The average company with 500 employees pays more than CZK 1,000,000 a month for Microsoft 365 with Copilot. Add it up over a year: CZK 10–15 million. And what do most organizations actually use it for? Roughly the same thing they used Office 2003 for: emails, spreadsheets, presentations.
This isn't a criticism of people. It's a criticism of how AI gets thought about.
Most managers assume that an AI layer at a company means a new project, a new budget, and a place in line at the IT department — which is already overloaded with infrastructure debt and has no capacity for anything with a horizon shorter than two years. So it gets postponed. The opportunity sits there.
But this particular solution doesn't need IT. It doesn't need new infrastructure. It doesn't need new licenses. Everything described below costs exactly what your company is already paying today, and runs on technology that's already on every computer in your company.
Human attention is a sieve, not a tank
In every meeting, you have a choice: focus on the people, or focus on the content. On the dynamics in the room, or on what's being said. On what the speaker is saying, or on what they mean by it. Essentially no one can do both at once. We always pay for it with a missed opportunity.
The result? We retain only a fraction of what gets said. Nuance, minority perspectives, indirect signals, seemingly minor comments that turn out to be crucial three months later — all of it slips through the sieve of attention.
An organization that relies purely on human memory and notes operates on a permanently degraded version of its own reality.
The problem is called a cognitive bottleneck
Managers at mid-sized and large companies spend 60 to 70% of their work week in meetings. That leaves minimal time for thinking — for actually processing what was said. Decisions get built on whatever people happen to remember. Strategy gets built on an incomplete picture. Implementation gets checked however time allows.
This isn't a failure of people. It's a failure of architecture.
An organization functions as a distributed memory made up of its employees' heads. This memory is, by nature, selective, biased, and unstable. Every colleague who leaves takes with them context that exists nowhere else.
Why the tools we already have haven't helped
The tools meant to solve this problem (CRM systems, project platforms, knowledge-management wikis) have, paradoxically, made things worse. Their architecture requires people to actively input information. Every new form, every mandatory field, every additional platform added extra work with no direct value for whoever had to fill it in.
The result is well known: systems end up incomplete, outdated, or abandoned. And on top of that, organizations added vendor lock-in — a state where you can't replace a tool without losing data and re-adapting your processes, giving the vendor permanent leverage in every price negotiation.
These tools don't solve the cognitive bottleneck problem. They just move it around.
What changes when you add an AI cognitive layer
A meeting transcript is available automatically in Microsoft Teams. That alone isn't enough. A document nobody reads has no value. Value is created the moment an intelligent layer sits on top of this data — one that can:
Synthesize for different roles. What from yesterday's product meeting is relevant to the CEO, and what's relevant to the lead developer? Those are different takeaways from the exact same transcript. AI handles this instantly, with no extra work.
Maintain context over time. A manager returning from vacation doesn't need to read 40 transcripts. They need a briefing: what got decided, what moved forward, where the open items are.
Detect signals. If the same topic — say, ambiguity around who owns customer onboarding — keeps coming back unresolved for the third meeting in a row, that's a signal. People miss it because they don't have time to compare transcripts. AI sees it automatically.
Frame decisions with context. The biggest value isn't in what AI knows. It's that it knows it at the right moment — exactly when a decision is being made.
The effect compounds over time
Here's the key asymmetry that changes the whole equation.
Accumulated human intelligence within an organization grows linearly, and drops sharply every time a key person leaves. An AI cognitive layer works differently: the longer it runs, the more data it has, the more accurate its syntheses become, and the more relevant its recommendations get.
After six months, it's a different tool than it was in week one. After a year, this layer becomes one of the most valuable things your organization has.
M1–M2: Quick wins | M2–M6: Foundation | M6–M18: Intelligence layer
IT won't get to it. And this time, that's fine.
Technology departments at mid-sized and large companies have their hands full: infrastructure debt, migration projects, compliance requirements, security audits. Anything waiting for real business value — anything that would help managers decide better, faster, with better context — sits at the back of the line.
Microsoft 365 with Copilot plays by different rules — not because Microsoft is a philanthropist, but because the architecture is different this time. Transcripts live in Teams. Documents live in SharePoint. Emails live in Outlook. Tasks live in Planner. Everything connected, everything indexable, all on a platform the company has already been paying for ever since it moved from local servers to the cloud.
The accumulated intelligence described here doesn't require any procurement, any IT approval, or any new project. It's a question of configuration and how existing work gets done — not technology.
This can't be delegated
Anyone can configure the technology. The problem lies elsewhere. Accumulated intelligence only works if management genuinely owns it — not as a project sponsor, but as the product owner of a system they use themselves every day. The change has to come from the top, because it's about how management takes in information, frames decisions, and tracks their implementation. This can't come from a coordinator or an IT specialist. Either the manager does it themselves, or it doesn't work.
Take your five most important recurring meetings: the weekly team meeting, the sales meeting, the project steering committee. Look at each one as a system: what goes into it, how information gets processed within it, and what's supposed to come out of it. Most managers, doing this exercise, discover that the inputs are random, the processing depends on the mood and energy in the room, and the outputs aren't defined at all.
Management that refuses to do this, and waits for someone else to set it up for them, or claims they don't have time for it, can't then complain that the organization doesn't know what it knows and keeps repeating the same mistakes.
Where to start
Accumulated intelligence doesn't appear overnight. But starting is simple, and results come fast.
Three steps that work as an entry point:
1. Turn on transcripts for all internal meetings. Without transcripts, there's nothing to synthesize from. That's the one real prerequisite. Microsoft Teams supports Czech surprisingly well. In Teams Premium, turning it on is a single click in the meeting settings.
2. Set up structured meeting minutes into Planner. Turn a transcript into actions, an owner, and a deadline. This workflow can be implemented today with no custom development — just Copilot in Teams and Power Automate.
3. Store the data wherever feels natural. Don't overthink the architecture. Don't design a SharePoint taxonomy. Start by saving transcripts into a folder that makes sense right now. You can refine the structure once you know what data you actually have. Deal with data-volume problems once you actually have them.
For years, we've talked about agile principles: ship fast, iterate, learn as you go. And for years, we've run into monolithic systems whose entire design and architecture had to be planned out completely before implementation, or they simply couldn't launch.
Today, we have the infrastructure right at our fingertips. You can turn on transcription today. You'll see the output tomorrow. In a week, you'll know if it's working.
And hidden inside this is a transformation that concerns each of you personally: you stop being the owner of outputs (documents, presentations, reports) and start being the owner of a portfolio of AI skills. Every workflow you set up is a skill. Every prompt that works is an asset. And this portfolio grows with you.
Where the return is highest: think in streams
When looking for your first use case, simple logic applies: the higher the frequency of repetition, the higher the payoff from automation. A monthly report has a different ROI than a weekly team meeting that happens four times a month over three years.
So look for the biggest, most regular streams: recurring meetings, periodic reports, routine decision cycles. Start there, because frequency is a multiplier.
And once the system is running, treat it genuinely like a product. That means challenging past decisions just as systematically as you accept new ones. Remove things just as readily as you add them. A simple system that actually works is more valuable than a complex one nobody understands.
These systems roughly double their capability every seven months. Instead of spending a hundred percent of your time on coordination and producing outputs, you suddenly spend a third of your time on designing and developing the system, a third on developing your own skills, and the rest on oversight and the activities that can't and shouldn't be automated.
Want to see how accumulated intelligence works for us at Kogi with MS Copilot? We'd be happy to show you, and talk through what a concrete implementation could look like at an organization like yours.