AI in crisis communications: why route knowledge beats the next new tool

I have spent this week getting from Barnes to Marylebone and back by every means London offers. A train into Waterloo, the Underground across the middle, a bus when the line was down, and a good deal of walking in heat the city wasn't built for. The first morning was bewildering. By the fourth I was reading the network without thinking, boarding the carriage that lined up with the exit I wanted, knowing which interchange was a long hot slog and which was barely a minute. The map had not changed. I had.

Somewhere on a delayed train I worked out that this is exactly what many businesses get wrong about AI. They keep buying new vehicles. A flashier model, a newer app, whatever launched last week. The value was never in the vehicle. It was always in knowing the routes well enough that the journey runs itself.

One moment illustrates the point. I was standing in a station with every tool I could want in my pocket. Two journey planners, live boards overhead, announcements, the lot. And I was faced with so many options that I gave up on all of them and asked a member of staff which platform I needed. The relief was immediate. The network was fully wired, and I still wanted a person at the decision point. Hold that thought because it is the whole argument.

The four moves

There are four things you do with a journey, and they are the four things most businesses never do with their own work.

Plan is reading the route before you set off. Which modes, which interchange, what the fallback is if a line goes down. In a workflow it means mapping every step before you run it, naming the inputs, the decision points, and what the person owns against what the machine owns.

Conduct is the journey itself. No single mode gets you from Barnes to Marylebone, and no single tool runs a real piece of work. The skill is the stitching, not the parts.

Track is knowing where you are while you move. A local feels the stops go by without checking. A workflow you understand tells you at a glance which stage it is at and where it is about to fail.

Trace is what you do afterwards. You notice one interchange is a hot slog and the next takes two minutes, so you choose differently next time. It is the review loop almost nobody runs on their own processes, and it is where the real compounding return sits.

Meet Donna

Which brings me to Donna. Donna is what twenty-one years of running the same routes finally produced. We have mapped seven of our core journeys end to end, sixty-three steps in all, each one with a named owner. Those steps run across eleven separate AI systems, and I am not going to name one of them, because the names are not the point and they change by the month anyway. The point is what happens when eleven systems are live at once across sixty-three steps. Left to themselves they collide. They duplicate work, contradict each other, and pull a single job in three directions.

So, we don't leave them to themselves. On the railway, the person who keeps two trains off the same stretch of line, who sits with the whole layout in front of them and decides what moves and when, is the signaller. 

On our network that is us, the team. We decide what runs, in what order, and what gives way. Donna runs the lines we have already learned, the journeys done so often they go to time without anyone thinking about them. She doesn't choose the route or call the moves, and I would not want her to. She runs the ones we have proven, so we can watch the parts that still need a person.

Everyone talks about keeping a human in the loop. We run it the other way round. The humans are the loop, and Donna works inside it, for us. It is AI in the loop, not human in the loop, and after two decades of doing the journeys by hand, it is the only way we would run it.

Issue, crisis, analysis, repair.

This matters most when something goes wrong, which is the work for which we are built. We have run the first hour of more crises than I care to count. Reputation work runs on the same four moves, only faster and with far more at stake.

Most clients reach us mid problem. An issue is already bubbling and nobody has read the route. By the time it is a full crisis, the first hour is the one that decides everything, and there is no plan for it. We have watched a financial services provider outrun by misinformation because no one had mapped the first sixty minutes. Then comes analysis, the quiet work of tracking what is being said, by whom, and where it is heading. One insurer's audience had moved on while the brand stood waiting on the old platform. And then the long haul, repair, rebuilding a reputation over months and sometimes years, with the legal and commercial realities sitting right alongside. We have taken an offshore entrant through a market they had never set foot in.

Issue, crisis, analysis, repair. Plan, conduct, track, trace. The same four moves. The systems make each move quicker. We keep them in the right order, and a named person owns each call. Donna runs the proven steps underneath. None of it removes the human from the decision. It frees the human to make a better one. It allows our team to achieve more.

We learnt to read a network in Auckland. London this week proved the skill travels because a seasoned traveller dropped into a new city is reading it within a few journeys. What carries is the knack of reading any network you are handed. It is the same reason the method holds across fintech, agritech and insurance. Different city, same way of finding your way.

So, if you are weighing up AI for your own business, stop admiring the trains. Put the effort into learning your own routes until you stop thinking about them and keep a person at the point where the decision matters. That is where the return is.