May 14, 2026
Start Human, Support With AI: Where AI Strengthens CX — and Where It Breaks It
According to Gartner research, 81% of senior marketing leaders measure AI automation success by time saved, while only 36% evaluate it through a CX lens. Leaders have long faced a trade-off between productivity and CX, and the usual outcome is that improving CX can wait for another day. But now, technology vendors tells us that, with some AI magic, you can have the best of both worlds. Unfortunately, the reality is that AI is not the fix for all CX problems.
A Simple Principle: Start Human. Support With AI.
Here’s the playbook CX leaders need to cut through the hype: Start human. Support with AI. Humans must own customer understanding and experience design; AI should be used selectively in delivery to reduce effort, instill trust, and build customer confidence.
This principle is practical because it tells you where AI belongs: not as the author of your customer truth, but as the accelerator and maintainer of it.
Foundation 1: Know the Customer (Don’t Automate Understanding)
Customer understanding is not a static artifact – it’s continuously updated insight into customers’ goals, intents, behaviours, and value realisation. Yet many brands are falling behind. According to a Gartner survey, 58% of UK consumers say brands fail to understand their needs and preferences.
Now, you’ll hear vendors highlight how their AI models can now generate your journey maps, and define your personas for you. Comprehensive journey maps in minutes, data-driven outputs, precision-crafted personas – it’s pretty tempting. But it’s also a risk. Here’s why.
Firstly, these tools are only as good as their training data – which isn’t specific to your company – and what you provide as an input. But companies often fail to combine their survey data with inferred and indirect data, like customer journey analytics, customer reviews and chat logs. We end up missing a lot of customers’ lived experiences with our company because survey response rates are low and only as good as the questions asked.
The second is false precision. Vendor tools are designed to deliver what looks like a finished output. So when AI produces a perfect-looking persona or journey map, we assume it is perfect and confidently build our CX around it. That is, until customers show us that it isn’t true for them.
And finally, we need to consider the impact of team atrophy. If AI ‘does the understanding,’ we stop doing it. But our remit – to design experiences that drive growth – depends on us recognising patterns, making trade‑offs, and driving a customer-centric culture across the org. If that muscle fades, the strategy fades with it. So we must avoid the temptation to automate initial understanding.
Instead, organisations must invest the time and effort in building a human-led baseline for personas and journeys, then use AI to maintain it: cluster verbatims, summarise VoC at scale, detect shifts in operational signals, and wire insight to action.
Foundation 2: Instill Trust (Design Customer Control, Not Containment)
Trust is a customer’s belief that a company will follow through on its stated intentions, especially in difficult moments. And trust is declining: UK consumer trust in big brands fell from 70% in 2021 to 60% in 2025.
That decline coincides with the period in which we’ve been investing heavily in AI for CX. I’m not going to pin all of our trust problems on AI, but I do think that it’s having a significant impact. Here’s why.
AI backfires when it removes customer agency. Customers want four things: disclosure, choice, explainability, and recourse. For example, a Gartner survey found 77% want to be notified when interacting with GenAI. But most of our deployments of AI in customer experience delivery fall foul of at least one (and often several) of these four tests.
So design AI to put customers in control – start human, support with AI – by labelling AI use, providing opt-outs, keeping humans in the loop for high-impact or high-emotion interactions, configuring tripwires, and making escalation/refunds/corrections easy.
Foundation 3: Orchestrate Personalised Customer Journeys
Personalisation should mean “the right message, at the right time, in the right channel.” But customers feel under siege: according to Gartner data, 48% of personalised digital communications are rated creepy, irrelevant, or both. And 62% of UK consumers would rather give up more relevant personalisation than have their digital behaviours tracked.
The fix isn’t “more personalisation with AI.” It’s changing where personalisation lives: shift from channel-by-channel optimisation to journey-level orchestration, supported by a centralised governance model and cross-functional teams. Then adopt an experience pattern that scales without making customers uncomfortable: Ask, don’t guess (use willingly provided signals), Orchestrate, don’t broadcast (optimise the journey, not the channel), and Assemble, don’t produce (build a modular, governed content library that AI assembles from).
What to Do Next
Start by targeting the moments that create the most friction and doubt — the “high effort, low confidence” parts of the journey — and use AI to detect patterns, not dictate strategy. In the next 90 days, audit customer-facing AI for transparency, choice, explainability, and recourse; then wire VoC signals into actions that reduce effort and restore confidence. Over the next 12 months, evolve your operating model toward journey orchestration and build the governed “content spine” AI can safely assemble from.
AI can absolutely improve CX — but only when it respects the human in the relationship.
Christopher Sladdin is a Senior Director Analyst in the Gartner Customer Service & Support Practice, advising customer service leaders at the world’s leading organisations on their CX strategies.
