Six months after foundational training, a lot of what I see is Copilot in Outlook being used as a glorified search engine. It’s one of those tasks that AI is pretty consistently good at. It’s a good first step that helps staff get comfortable with the tech.

I also talk to people with big ideas that will change how they work. They’re asking “Can Copilot help with that?” But when they go into Chat to see if it can pull the information they need, it takes a lot of effort, if they can get it to work at all. Results can be inconsistent, and people end up frustrated with the tech. Adoption doesn't climb.

Foundational training did what it was supposed to do. It got staff aligned on what they can safely do with AI, gave them ideas about where it might fit, and taught a basic prompting technique. How impactful is that to the business? Enough to justify the licenses. Probably. Enough to move the needle on your EBITDA? Unlikely.

The instinct is often to pay someone else to figure out the solution, build it, and then roll it out.

For some problems that is the right call (more on which ones in a minute). But if everything routes outside, it gets expensive fast at the volume of work that actually needs doing, and nothing compounds. Staff never learn the skills they need to apply AI to their work: context engineering, coaching AI, writing instructions and skills, and understanding tools like Notebooks.

Short term, a group at intermediate and advanced skill levels can act on the low-hanging fruit without bringing in outside consultants. Real business problems that happen to have straightforward AI solutions, which is more of them than you would expect.

These are the problems that require some experimentation and skill. It’s not the level of output you can expect from an AI 101 course.

Six months into a champions program I run, roughly 70% of the AI work planned for H2 is impactful at the organizational level. These aren’t personal pet projects. They’re notebooks for the entire team. Agents. One person organized their team’s information so anyone could start a conversation with Copilot without fumbling through folders to add the relevant attachments and type out the context AI needed.

This work isn’t accidental. They were taught the technical skills, Copilot’s capabilities, and how to spot an impactful opportunity. They put together individual AI Action Plans with their managers’ input.

Very little of this happens when a consultant builds the solution and hands it off. When the engagement ends, the capability leaves with it.

This is not an argument against hiring specialists. I would hire them.

For the projects with the highest risk and the most potential to transform the business, they make sense. Those are your big bets, and you can likely quantify how much to invest in solving them.

But nobody is scoping a statement of work for the thing that costs your ops team four hours a week, and if you do not have a tech team, there is no internal queue for it either.

So the long tail stays in-house. And it stays as an idea, a question, “Can Copilot do this?”, because your people don’t have the skills they need to figure it out.

Invest in building skills internally

At least one advanced person on every team. That’s the goal I advise my clients to target. These are the people building simple agents, the ones that do one specific thing well. They’re working in Copilot Studio and maybe even Power Automate or Power Apps, but they’re not getting into data engineering or writing and deploying code. A group of advanced people sitting together in IT won’t cut it. They don’t understand the work well enough to build a good solution, at least not without a lot of back and forth.

In most organizations the goal is to get most of a team to intermediate. Intermediate is someone who is good at figuring out how to solve a problem with AI in chat and notebooks. They can set up persistent context so they are not re-explaining themselves every morning, coach a mediocre first output into a usable one, and tell the difference between a limit of the tool and a problem with how they asked. That’s the level where a lot of the magic happens, but it tends to stay at the individual level.

And of course you’ll have a couple of stragglers. People that struggle with change, new technology, or just plain hate AI and want nothing to do with it.

Now building all this skill isn’t easy and it doesn’t come without a cost. I’m not talking so much about the cost of hiring someone awesome like me to teach workshops. I’m talking about the time it takes your people, not just to attend a training once a month, but to carve out time to apply the skills they’re learning to their work rather than defaulting to the fastest way, which is the one they already know. If nobody will protect the time, it fails. If teams don’t have the capacity for experimentation, it fails.

Should you invest in building the real, practical AI skills that change how work gets done?

Not if you haven’t launched a foundational curriculum. First, start there.

Not if you want full control over what gets built. In that case policies and procedures aren’t enough. Build it centrally.

Not if the value of AI sits in one or two hard technical problems instead of a long tail. Just hire the consultants or in-house talent.

But if your organization believes that the real transformation in your business will come from the subject matter experts that do the work, invest in building their skills. Invest in the workshops. Invest in AI Champions programs. Invest in hackathons. Invest in your people.

And yes, I’m here to help, if you want it.