Originally published by Johnny Campbell CEO at Social Talent. Read the original article here.
You’ve built the same report every month for years, pulled from half a dozen systems, stitched together by hand, meant to tell leadership what’s actually happening across the business: where the friction is, where the risk is, where the patterns are hiding. Days of it. Every month.
If any of that sounds familiar, you already know where this is going… This is exactly the kind of work AI should be doing for you: structured data, the same questions asked on a loop, patterns sitting there waiting to be found. Textbook case for what the tools are good at.
So why isn’t it happening?
I sat down recently with a Recruiting Ops leader supporting the TA function at a global company, tens of thousands of hires a year, who manually builds a Power BI report that lands on her VP of TA’s desk every month. She’s not junior. She has a team under her who look to her to know this stuff. And she still didn’t have it. Not for lack of access (she has AI). Nobody’s ever shown her, on her own report, what doing it differently looks like. And the instinct when people notice a gap like that is almost always wrong: send her on a course, teach her prompting, hope it sticks. It won’t.
What are we getting wrong?
Most advice on “getting your organisation using AI” is aimed at the wrong target.
It’s not a prompting problem. Nobody needs a course in clever phrasing anymore, the models have moved past needing that. What actually matters is knowing which task is worth pointing the tool at, and having your data ready when you do.
It’s not a training-programme problem either. MIT’s Project NANDA reviewed over 300 enterprise AI deployments and found that 95% of AI pilots showed no measurable impact on the bottom line. The reason: only 40% of companies have bought an official AI tool, yet employees at more than 90% of them use personal tools like ChatGPT or Claude for work every day anyway.
The MIT report isn’t really about whether AI works (clearly it does, for that 90%). It’s about whether a win generalises past the person who found it: individual gains everywhere, almost none of it inherited by the team.
You can’t teach that like wiring a plug: read the diagram, follow the steps, done. It’s closer to teaching someone how to bypass a blocked heart valve – you’re in the room, watching the hands, catching the moment it doesn’t go the way the textbook says, with someone who’s done it a hundred times before.
I was teaching my eldest to drive last night. I could not, for the life of me, explain the clutch to him in words. “Ease off slowly, feel it bite” means nothing until the car is stalling under you and someone who can see what’s happening says: “There. Right there. That’s the bite-point.” That’s the actual learning moment. Not before it, not in a manual. Right there, live, with someone watching and correcting in real time.
One caveat: this doesn’t apply to everyone. Some people never needed the passenger seat driver – they just needed to see what’s possible. A demo, a colleague showing off something they built, a headline about what someone else automated – enough of a spark for plenty of people to go figure out the rest themselves.
A smaller number don’t even need the spark – poking at Claude or ChatGPT on a Tuesday evening out of curiosity, they’ve built their own version of the report or the intake notes, unprompted. They’re a slice of that 90%+ shadow-usage number, not all of it – most of that number is scattered, small scale use, not people who’ve solved something, end to end. Your job here isn’t the driving lesson. It’s noticing what they’ve built and making sure it survives if they change jobs.
The sit-down is for everyone else – which, in most organisations, is most people.
So what does work?
Sitting with the Recruiting Ops leader, watching her build the report the old way, once. Then rebuilding it together: connecting the systems, showing the AI what “done” looks like, running it, fixing what’s wrong, running it again. Not a demo – her actual report, her actual data.
That’s stage one of a five-stage pattern I keep seeing work, whether it’s her monthly report, a sales rep’s call follow-ups, or a recruiter’s intake notes:
- Capture the exhaust. Record the calls. Keep the transcripts, the spreadsheets, the half-built reports – and keep an eye on what your self-taught people have already built. Most teams have the raw material for this sitting around unused.
- Find the task, not the tool. Look for the thing that’s high-frequency, high-pain, and still fully manual. Her monthly report. Your churn analysis. Someone’s intake notes.
- Build it with them, live – for the people who need it. This is the clutch moment. Sit down, connect the systems, build it together, watch them hit the snag, correct it in real time.
- Prove it, then package it. Once it works, the job changes. What was a live, one-off session (or someone’s private workaround) gets turned into a template, a saved skill, a scheduled run: something the next person can pick up without redoing the session from scratch.
- Let it become infrastructure. The real finish line is when it stops being “the AI thing we tried” and just becomes how the report gets made. Nobody mentions it in a meeting anymore, because nobody has to.
Scaling this past one desk
How does this go from one person’s report to something that carries across a business with thousands of people?
Not by making the sit-down bigger and calling it a “rollout.” Run it more times, in more rooms – fewer than you’d think, since some of your people already did stages 1-3 on their own, unasked. Find them first, then spend your time on the rest.
MIT NANDA’s data on the ones who don’t self-solve is uncomfortable for large organisations. The biggest companies, with the most budget and the most pilots, convert the fewest into anything that survives – nine months from pilot to production on average, and often they never get there. Mid-market companies, doing less, closer to the work, average 90 days. NANDA’s explanation: the ones who made it kept ownership with the people doing the job, not a central AI team running pilots at a distance.
That’s stages four and five, in practice. Not a bigger pilot, but a repeatable version of the thing that already worked, handed to the next person, function by function.
Why leaders are the lever
None of this spreads on its own. Someone has to be willing to sit next to the Recruiting Ops leader for the two hours it takes, instead of a course. And someone has to notice, first, who doesn’t need that at all.
That’s where leaders come in – the Recruiting Ops leader in this story is one. She runs a team, and she still needed the two hours. Nobody’s exempt from the “aha” moment on the strength of their title; pretending otherwise is exactly how a business ends up with senior people running behind on the one skill everyone assumes they’ve got.
It’s also why she’s now positioned to do for her team what was just done for her: “Come sit with me for twenty minutes,” lands differently from someone who’s sat in the passenger seat. The standing doesn’t come from the org chart.
It also comes with a vantage point: enough to notice her monthly report and a sales VP’s churn report are the same problem in a different label, and to spot the person two desks over who’s already solved theirs.
One leader, doing this properly with two or three people moves an organisation further than a platform licence ever will.
One thing to try this week
Find two people: one already building something with AI, unprompted -find out what it is, and make sure it doesn’t stay theirs alone. And one still doing the monthly-report equivalent of the Recruiting Ops leader’s Power BI ritual by hand – sit down with them, once, and build the first version together.
And if you’re the one running the team: put yourself on that second list too, before you put anyone else on it!
Wondering what’s next for AI in talent?
I’m hosting a packed SocialTalent Live webinar on Automating 40% of Recruiting that’s going to be a must-watch for the future-focused leader.
Register here to make sure you don’t miss out:
https://socialtalent-4.wistia.com/live/events/1ml3he5imc
See you there!