VFS Global’s story starts with a fairly simple problem. Governments were spending enormous time and resources on the administrative, non-judgmental side of visa and consular services: receiving documents, booking appointments, capturing biometrics, while their real expertise, and their sovereign responsibility, lay in the actual decision. We were the first to build a public-private partnership model around that gap, back in 2001, and it has scaled from a handful of centres to more than 4,200 application centres across 168 countries today, working with 71 client governments.

What makes us different isn’t really a feature or a price point, because we’re not competing in that kind of market. It’s trust, built over two and a half decades, at a scale almost nobody else in this space operates at. We work across every language, every culture, and every regulatory environment a government could ask for, and we have never once tried to move into the part of the process that belongs to the government: the actual adjudication. That discipline, staying strictly non-judgmental, is what lets 71 governments hand us the front door to their citizens and trust that we won’t touch what isn’t ours to touch.

Honestly, most of it comes from governments we already serve, expanding the scope of what we do for them. We added a whole set of miscellaneous services: power of attorney, attestation, life certificates in the US market last August, and that alone is on track to add close to 70,000 applications a year. The same pattern plays out globally: a government that trusted us with visas starts trusting us with passports, then consular services, then something new entirely. We also win new government partnerships through open, competitive tender, which is exactly how it should work in this business. Where I wish more of it came from is earlier conversations. Too often we get called in once a government is already under strain: a backlog, a policy shift, a crisis that exposes how thin its own administrative capacity really is. I would love more of our new business to come from governments thinking about this proactively, building resilient citizen services before they’re forced to, rather than after.

The one I would point to first is demand forecasting. Predicting appointment volumes, staffing needs, and peak season pressure used to rest almost entirely on institutional memory and spreadsheets. We have genuinely handed a meaningful part of that planning over to AI now, and it has made us noticeably better at getting ahead of a surge instead of reacting to one.

The one we tested and chose not to take live was letting AI make the final call on whether an application was complete. We ran it as a contained pilot, the idea being that the system would wave through anything it judged sufficient. It looked efficient on paper, but the pilot surfaced enough edge cases, applications that were unusual in some small but important way, for us to see clearly how easily those could be missed without a trained officer’s judgement. So we made a deliberate call not to deploy it that way: AI flags what deserves a second look; a person always makes the call. That wasn’t a step backward. That’s exactly what a pilot is supposed to tell you before you scale something.

Applicants absolutely arrive more informed than they used to, and a good part of that is AI-driven, though maybe not in the way people assume. We’ve built our own guided tools, including a virtual assistant, and worked on things like the UKVI chatbot, and those genuinely change how someone shows up: fewer basic questions, better prepared documents. General-purpose assistants like ChatGPT are starting to enter that picture too. We occasionally see applicants reference something an AI assistant told them, and it’s a mixed bag: sometimes accurate, sometimes confidently wrong, because visa requirements change by government, by category, and by month, and a general model doesn’t always know that. That is actually pushing us to think harder about being the verified source those systems pull from, rather than leaving it to chance. If people are going to ask an AI assistant before they ask us, we would rather that assistant be drawing on accurate, current information than a three-year-old forum post.

Do this: get your data discipline right before you get ambitious with AI. Minimise what you collect, be explicit about the purpose, control access tightly, and only then start layering in AI for guidance, planning and internal productivity, the areas where it genuinely helps. Pilot with a clear use case and clear boundaries, not a vague sense that you should be doing something with AI.

Ignore the pressure to chase every new tool that launches. That’s noise, not strategy. And ignore any pitch that quietly moves AI toward the actual decision, whatever that decision is in your industry. In ours, that is a visa outcome, and it will never sit with a model. In yours, it’s whatever your customers are trusting you to get right on their behalf. The moment AI starts making that call instead of supporting the person who does, you have traded a genuine advantage for a liability you won’t see until it’s already a problem.