Opinion · MarTech · Real-time decisioning

You don't have a data problem. You have a timing problem.

Enterprises keep buying more data. The customer moment keeps passing.
The gap is not what you know. It is when you can act on it.

Think back to the last time your bank got the timing wrong. The loan offer that landed the week after you had arranged finance elsewhere. The "we've missed you" message that arrived once you had already moved your salary. The fraud alert that reached you after the money was gone. None of these failed for want of data. The bank held everything it needed. What it missed was the moment.

It is the same story inside the business. Walk into almost any large enterprise and ask why a customer churned, why an offer missed, why a transaction slipped through, and you will be shown a data project: more sources, a bigger warehouse, a new platform to unify it all. The instinct is to collect more. But in most of these stories the data already existed. The signal was captured. What failed was the clock.

A decade of customer technology spending rests on a misread. The problem was rarely a shortage of data. If anything there is too much of it: Forrester estimates that 60 to 73 per cent of enterprise data is never used for analytics, and surveys of business and IT leaders routinely find most organisations treat more than half of their data as "dark." The constraint was never collection. It was the lag between the data arriving and anyone being able to act on it.

The concierge already solved this

The best hotels worked this out decades ago, and they did it without a data lake. A guest mentions once, in passing, that they like to walk before breakfast. The next morning the forecast turns to rain, and as the guest crosses the lobby at eight the concierge is already waiting with an umbrella. Nothing about that moment required more data. The hotel knew the guest's routine and the hotel knew the weather. What turned two ordinary facts into a moment the guest remembers for years was that they were brought together and acted on at exactly the right time, unprompted. Ritz-Carlton has a name for this, anticipatory service, and a daily discipline behind it that staff call "radar on, antenna up." The instruction is never collect more about the guest. It is notice the signal, and act on it now.

Banking holds incomparably more data about its customers than any concierge holds about a guest. What it usually lacks is the umbrella moment: the ability to put two known facts together and act, unprompted, in the few seconds while it still matters.

An old promise, rarely kept

None of this is a new idea, and that is the point. The industry has promised it for thirty years, from one to one marketing in the 1990s, through "moments of truth," to micro-moments, to today's "segment of one." What has not changed is the delivery. Around 90 per cent of companies say they invest in personalisation, yet only about a quarter believe they achieve it at scale, and only one marketer in three can activate the data they already hold. Brands estimate they personalise roughly 60 per cent of experiences; customers put the figure closer to 40. A promise repeated for three decades against a gap that has barely moved does not mean the ambition is wrong. It means something structural keeps defeating it. There are two such things, and the better known is the smaller of the two.

The batch was never a strategy

Overnight data loads, weekly campaign calendars, monthly model refreshes: these were not deliberate design choices. They were workarounds for the compute, storage and bandwidth limits of an earlier era. Processing was expensive, so you batched it and ran it at night. Those limits have largely gone. The habits have not. In most enterprises, customer data still moves on a schedule that has nothing to do with when the customer actually does anything. The customer behaves in real time. The stack responds in batch, and calls it a strategy.

This is where the conversation usually goes wrong. "Real time" is on every vendor's slide; it is sold as a feature you can buy, a faster setting on the same machine. It is not. Timing is a property of the architecture, not a tactic in a campaign. A stack assembled around overnight loads and scheduled sends cannot be made to act in the moment by turning a dial, however much intelligence is layered on top. The honest question is not whether a platform says real time. It is whether anything in the path between a customer's action and the business's response was built to operate in the moment at all.

Where the time goes

Trace a single customer event through a typical stack. It is captured, then waits for the next ingestion job; it lands in the warehouse and waits to be queried; an analyst, eventually, builds a segment; a manager, eventually, approves a campaign against it; the message goes out, often days after the behaviour that prompted it. Each step is reasonable on its own. Together they make a system that answers in hours and days, for customers who decide in seconds and minutes.

