within commercial and brand context. Together, they make AI decisioning credible rather than superficial.
5. What this means for CRM leaders now
Taken together, Stuart’s framework and the panel discussions point to a clear conclusion. AI decisioning is not a shortcut around disciplined CRM. It is an evolution of it.
The starting point is not a tool. It is a decision. Which choices in your customer journey are constrained by manual rules, competing priorities or incomplete data. Where would better learning genuinely change outcomes.
From there, the focus shifts to foundations. Identity must be resolved at a customer level. Signals must be prioritised rather than accumulated. Measurement must be structured around incrementality rather than surface metrics. Without these elements, automation simply scales existing weaknesses.
There is also a sequencing question. Several panellists described beginning with operational use cases, removing repetitive tasks and reducing cognitive load before attempting full decisioning. That progression builds confidence and frees capacity for more strategic experimentation.
Finally, there is the human dimension. As AI takes on more execution, the scarce capability becomes judgement. Framing problems clearly. Interpreting evidence responsibly. Balancing short-term gain against long-term value.
AI-assisted decisioning will continue to mature. The organisations that benefit most will not be those that adopt fastest, but those that build strong foundations, measure rigorously and treat judgement as a strategic asset.
The upcoming Plinc webinar on AI Decisioning, will expand further on these themes. Follow Plinc on LinkedIn to find out more.
