Closing the loop.

An open loop system runs blind. You give it an instruction and hope it follows through. A toaster is open loop. It heats for the time you set, regardless of whether the bread is golden or charred. There is no feedback signal coming back to tell it what’s actually happening.

A closed loop system measures what’s happening, compares it to what should be happening, and corrects itself in real time. A thermostat is closed loop. It senses the room temperature, compares it to your target, and adjusts. Cruise control is closed loop. Self-driving cars are closed loop. The legged robots I used to work on at Imperial and Ocado are aggressively closed loop. Without that constant stream of feedback, they would fall over within milliseconds.

Every reliable system in the modern world runs on closed loop control. With one strange exception.

Most large enterprises run their frontline teams in open loop mode and don’t realise it.

Here is how the open loop works in practice. A company builds a new product. It writes a sales playbook. It rolls out training and asks frontline staff to sit a multiple-choice exam. Pass marks are recorded. Compliance is satisfied. The product launches. Months later, sales numbers come in. They are below target, or above, or somewhere confusing. The company tries to work out why.

It looks like a feedback loop. It isn’t.

A multiple-choice exam tells you whether someone can recognise the right answer. It tells you almost nothing about whether they can deliver it to a sceptical customer at four in the afternoon on a Wednesday.

The exam is testing recall. The job requires execution. These are different skills, and conflating them is exactly the kind of measurement error that, in any engineering discipline, would get flagged and corrected. You can’t control a system by measuring something other than what you actually care about.

And the sales numbers that arrive months later? Those are real signals, but they are noisy, lagged, and conflated with everything else: market conditions, pricing, competitors, seasonality, the weather. By the time you can untangle them, the next product is already in the pipeline. You’re not closing the loop. You’re doing post-mortems.

This is the gap. And it’s the gap robotics solved a long time ago.

In a robot, you don’t wait until it falls over to learn that the controller is wrong. You don’t wait until the warehouse is on fire to learn that the navigation stack has a blind spot. You build in sensors. You generate a high-quality feedback signal at the point of action, you compare it to what was supposed to happen, and you correct.

Apply that thinking to a frontline workforce and the architecture is the same. The company is the system. The strategy is the input. The frontline workers are the actuators: the part of the system that actually does the work. Revenue is the output. And what’s missing, in almost every enterprise we look at, is a real sensor between the actuators and the output.

That sensor is what Seebird builds. We put frontline staff into bespoke, voice-driven simulations of their actual job, with the real product, the real objections, the real customer. Not a multiple-choice exam. A five-minute conversation that mirrors what they’ll be doing the day after launch. Every session is scored against criteria the company actually cares about. And critically, this happens before the product goes live, while there is still time to act.

If a region is misunderstanding the product, you see it. If a script isn’t landing, you see it. If a particular frontline worker needs more support, you see it. You see it in days, not quarters, and you see it cleanly, before the market signal arrives, conflated with everything else.

That is what closing the loop looks like.

None of this is a claim that simulations are reality. They aren’t. The market is its own control problem, with its own dynamics. But there is a layer of the system that companies can control. Whether their frontline is genuinely ready to execute the strategy they designed. And right now, most companies are leaving that layer to chance. They’re running open loop on the one part of the system that’s most within their grasp.