What changes when you act in time

The value of a customer signal is perishable, and the window is short. Customers spend only a few minutes a day in an app. Map those minutes across a day and the openings are everywhere, exactly as the concierge's are. The balance check before the school run. The contactless tap on the commute. The low balance found at lunch, the day before a direct debit is due, a moment to move money or extend a small buffer before a charge bounces, not to apologise for it afterwards. The salary that lands mid afternoon, the moment to nudge a transfer to savings. The loan product browsed on the sofa in the evening, where an answer in the session converts and a follow up email two days later does not. The monthly review that never quite happens unless something prompts it. Each is an umbrella moment: two known facts, brought together and acted on while it matters.

And the highest value moments cannot wait at all. A fraud signal is worth acting on before the transaction settles, not in tomorrow's report. A sharp drop in a good customer's activity is worth meeting before they call to cancel, not in next month's churn analysis. An abandoned onboarding step is worth fixing in the session, while the customer is still there. In every case the data existed. The only question that mattered was whether the bank could act inside the window. Most cannot.

Collecting more data does not extend the window. The concierge did not need more data. The concierge needed to act before the guest reached the door.

Why the stack cannot do it

This is not a failure of any one product. The data platform is built to store and unify; it does not act. The analytics layer is built to explain what already happened; by the time the insight surfaces, the moment is gone. The campaign tools are built to fire on calendar dates, not on what a customer just did. Each component is good at its job. None of them closes the loop in the moment, because the architecture as a whole was designed around the batch, and a batch architecture produces batch outcomes no matter how much data you feed it.

The technology moved. The operating model didn't.

Architecture is only half the reason, and the smaller half. Suppose the pipes were instant. A signal still has to become an action, and in most organisations that path runs through people: someone defines the segment, someone writes the message, a manager approves it, brand and compliance sign off. That chain runs on human time, measured in days, not on machine time, measured in milliseconds. No enterprise can manually approve ten million separate decisions in the moment, so the bottleneck stopped being compute long ago and became coordination. The technology got faster. The decision rights and the workflow did not.

This is why the umbrella moment is rare even where the data and the tools already exist. Ritz-Carlton solved it not by giving the concierge more information but by changing the operating model: any employee may spend up to two thousand dollars to put something right for a guest, in the moment, without asking a manager. The result is anticipatory service at the front line.

Acting in the moment is an authority problem before it is a technology problem.

The concierge that never sleeps

This is where artificial intelligence earns its place. A human concierge can keep radar on and antenna up for a few hundred guests; an AI agent can do it for ten million customers at once, reading every signal, weighing intent, propensity and risk in the moment, and acting unprompted while the window is open. That is the umbrella moment at the scale of a national bank, and it is new.

It also does what no front line could at scale: it collapses the approval chain. People set the policy and the guardrails once; the agent makes the decision within them; people audit the outcomes. The human role moves from approving every action to governing the system that takes them. That is the only way the coordination bottleneck is ever removed, because you cannot hire your way to ten million decisions a day.

But intelligence does not fix timing on its own, and that is the trap. The smartest model still loses the moment if the signal reaches it tomorrow; an agent bolted onto a batch stack is a brilliant concierge handed yesterday's guest list and last week's weather. AI is necessary, not sufficient, and it has to sit on a loop that already operates in the moment. In regulated industries there is a further condition: an agent that acts autonomously in under a second must also be auditable after the fact, able to show which signal it read and which rule and model it applied. Real time and accountable, not one at the expense of the other.

What to do about it

The diagnosis points to a short, practical agenda, and none of it begins with buying more data. Measure time to action, the elapsed time from a signal appearing to something happening for the customer, per journey, on the dashboard. Separate the moments that truly need real time from those that do not: fraud, churn and onboarding live or die in the window; a quarterly statement does not, and pretending otherwise wastes the effort. Work backwards from the highest value perishable moments rather than trying to rebuild everything at once. Move the decision to the event, not the schedule, as a principle rather than a project. And start with one moment, prove you can act inside its window, then widen. The aim is not to be fast everywhere. It is to be in time where being in time is what creates the value.

If that number is measured in hours or days, no extra data source will change it, because data was never the constraint. The next data project will produce a richer, more complete, more unified picture of a moment that has already passed.

You don't have a data problem. You have a timing problem.

What kind of layer is fast enough to act in the moment and accountable enough to be allowed to is a question this series will keep returning to.