The last mile of AI transformation is a conversation
I’ve sat in enough “digital transformation” meetings to recognise the pattern: we talk about platforms, process redesign, data maturity, governance, adoption. We map the operating model. We choose the tooling. We measure efficiency. And then—somewhere between rollout and reality—everything hits the same bottleneck.
A person has to speak to another person.
I’m an NHS psychiatry doctor by training, so I’ve spent years watching how small shifts in language can change outcomes and even save lives—especially when someone is frightened, angry, or overwhelmed. That perspective has made me increasingly convinced that the last mile of AI transformation isn’t the technology. It’s the conversation.
It might be a council officer speaking to someone facing homelessness, an adviser in a government contact centre trying to de-escalate a call that starts with “I’ve been waiting for hours,” a teacher handling a safeguarding concern, a financial adviser explaining a pension outcome that doesn’t match expectations, or a solicitor navigating a divorce conversation where every sentence can inflame—or steady—the situation. In every case, the outcome is shaped by the same human fundamentals: clarity, empathy, structure, and emotional judgement under time pressure.
What’s striking to me is how often organisations treat that last mile like an assumption. We modernise systems, but we assume the conversation will take care of itself.
Work has shifted. Even in roles that used to be face-to-face, more critical interactions now happen without visual cues: over the phone, in hybrid settings, through call-backs, and in time-pressured moments where someone is trying to solve a problem while keeping a queue moving. For many organisations, this is the real front door of public sector services and customer-facing delivery.
Audio-only interactions raise the cognitive load. You don’t get the subtle reassurance of eye contact. You can’t read the room. Silence becomes ambiguous. Interruptions land harder. Tone becomes the message.
And yet, much of our communication skills training still looks like it did years ago: classroom role-play, occasional workshops, shadowing, or “learning by doing” in real situations where the stakes are already high. Practice opportunities are scarce. Feedback is inconsistent. And repetition—arguably the most important ingredient for skill-building—is the first thing to go when calendars are tight.
If you want a simple litmus test for modern workplace readiness, it’s this: can someone hold a high-stakes phone call where the other person is upset, confused, frightened, defensive, or simply exhausted?
We call communication “soft” right up until it becomes a service failure, a complaint, a safeguarding escalation, a regulatory issue, or a resignation. The cost of a poorly handled conversation rarely shows up as a single line item. It shows up as rework, delays, repeat calls, complaints, escalations, avoidable conflict, and staff burnout.
In an AI-accelerated world, this matters even more. Automation can compress timelines and speed up decision-making, but it can’t shortcut trust. Faster systems can amplify the impact of a misstep in human communication: when people feel unheard or confused, they don’t blame the workflow—they blame the organisation.
That’s why I increasingly see communication as a core part of workforce upskilling. Not as an “extra,” but as capability infrastructure.
I am convinced that if we want communication to improve at scale, we have to treat it like any other capability: something you can practise safely, frequently, and with high-quality feedback. A modern approach I’d summarise as voice-based simulation plus coaching-grade feedback fits the world we’re actually working in.
Voice-based practice matters because it mirrors phone-first reality. It forces you to manage pace, tone, hesitation, interruption, and silence—the things that shape outcomes but rarely appear in a written rubric. It also makes repetition possible: you can practise the conversations that derail real work—complaints, boundary-setting, safeguarding concerns, explaining complex decisions, handling anger, delivering information that will disappoint someone—and then try again with a different approach. With AI-supported feedback, practice isn’t limited to when a supervisor, coach, or facilitator is available; the feedback can be consistent, detailed, and tied to the behaviours that make conversations safer and clearer.
And then there’s the delicate topic: emotional sentiment analysis. I don’t see it as a verdict on what someone “really felt.” I see it as a reflection tool—another signal that can help you notice patterns, like where empathy drops when tension rises, or where reassurance lands as minimisation. The guardrail is crucial: this should support learning and self-awareness, not surveillance.
New graduates entering the workplace often have strong technical knowledge and digital confidence, but limited experience handling real-time emotional complexity. They may be brilliant on paper and still freeze when a call becomes unpredictable. Practice helps them build a repertoire of phrases, pacing, and repair moves—what to do when you’ve said the wrong thing, or when the other person shuts down.
Mid-career professionals face a different challenge: their instincts were shaped in a different environment. They’re now leading hybrid teams, managing AI-enabled change, and navigating higher expectations for psychological safety and inclusive communication—often without being given time to recalibrate their conversational habits. They don’t need theory; they need efficient, relevant practice with precise feedback.
If I had one message for digital leaders, it’s this: communication isn’t a “nice to have” alongside transformation. It’s part of the infrastructure.
If you’re investing in AI and new operating models, invest equally in the human layer that makes them work. Create safe places to practise the conversations that define outcomes. Build feedback loops that make improvement normal. Treat phone-first, real-world practice as capability-building, not remediation.
Because the last mile of AI transformation isn’t a dashboard.
It’s a conversation.
Read More Workforce & AI Skills
