Every executive I talk to asks me some version of the same question. Where do we start with AI.
Wrong question. The one that actually matters is which bucket, and how many of them does this touch at once.
The Four Buckets.
I've come to think about every AI use case in four buckets. Augmentation. Automation. Optimization. Prediction.
Augmentation makes a person better at what they already do. Automation removes the person from a task entirely. Optimization improves something that already runs, a process, a spend, a schedule. Prediction tells you what is coming before it arrives.
Most operators pick one bucket and stop there. That's the mistake. The workflows worth funding first are the ones that land in two buckets or more at the same time. A task that only automates is nice to have. A task that automates and predicts, that's the one that moves the P&L.
"One bucket is a feature. Two buckets is a priority."
Where Speed Pays for Itself.
I've watched marketing teams spend weeks with agencies and studios producing a single campaign, then wait another two weeks to learn whether it worked. That timeline used to be the cost of doing business. It no longer is.
With good data science on what buyers actually respond to, and AI doing the creative production, a team can launch several distinct, tailored ads within hours instead of weeks. That's augmentation and automation stacked on top of each other. The team still owns the strategy. The machine owns the production line.
The Words That Change a Room.
The best salespeople I've worked with have one thing in common. They know exactly what to say the moment a prospect raises a concern. Most reps don't. They freeze, they improvise, and the close rate shows it.
Engineer the rebuttal in advance, arm the team with the exact phrase for the exact objection, and close rates move in a way that coaching alone rarely produces. The same logic runs through customer care. Next best action, next best statement, delivered at the right moment in the conversation, is the difference between a call that ends in frustration and one that ends in loyalty. I've seen NPS move by more than sixty points off exactly this kind of change.
Timing Is the Whole Skill.
A referral asked for at the wrong moment is worse than no ask at all. It reads as desperate, sometimes as entitled. A referral asked for at the right moment, with the right phrase, feels like the natural next step in a conversation that already went well.
AI is exceptional at knowing which moment is which. In a legal services and veteran benefits business I ran, getting that timing right at scale, across tens of thousands of interactions, moved customer acquisition cost down by more than half.
The same principle runs through every automated text and voicemail a customer receives after a missed connection. Most of them sound identical, and customers know it. Tailor the message to the actual reason for the outreach, let AI choose the moment to deliver it, and engagement changes completely. Settlement conversations and payment plans run on the same logic. It was never really about the number. It's about the phrase, the timing, and whether the offer feels built for that person's actual situation. AI does that extraordinarily well, more consistently than any script.
The Data Lake You Don't Need.
Every business I've operated has the same problem underneath it. The truth about the business is scattered. Some of it lives in the CRM. Some in the communication logs. Some in social. Some in the financial books. The traditional answer is to unify everything into one data lake and then start asking questions. That project alone can take a year.
AI can shortcut it. Given access to the disparate sources directly, it can answer the question without waiting for the unification project to finish. That doesn't replace good data architecture. It buys back a year of better decisions while the architecture catches up.
Buy Back Your Own Time.
I hold operators to the same standard I hold myself to. If you're asking your teams to use AI to work smarter, you should be doing the same in your own calendar.
Every meeting I run now has AI handling the agenda, the notes, and the follow up. That isn't a convenience. It's hours a week returned to actual thinking. I've gone further and built agents for parts of my life outside of work too. The goal was never to automate away the job. It's to buy back the time the job was quietly taking from everything else.
"The best use of AI is not the task it finishes. It's the time it gives back."
That's the real test I apply to every AI conversation I have with a board or a leadership team. Not, is this impressive. Which bucket does it land in, does it land in more than one, and what will we do with the time it returns.