From Open Loops to Closed Loops: How AI Helps a Business Learn From Its Own Work

Most businesses are not short of AI tools. They are short of systems that help them learn from the work already happening.

Every day, customers ask useful questions. Sales calls reveal objections. Support tickets expose recurring issues. Teams find shortcuts. People work around broken processes. Decisions are made, repeated, challenged, forgotten and occasionally improved. But too often, the learning does not compound. It gets trapped in meetings, Slack threads, documents, dashboards, inboxes, call notes and people’s heads.

A customer says something important on a sales call, but it never reaches the right person. A support issue repeats five times, but nobody connects the pattern. Someone finds a better way to complete a task, but it stays inside their own workflow. The business is doing the work, but it is not always learning from the work. That is what I mean by an open loop.

Recently I watched Y Combinator partner Diana Hu talk about closed loops in AI-native companies, and how they can rapidly speed up the rate of development. The real opportunity is not simply that AI helps individuals move faster. It is that AI can help a business learn faster.

A closed loop captures what happened, feeds that information back into the system, and improves the next cycle of action. In business terms, the company gets better at noticing what is happening, understanding what matters, and turning that learning into better workflows, better decisions and better customer experiences.

The promise of closed-loop AI is that individual insight stops dying in private workflows and starts becoming shared intelligence. That is a very different conversation from “which AI tool or agent should we use?” It is about building a learning engine for the business.

I am building a new company at the moment, and one of the things that excites me most is the chance to build from the ground up without some of the legacy I have seen slow businesses down for years. No inherited communication habits that get in the way. No default reliance on meetings to move information around. No outdated ways of working where progress depends too heavily on the right person remembering to update the right person at the right time.

That is not because people are bad at communicating. It is because most businesses were built around systems that leak context. Information has to be carried from person to person, often through departments with different priorities, budgets and incentives. I have seen this pattern repeatedly in digital transformation work. The tools change, but the organisational problem often stays the same: useful knowledge sits in pockets, teams create local workarounds, departments optimise for themselves, and leaders get a polished version of reality, often too late to make the best decision.

It is the old silo problem. Except AI can make the silo one person deep.

Someone in sales finds a better way to produce proposals. Someone in service finds a better way to summarise calls. Someone in operations builds a prompt chain that saves hours each week. All of that is useful. But if the learning stays hidden, the business never gets the full benefit.

This is where culture matters. I believe that people need to be given a good reason to share what they are learning. They need to believe they will not be penalised for being efficient, loaded with more work, or made more vulnerable because they found a smarter way to do part of their job. If they can’t have confidence that leaning in will make their work more meaningful, then as leaders, what are we doing?

Anyone who has spent time around transformation work knows how quickly teams defend their patch. Marketing wants its budget. Sales wants its tools. Operations wants control over the process. Technology wants governance. Finance wants proof. Everyone has a rational reason to protect their own world.

A closed-loop way of working challenges that. It shows where work is duplicated, where handoffs break, where the real blockers are, who is improving the workflow and who is mostly controlling the narrative. That can feel uncomfortable, because it flattens some of the old sources of power.

I remember seeing this clearly around the R&D of a new product. The choices were being shaped largely by the hunches and preferences of the head of R&D. Some of that instinct was invaluable. Experienced people often have pattern recognition that should not be dismissed. But the brief did not properly include the context sitting elsewhere in the business. The information existed, but the process had no reliable way to bring it into the brief.

Sales had a close handle on what customers were asking for. Customer-facing teams could hear the frustrations, repeated requests, objections, moments of excitement and points of confusion. But that knowledge was treated as anecdotal. It did not carry the same weight as the opinion of the person who owned the R&D process.

This is where businesses get stuck. Strong, opinionated people can be brilliant. They can also become bottlenecks when their opinion becomes the operating system. If they do not want anecdotes, contradictory data, or empirical evidence that challenges the direction they already favour, then the business is not really learning. It is orbiting the strongest voice in the room.

AI changes what is possible here. It is now much easier to gather perspectives quickly. You can analyse customer calls, run external and internal polls, cluster sales feedback, compare support tickets, summarise objections and look for patterns across CRM notes, lost deals, surveys, transcripts and product feedback. What used to be dismissed as “just anecdotes” can become a body of evidence.

That does not mean the head of R&D loses their role. It means their judgment gets better inputs. The wall of stubbornness becomes harder to maintain when the business can move from “someone said this once” to “this pattern is showing up across customer calls, sales feedback, support tickets, lost deals and product friction.”

That is one of the big benefits of closed-loop working. It gives the business a better way to test hunches, challenge assumptions and shape decisions around evidence.

It also changes whose voice gets heard. The salesperson who spends all week listening to customers does not have to rely on being the loudest person in the meeting when, frankly, they probably do not care that much. They want to be selling. The support person who hears the same customer irritations every day does not have to feel defeated or powerless, quietly hoping that someone will finally address the problem that keeps reappearing.

