Governance · Event notes
If AI is being embedded into day-to-day work, then leadership has to take responsibility for guardrails, trust, and the foundations that make AI safe to run at scale.
I went along to Manchester Digital’s Tech Leader Talks on “Responsible leadership in an AI era”, hosted by Bruntwood at Circle Square and sponsored by Softwire. The panel brought together leaders from Softwire, DWP Digital, Zopa, Arm and Kao Data.
What I liked about the discussion is that it didn’t get stuck in model hype. It moved quickly into the practical reality: if AI is being embedded into day-to-day work, then leadership has to take responsibility for guardrails, trust, and the foundations that make AI safe to run at scale.
“Pace vs practicality” is really about risk and operating discipline
The chair, Cally Clifft (Softwire), asked about pace versus practical reality. Arm’s perspective was refreshingly grounded: there are new models arriving constantly, but the important question is whether they are viable in real deployments — how they run, how they scale, and whether they work outside of a demo.
DWP’s contribution put a sharper edge on the same point: there’s pressure for government to adopt AI quickly, but when you serve 20 million people the standard is different. The focus becomes adopting AI safely, protecting vulnerable customers, and sorting out foundations like structured data, a responsible AI framework, and legacy technology.
That’s where “pace” often gets misunderstood. The constraint isn’t ambition. It’s that once AI touches real people, real money, real eligibility, or real vulnerability, the failure modes stop being theoretical. At that point, governance becomes part of how you operate safely.
Trust and transparency are not optional in public services
One of the more concrete examples from the panel was DWP’s AI question-and-response tool designed to help work coaches navigate Universal Credit guidance. The point is simple: reduce time spent searching and increase time spent talking to the person in front of you. It was tested in 30 job centres in Wales and is now being extended across the UK.
DWP also talked about the reality of volume: they receive 25,000 physical letters a day. They’ve introduced a vulnerability scanner so a process that used to take weeks can now be run in a day, scanning letters and fast-tracking vulnerable customers.
This is exactly where “responsible AI” becomes a serious engineering and governance problem. It’s not enough for the tool to be helpful on average. You need to know what happens when it’s wrong, what it misses, how staff challenge it, and how quickly you can intervene if failure patterns show up at scale. Public trust doesn’t fail gradually — it collapses when people feel the system is opaque or unaccountable.
Fintech: better experience is the right goal, but execution risk is real
Zopa’s example was a good illustration of AI in a consumer setting that’s not just a chatbot bolted on for marketing. They’ve launched an AI voice assistant that lets users interact with their banking app in a more intuitive way, with a specific accessibility benefit for people with disabilities. They also mentioned higher-value tasks like splitting a bill by talking through items ordered, with the app working out the shares.
That’s a genuinely useful direction — but it’s also where governance and security matter even more, because the moment AI moves from “help me find” to “help me do”, it starts to influence outcomes. In fintech, “responsible” is less about nice wording and more about controls: verification, step-up authentication where needed, fraud patterns, and avoiding the classic failure mode of “right amount, wrong person”.
Infrastructure: the AI era is constrained by megawatts and geography
Kao Data brought the conversation back to the part that often gets ignored: infrastructure is physical, and AI growth is constrained by where compute can actually sit. The panel highlighted how delivering AI infrastructure is constrained around West London and Slough — noting that Slough, one of the largest data centre clusters, is now at capacity. They talked about the need to diversify, including a site under development in Stockport, and the importance of central government stewardship to capitalise on the opportunity for UK compute.
This matters for governance too. As compute becomes more strategic and more expensive, the operational disciplines you’d expect in any critical infrastructure environment start to apply to AI: access control, audit trails, accountability, and cost visibility.
A quick point on “AI done well”
The “AI done well” examples were useful because they weren’t framed as AI replacing people. They were framed as AI improving outcomes. Arm mentioned Arm CPUs used with Raspberry Pi drones in conservation work — monitoring animals and potentially detecting poachers. Kao Data referenced a client using AI to improve cancer treatment — a reminder that AI isn’t only about chat interfaces. Zopa’s example about tailoring tone-of-voice outputs was a practical nod to how these tools are being used in real management workflows.
What sits behind all of those examples is the same requirement: you need a way to deploy AI that is repeatable, measurable, and safe — not just “available”.
Why we built Conductor: governance that sits inside the workflow
Listening to the panel, the underlying issue was familiar: AI adoption is already happening, but governance often lags behind it. In most businesses that means AI use becomes fragmented, informal, and hard to see centrally — which is exactly when security and consistency degrade.
This is why we built Conductor. Conductor is an AI governance and orchestration layer. It sits between your business and the AI models and provides visibility, guardrails, cost control, and auditability — while still allowing teams to apply AI to real operational workflows.
The point isn’t to replace tools like ChatGPT or Copilot. The point is that those tools don’t govern themselves in a way an organisation can rely on. Conductor is the control layer that governs how those tools are used across the organisation — with central oversight, role-based guardrails, cost monitoring, audit traceability, and workflow integration. And security can’t be an afterthought: Conductor is built with enterprise-grade architecture including role-based access control, secure authentication, audit logging, and token usage monitoring.
Responsible AI leadership is mostly about being in control. The organisations that scale AI safely won’t be the ones with the best slogans — they’ll be the ones that can show, clearly, how AI is governed inside the workflows that matter.
Want to see what governed AI looks like inside your own workflows? A short discovery call is usually enough to find where Conductor pays for itself.
