Your team already uses AI. Your company doesn't yet

A few days ago, the CEO of an organization called me. He called me with a deadline looming.

AI wasn’t in his plans. He had an organization to run, targets to hit and other priorities on the table. But the board had already decided: the organization was going to adopt AI, and he had three months to make it happen.

It’s not a comfortable position. You’re asked to lead a change you didn’t choose, on a topic you haven’t mastered, with a clock that starts running on day one. And any mistakes are visible to the people who set the deadline.

He started studying. And the more he read, the more overwhelmed he felt. New models every week, tools promising to change everything, terms he didn’t know.

On top of that, he knew something that worried him: people in his office were already using AI. Each in their own way, with whatever tool they wanted, with no policy and no approved tool.

AI had come into his organization the way WhatsApp came into the workplace: everyone on their own phone, with their own chats with customers, without anyone deciding it.

What does it cost when everyone uses their own AI?

Informal use has a very good side: the team wants to use AI, and that’s the hardest thing to get. But it has costs you don’t see.

The best agent in the department was built by one person using their personal account. When they go on vacation, no one else can use it, and if they leave the company, that knowledge leaves with them.

Company information moves through personal accounts, without anyone having decided what can be uploaded and what can’t.

And everyone moves at their own pace: some people already delegate half their work, and others have never opened the tool.

Everything changes every week

AI keeps advancing every week, and every advance comes with headlines promising that now, this time, everything changes. On top of that, every tool the company already pays for announces its own AI feature.

For a leader who is just starting, every piece of news feels like a pending decision. That was what overwhelmed this CEO: he had more information than he needed, and all of it seemed urgent.

Two paths that seem logical

Wait for things to settle down. If everything changes so fast, it seems sensible to wait until it stabilizes. But while you wait, your team keeps using AI anyway, just without rules.

Pick the best tool first. This is where most companies get stuck: comparing, asking for demos, looking for the perfect option. The best tool this quarter probably won’t be the best one next quarter.

How do you bring AI into a company?

In the past year, I’ve worked with dozens of companies on their AI adoption and designed more than 110 processes with AI. At Kavanta, we’ve also spent thousands of hours researching and analyzing how these tools fit into companies.

There’s more than one way to bring in AI. The one I recommended to him, because it gets the most out of the tools for what they cost, is to turn them on in order.

That order has one rule and four layers.

First, the rule

The policy has three parts: ban free AI tools for work, buy business plans and don’t pay for anyone’s personal plan.

No free tools for work. On a free account, the company loses control over the information that gets uploaded. Depending on the tool and its settings, those conversations may be used to train models.

Business plans. With a business plan, the organization manages who has access and decides what happens with the information, and what the team builds (skills, agents, projects) belongs to the company.

Don’t reimburse personal plans. Paying for each person’s subscription looks like a generous shortcut, but the account still belongs to them. When they leave, they take their conversations, their agents and everything they learned to do with them. The company pays for something it can’t manage.

Layer 1: AI where your team already works

Almost every company works on Google Workspace or Microsoft 365, and each comes with its own model: Gemini in Google, Copilot in Microsoft.

This layer is often underestimated, because it seems basic. But it’s exactly where your team spends its day.

Email, documents, spreadsheets and meetings are already there, and AI can summarize a fifty-email thread, draft the reply, analyze the sales spreadsheet or take meeting notes, without anyone switching tools.

This layer also puts everyone on the same foundation: the same AI, with company accounts and under the same rules. Many plans already include it, and even so, many companies have it and don’t use it.

If you work with Google, Gemini does very good work, and it costs less than Microsoft’s AI.

Layer 2: an agentic platform to delegate digital work

Today, a large share of the digital work people do can be delegated to AI: researching, analyzing, writing, calculating, preparing, following up.

An agentic platform like Claude or ChatGPT plans, works toward a goal and carries out tasks with your tools. And for a company, the most valuable part is that it captures how the work gets done:

• With skills, the way to do a task is documented and repeated the same way every time.
• With agents, a multi-step process runs from start to finish.
• With routines (scheduled tasks), the work runs automatically at the right time, like the sales report on Monday at 8.

That’s the fundamental change. Knowledge that used to live in one person’s head or chat history now belongs to the company.

In the processes we’ve measured after they went live, digital tasks are delegated 75% on average, across marketing, sales, finance, human resources and operations. Some reach 99% and others stay at 50%. I explained why in this article.

We prefer Claude for business, for the quality of what it produces and how easy it is to use.

Layer 3: the AI you’re already paying for

Your CRM, your ERP and your prospecting tool are almost all adding AI today. Some have done it very well; others added it so they wouldn’t fall behind. And in many companies those features are there, paid for, with no one using them.

Skipping this layer has a double cost. You can pay twice for the same thing, building or buying something your current tool already did. Or you can stay stuck with a tool that fell behind and holds everything else back.

This layer requires piloting: take the tools you already use, test their AI features on a real process and decide with one question: does it connect to the platform you chose in layer 2?

If it doesn’t connect and doesn’t bring useful AI either, it’s worth replacing. An isolated tool usually ends up costing more in hours than it saves in license fees.

It happened to us at Kavanta. Our CRM launched built-in smart chatbots, but we had never used them, because we had built our own outside the CRM. When we piloted the CRM’s chatbots, we found they worked better and, because they were integrated, they lowered our costs.

Layer 4: connections, the things that have to happen on their own

The first three layers cover the work someone asks for. What’s missing are the tasks that pass information between platforms: the data that comes in through a form and has to move to the CRM, or the closed sale that has to reach billing.

That’s what workflow tools like n8n, Make and Zapier are for. They connect to multiple platforms and can also use different AI models along the way.

An agent works inside a conversation: someone asks it to prepare tomorrow’s meeting, and the agent researches, puts the document together and delivers it.

A workflow doesn’t wait for anyone: it’s triggered when something happens, like a new form or a payment received, and it always follows the same path, at any hour.

Agents are for work that requires judgment in each case. Workflows are for handoffs that have to happen the same way every time.

Laboro, a leader in home improvement services in Mexico, was growing faster than its team could absorb. It automated the handoffs between the platforms it already used and added a quoting tool that generates estimates on its own.

As its CEO, Daniel Ricchiuti, puts it: “Today we handle in hours what used to take days.” Its conversion from leads to sales rose 66%, with the same team.

What’s next for that CEO

That CEO called me with three months ahead of him and a long list of news. Today he has a plan: which tools to bring in and in what order.

But putting the tools in place is only half the work. The challenge ahead is for his team to learn to get the most out of them across the company, to delegate the hard tasks to AI and, that way, free up time to grow.

Where should you start?

It depends on where you are. If you don’t have anything yet, start with the rule and layer 1. If you already have the foundation, go deeper and move on to the agentic platform. After that come the tools you already pay for, and the connections.

Each layer builds on the one before it.

The tools can be turned on in weeks. What turns them into time to grow is your team learning to delegate to them.