How to build AI Skills your team will actually use

I talk to a LOT of marketers in my role. And one of the most common things I hear across industries is that orders from above sound like, “Find a problem for this AI solution we already paid for.”

That inevitably leads to performative use cases that don’t improve anything and don’t stick.

If you feel the same way, today’s expert shares how to build helpful workflow automations you’ll actually use.

If you don’t feel that way, that’s still today’s topic; you just won’t relate to the snark.

Our resident AI engineer, Tim Biondollo, is back with the story of how he helped one team reduce content creation time by up to 38%. Plus, tips for how you can apply this process to almost any kind of marketing task.

The skillz to pay the billz.

One of our marketing teams faced a double whammy: A looming deadline and a lack of writer bandwidth. Plus a whole toolbox of AI solutions that weren’t helping either problem.

Leadership called in… Tim. (I like to imagine it on a big, red Batphone.)

“What the writers were complaining about was unnecessary friction. Things were adding time without adding to the quality of the writing. [The problems] were mostly on the technical or research side, not the writing side itself.”

Tim realized the issue was a lack of AI Skills. Not in the sense of “knowing how to use AI,” but Skills with a capital S.

An AI Skill is a set of saved instructions that tells an AI tool how to handle a specific task. Instead of having to manually type out a prompt each time, the Skill recognizes the context and automagically works its mojo.

“A Skill recognizes that it needs to fire from what you say in the chat and does so automatically. The [user] doesn’t have to remember anything.”

Together, they created a suite of automations that fetch live product info, surface competitor research, catch hallucinations, and flag editorial standards. The result was an average of 112 hours returned per month.

Cool story; now, how do you do it?

Since those specific automations are customized to the team’s needs, sharing the exact Skills would be pretty niche. The useful part is how Tim finds and creates useful Skills. And you can apply this to almost any of your workflows.

1. Discovery phase

“Most AI users are trying to work bottom-up — fix the immediate friction point — when top-down is actually the more powerful approach,” Tim tells me. “If you’re looking at the big picture and working down from there, it changes what you decide to do. It becomes less about fixing and more about generating value.”

To that end, Tim started by talking with the leadership team.

“At that point, I was in order taking mode, which is really where strategic work begins. Our first order was, ‘We’d like our writers to be able to hit our volume goals [but] they have limited bandwidth.”

From there, “I went down a level and spoke with the manager of the writers. Asked what problems he’d been hearing, what came up frequently, and what he cared most about writers still owning.”

Though each marketer had an individual process, this allowed Tim to understand what common tasks created bottlenecks.

2. Work out the bottlenecks in chat.

The next step is to work through those problem areas “manually” within the AI’s chat function.

“You don’t have to build the Skill first. Go spend an hour in a chat working on the [task] you want to automate,” he advises. “Get to what ‘good’ looks like, then say: ‘Turn this into a skill so you remember how to do it next time.’”

For example, a common bottleneck for me was cleaning up interview transcripts. First, I might ask an LLM to correct any transcription errors or attribution mistakes. Then, I’d tell it to scrub out filler words like “um” or “y’know.” Finally, I’d ask it to strip timestamps and formatting that makes it difficult to read.

Once I was happy with the result, I asked it to turn all of that into one Skill so that I didn’t have to prompt all those steps each time.

“It might take a few tries, but you’ll get a much better result than trying to spec out a full workflow from scratch.”

3. Start a feedback loop.

One of the coolest things about Skills is that they don’t have to look the same for everyone and they don’t have to stay the same over time. You can iterate and test new workflows and features simply by asking nicely.

“We’d had adoption problems with AI tools in the past,” Tim explained to me. So, when measuring the success of each new Skill, he started by asking, “Did it do what it promised? Did it save time? What were the most useful parts? What weren’t?”

But then a funny thing happened.

“One thing we didn’t expect: some writers started customizing the skills to their own workflows. Since the AI can help you update or modify a skill, they’d see it saving them time and think, ‘If I tweak it a little for how I work, it’ll save even more.’ That’s a good sign for adoption metrics.”

Tim’s Tricks for Skills that Stick

Even though the process is literally as easy as talking it out, I asked Tim what separates a really good Skill from a mediocre one.

Here’s the advice he gave:

• Map the existing workflow first.

“Understanding the process and breaking it down into a workflow of steps is harder than it sounds, because writing is a very personal kind of work. But that’s how you save time on things [you’re] measured on, not just the things that are easy to automate.”

• Scope what’s worth automating.

“It’s easy to get caught up in doing things faster rather than asking whether the thing is worth doing at all,” he notes. “Before you go deep, zoom out. Think about what the long-term goal is, what the highest-value task is, and where AI time savings actually translate to business outcomes.”

That last point is one of the most crucial.

“Let’s say you save a bunch of time on a newsletter that has zero audience. You’ve saved time, but you haven’t generated any value.”

• Automate tasks, not judgment.

“Use the AI as a tool. Don’t trust anything it says until you’ve gotten it to a place where it feels trustable for that specific task,” Tim cautions. “Then it’s about applying your judgment at the highest leverage point. Not throughout. Just where it matters most. That’s the actual efficiency gain.”

Even though Tim’s job is building AI processes, he’s quick to hammer on the importance of human processes.

“We talked a lot about how to use AI and automation thoughtfully, not to replace writers. [Marketers] have a deep, tribal knowledge of your product, your editorial standards, and your audience. That’s not easy to extract. Lose them and you lose a significant piece of knowledge that was working for your team.”

• Revisit past workflows.

One of the most powerful ways to apply this process isn’t in looking forward, but revisiting workflows you’re already using.

“If you have past-chat search turned on, you can open a new chat and say, ‘Go look at what I did last week. Help me map out the end-to-end workflow, identify the friction points, and suggest where Skills could save me time without trading off quality.’”

And even here it’s about augmenting human judgement.

“Let the AI do the analysis for you, then you make the call about what’s worth building.”