The work itself starts to speak.

This is the thread running through all of it: AI is only transformational when individual insight becomes collective intelligence.

The Formula 1 analogy helps me here. The driver gets the fame, but the race is not won by the driver alone. Listen to how often drivers praise, or complain about, the team. The performance of the car, the race strategy, the pit stops, the tyres, the data, the engineers and the communication are all part of the outcome. Probably more than in any other sport, Formula 1 makes the system visible. You can be the best driver on the grid, but if the car is not up to scratch, you are not on the podium.

A good closed-loop business should feel more like that. Not every department is fighting for the limelight. Not every senior person is defending their fiefdom. Not every decision depends on who can argue most persuasively in the meeting. A unit working around a shared outcome, where contribution is visible because the system shows what changed.

That is the better vision. I don’t want to work with clients who are just “using AI.” I want them to be building a learning engine that helps the business learn from its own work. An engine that makes useful ideas easier to see, reduces tedious work, gives people more time for judgment, creativity, relationships and problem-solving, and gives more weight to the people who understand the work, not just the people who are best at controlling the narrative.

That is the part I think leaders need to communicate much better. Instead of people hearing, “Share your prompt so we can squeeze more out of everyone,” they need to hear something more credible: “Show us what you have learned so we can improve the process, protect the quality of the work, and make this better for everyone.”

Will people always get the credit they deserve? No. Many organisations are not good at that. The grafters who quietly hold workflows together are often the last to be recognised. But that is exactly why culture matters. If leaders want people to share what they are learning with AI, they have to make it safer and more worthwhile to do so. People need to see that sharing a better workflow will not simply lead to more work, less security, or someone else taking the credit. And to do that, there’s a massive mindset that needs to shift. There’s a lot to lose; but there’s so much more to be gained. Be brave.

The aim is not individual heroics. It is collective progress. That is when people are more likely to lean in. Not because they have been told to “embrace AI,” but because they can see a better outcome: less tedious work, more meaningful contribution, better customer experiences, clearer processes, fewer fiefdoms, and less power sitting with those who know how to wield influence.

The best insight in a business is often not sitting at the top. It is sitting with the person who has done the same painful task two hundred times and finally knows exactly how it should be fixed. I have seen versions of that person everywhere: the operations person who knows the handoff is broken before anyone else admits it; the customer service person who can name the complaints that never make it into the board report; and the salesperson who knows the proposal template is losing deals because it answers yesterday’s questions.

Those people are often the real operating system of the business. AI can ignore them, automate around them, and make them more anxious. Or it can surface what they know, reduce the work that drains them, and give the business a better way to learn from them. I believe that is a genuine, understated rallying cry many people would respond to, provided, of course, that it is true.

What true closed-loop development looks like

True closed-loop development is not just “we use AI more.” It is a system where work creates artefacts, those artefacts create insight, and that insight changes the next decision.

An artefact is simply a useful trace of work: a call transcript, meeting summary, support ticket, proposal draft, CRM note, product brief, customer survey, internal poll, decision log, quality review or workflow map. The point is that the work leaves something behind that the business can learn from.

A closed loop might look like this:

1. Customer conversations are captured as artefacts. Sales calls, support calls, emails, chat transcripts and meeting notes are recorded, summarised and made searchable, with the right privacy and consent boundaries in place.
2. Repeated issues are clustered into themes. AI groups recurring objections, complaints, requests, delays or misunderstandings so the business can see patterns rather than isolated anecdotes.
3. Internal and external perspectives are added. Sales, support, operations, delivery and leadership can add context. External customer polls, internal staff polls and short surveys can test whether the pattern is real, widespread or urgent.
4. The team reviews the pattern, not just the anecdote. Instead of one person saying, “I heard this from a customer,” the business can look at evidence across calls, tickets, CRM notes, surveys, lost deals and frontline feedback.
5. A decision is made and recorded in a decision log. The business changes a workflow, service, product feature, onboarding process, support handoff, proposal template or escalation rule. The reason for the change is captured so people understand why it happened.
6. The outcome is measured. Did response time improve? Did customer frustration reduce? Did the proposal turnaround speed up? Did quality hold? Did rework fall? Did conversion improve? Did the team save time without increasing risk?
7. The learning feeds the next cycle. The new workflow becomes the baseline. The business keeps monitoring what happens next, using the latest artefacts to improve again.

That is not bureaucracy. It is how the business stops losing what it already knows.

This is why I find the idea so compelling as I build new products and services myself. I do not want AI to become another layer of noise, or a private productivity party trick that never changes how the work actually happens. I want to build with this thinking baked in from the start: fewer leaky handoffs, less dependence on memory and status updates, and a better way to capture what is being learned as the work develops.

The promise of closed-loop AI is not just that the business gets faster. It is that the business gets more honest. More honest about what customers want, where work breaks, who is improving it, and what needs to change next.

And that is an aspect of AI I find incredibly exciting.