So what’s the problem, is the AI overhyped, or was it the wrong fit? or is there something deeper at play?
This isn’t just in the big tech world, increasingly mid-market B2B companies have made similar investments. Reports look better, there’s a wealth of written content explaining things, there’s more stats, but has anything in the underlying business actually improved? It can feel like a replay of many CRM upgrades, marketing automation rollouts, and other digital transformation projects in the past couple of decades.
Most of the time, it comes back to a lack of clarity on how the systems underneath actually work. Putting more tools on top of that can magnify and speed up the same confusion and mistakes, rather than fixing them.
The emerging competitive advantage
For years, having a company that is systemised has been a clear advantage. Standard operating procedures, documented processes, defined roles, repeatability are all essential for scaling up – or selling. But there’s a level above that that most companies don’t see – and that’s a real and complete picture of how your whole company works, as a system.
This means the knowledge that’s actually in an offline spreadsheet somewhere, or distributed across your sales reps’ brains, or in the shortcuts your staff use to get around the things that don’t quite work as they should, or in all the customer and prospect interactions that aren’t tracked because they’re not as easy to measure as clicks and likes.
It all works – but if AI can’t see this system, it can’t improve it. The companies who can see how their system works from end-to-end, will increasingly be able to use AI tools to scaleup effectively.
As AI becomes more powerful, enhancing and threatening every industry, those companies that are legible and coherent will have an enormous advantage, and those that aren’t risk being left behind, or even worse rolling out a failed implementation that actually magnifies problems.
What do you even mean by a system?
When people talk about systems they usually mean documented Standard Operating Procedures, or some software, or an org chart – but these are all just pieces of the whole. A system is made up of the elements involved, the way those elements connect, and the purpose the whole thing ends up serving.
It’s the connections within a system that define it – not just static components. This means how information moves, how decisions get made, what happens when something unexpected comes in, where the real information is stored, and how it is used.
All systems have feedback loops; where something that happens downstream feeds back into the system earlier and changes things. Depending on how they are set up they can have both positive and negative effects.
In a business system, a key aspect of feedback is getting useful information in a timely way so that you can act on it. We’ve all stayed in a hotel where the hot water isn’t quite up to scratch. It’s freezing cold, so you keep turning up the hot tap to no effect, then after a few minutes, it becomes scaldingly hot, so you turn it down. At first nothing happens, then it’s freezing again. There is a delay between the action and the feedback, making it almost impossible to adjust the system effectively.
Systems rarely behave in exactly the way they were designed on paper. People adapt, workarounds appear, small shortcuts get normalised, conflicting incentives arise. These could all be good and useful in themselves – but over time, you end up with something that functions well, but is harder to describe than anyone would like to admit.
What this looks like in a real company
Ending up in this situation – where the understanding of the system is fragmented and unclear – is very normal, particularly for successful, fast-growing companies. They start simple, then they grow, they add people, they add software, they document things. But over time, things change, people move on or find better ways, the market shifts, new software is built on the remains of the old.
Everything makes sense at a local level, but the broader system is no longer completely visible or understood by anyone.
Which means that when things go wrong, whether dealing with a new external threat, or even just minor internal confusion or misalignment, the root cause is not always clear – meaning fixes often don’t have the expected results. Companies end up managing the symptoms rather than the causes, and then building reporting to show how well they’re managing those symptoms.
Looking closely at typical Australian B2B companies, these signs of a system under strain show up in situations like;
- You have a CRM but people aren’t using it quite the way you would like
- You’ve got some nice dashboards, but you’re not completely confident they report onwhat you really need to know
- Sales are going well, but the pipeline is unpredictable
- Marketing is focussed on activity and numbers, but you’re not completely sure how itconnects to revenue
- Teams like sales and marketing disagree with each other on what’s important
- There are probably more opportunities out there that you are missing
- There’s interest in, or pressure to use, new automation and AI tools, but you’re not surewhere to start or if it’s really a good idea
These issues occur in profitable, well-run, businesses with long-standing customers. Which is part of the problem. Things work well enough that no one is forced to fully map how they work.
The first-order result often looks good; numbers go up, reports look busy. The second-order effect is where the damage sits – misaligned incentives, inaccurate data, duplicated work, and a system that becomes harder to fix every time you “improve” it.
The AI danger
AI tools have arrived with promises to remove friction, see patterns humans miss, save time and improve decisions. All of these are true and achievable, with the right implementation. But when tools are added to fragmented systems, the results can be very different.
AI can amplify and speed up everything, but if it’s trained on incomplete data, all you get is better-looking reports further confirming the wrong thing, but doing it with more confidence. Or marketing reaching even more of the wrong people. You end up with processes that are very good at giving you slightly wrong answers, and they do it so well, and the implementation costs have been high enough, that people are even more likely to trust those wrong answers.
If the underlying system is coherent, the results can be impressive. If it isn’t, you tend to scale confusion in a more sophisticated way.
What legibility means
This is where the idea of legibility becomes extremely useful.
Legibility is the ability to see and understand how your system actually works, end to end, without relying on guesswork or folklore. Whether you are a human or a bot.
It doesn’t mean perfection. It doesn’t mean eliminating every edge case, or forcing things into a rigid structure.
It does mean you’ve got enough clarity that you can see and understand how your company works in reality. Problems can be traced back to their source. You can identify the metrics and measures that matter, the ones that will actually show you if something is going right or wrong. And you can implement fixes, improvements and software upgrades with confidence that they will work, and that you can see and understand why they’ve worked, and where to test them.
There are simple ways to test how legible your systems are. Pick a recent deal, or a recent problem, and try to map it properly. Not the ideal version. The real one. See how many points rely on “it depends” or “that’s just how we do it”.
That gap is where legibility breaks down.
Even the most coherent and legible companies don’t stay in this state easily.
People move roles. New tools get introduced. The market shifts. Incentives change slightly. Small exceptions get made for good reasons.
And over time, the system drifts again. Legibility fades in the background, just gradually enough that no one quite notices until things start to feel harder. This is just what complex systems do.
On one path; fragmentation leads to more patching. Each patch adds a bit of complexity. That complexity reduces visibility, which leads to more patching. It can go on like that for quite a while.
But keeping a focus on maintaining legibility; checking reality against the map regularly – and updating the map when it’s needed, and making changes in the real world when that’s needed – will keep things as coherent as possible.
The shift that’s coming – oh wait, it’s here
In the past, the best companies were often the most efficient. They executed well. They scaled what worked. That still matters, but it’s no longer the full story.
As tools become more powerful and more accessible, the constraint shifts. All of a sudden, small nimble entrepreneurs can effectively deploy an AI workforce if they have a working concept. It’s less about what you have, and more about how well you understand what you’re doing with it.
The companies that can see themselves clearly will be able to test ideas, integrate new tools, and adapt – taking what they’ve always done well, and doing it on a much bigger scale.
