Your company can see faster than it can act

That was true when information was scarce and analytical capability was a genuine advantage.

But the advantage has shifted.

Data itself has not become less valuable. Proprietary data is more valuable than ever. What has become commoditized is the ability to analyze information and generate insight. Capabilities that once required years of investment now arrive as products that every competitor can buy.

When everybody can see, seeing stops being the advantage.

The finding arrives. It waits for triage. Then an owner. Then prioritization. Then a change window. Then an approval from someone waiting on another approval. Very little of that time is work. Most of it is waiting. That gap deserves a name because you cannot manage what you have not named. Call it decision latency: the time between signal and action.

Every approval gate was introduced for a good reason, usually after something went wrong. A production outage led to change reviews. A bad contract meant legal approval. A costly purchase created a committee. A supplier issue added another form. Each of those was a sensible response to a real incident. And every single one of them is still there.

That is the ratchet. Organizations add controls after failures but rarely remove them once the risk has passed. The benefit of deleting an approval is shared by everyone; the risk of deleting it belongs to whoever signed it off. As a result, layers accumulate.

Ten years of incidents becomes ten years of process. Every single delay has a defensible answer. The total does not. That is why decision latency is so difficult to recognize from the inside—and why it has become one of the biggest competitive differentiators.

Nobody designed this. It accumulated.

Start with the clock: how long does it take to move from knowing something to doing something?

Then break that time down into four measurements:

Now apply that same clock to security findings, customer complaints, pricing decisions and capacity requests.

Almost nobody measures these. Everybody should.

Better dashboards. Richer insights. More recommendations. More risks identified. More anomalies detected.

Then every recommendation enters the same approval process that took four months last time.

That is not an improvement. It is simply increasing the arrival rate at a queue you never widened.

The organization sees faster. It still acts at the speed of the change board.

There is also a second-order effect that rarely appears in the business case. Every recommendation that goes unactioned teaches people to ignore recommendations. Generate enough of them and you do not improve decision-making—you create alert fatigue for your most valuable staff.

Which brings us to what happens when the warnings stop being read.

Knight Capital provides one.

In 2012, the firm deployed new trading software to eight servers. One server was accidentally left running an older version of the code. A dormant testing function, never removed from that server, was reactivated when the market opened.

The systems had already detected the problem. Between 8:01 a.m. and the opening bell, they generated 97 warning emails identifying the affected system. Nobody treated them as alerts, so nobody acted.

When trading began, the faulty server executed millions of unintended trades. Forty-five minutes later, Knight had lost $460 million and the company was effectively finished.

Look at what actually failed.

Not automation. Dead code that had never been removed. A deployment that wasn't verified. No kill switch. And 97 warnings that nobody read.

Equifax had four and a half months. Knight had 89 minutes. Same problem. Different clock speed.

So no, the answer is not to take the humans out.

The answer is to take the waiting out. Those are very different things, and many organizations confuse them.

But regulation does not require slow decisions. It requires control. 

Modern regulations such as DORA (Digital Operational Resilience Act, since 17 January 2025) focus on whether organizations can demonstrate what happened, when it happened, which policy governed the decision, and whether the process can be audited afterwards. Nowhere does it require that a human hand performs the action.

So think about which of these two is genuinely easier to evidence.

A policy engine that evaluates decisions against defined rules and records every action. Or a change advisory board that met on Thursday, where the reasoning lives in the memory of whoever chaired it and the permanent record is a ticket comment saying "approved".

One of those is auditable. The other is a story people tell afterwards.

Regulation is not the reason your decisions are slow. It is the reason you were given for why they are slow.

Judgment is deciding whether a trade-off is acceptable: take the risk, accept the outage, spend the money, disappoint the customer. Humans should own those decisions. They should do it deliberately, in advance, once, and write it down as policy.

Approval is confirming that an existing policy applies to a specific case. It is administration.

Encode the first. Automate the second.

Jeff Bezos, the founder and former CEO of Amazon, described this distinction through the idea of one-way and two-way doors. 

“Some decisions are irreversible and deserve careful consideration. Most are reversible and should move quickly.”

Many organizations have accidentally applied the one-way-door process to everything.

The question is simple: how many of your approvals are protecting decisions that could easily be undone?

A system that acts within clear boundaries, logs its actions and escalates exceptions is not the Knight Capital scenario. The lesson from Knight was not "do not automate." It was: “automate with guardrails, limit the blast radius, verify deployments and make sure warnings reach people who will act on them.

When I priced the same scope using traditional delivery models—day rates, project plans and normal organizational processes—the estimate was many times higher and months longer.

The estimate was not wrong about the work. It was correct about the machinery surrounding the work.

That is the gap.

Not skill. Not tooling. Latency.

Measure:

It is how quickly you move from knowing to doing—and how many people need to say yes along the way.

Your competitors are about to see everything you see.

That was never the hard part.

Fujitsu’s own Uvance work points to the same challenge: helping organizations shorten the distance between data, decisions, and action. Our Data & AI approach is designed to compress decision cycles from days to hours, while our agentic AI model reflects the distinction at the heart of this article: 

“Humans remain accountable for judgment, while systems execute within defined, recorded, and auditable boundaries.” 

The technology to do this already exists. The question is whether your organization knows where its decision time is actually going.

If you want a second pair of eyes on those numbers, or on what could safely be encoded as policy in your environment, get in touch. Bring the twenty examples. We will start there. From there, we can identify where AI, automation, and better decision design can safely shorten the path from signal to action.