A working session for Republic Financial executives — frameworks, not jargon — built around 25 questions I bring into every executive workshop, and two live engagements pulled from Republic's own portfolio.
My job today is not to convince you of anything. It's to give you three ways of thinking that will sharpen every AI conversation that hits your desk this year.
Most of what gets called "AI strategy" is really just figuring out how your business actually works. The decisions that matter — AI, automation, or neither — only get good once the questions in your senior people's heads are out, written down, and answered. That's the work.
These are 25 questions I bring to every executive workshop — they're not Republic's, they're mine. We won't address all 25 today; we'll hit roughly eight, organically, through two real cases. At the end I'll come back to whichever the room ranked highest. This is your workshop to direct.
How do I tell the difference between AI work that looks impressive in a demo and AI work that actually saves us time, money, or errors?
Which of the things we're good at today get weaker as AI tools become widely available, and which actually get stronger?
How do we run a coordinated company-wide approach instead of a pile of disconnected experiments nobody can tie back to a goal?
Is it the company's responsibility to teach employees how to use AI tools, or is it on each employee to figure out on their own time?
How do we redesign work from the ground up around AI, instead of adding AI on top of how we already do things and calling it transformation?
How do we explain what's changing to employees in a way that's honest about the impact without setting off panic or sounding like a press release?
"AI transformation" gets thrown around. What does it actually look like in our business? What are the concrete options on the table?
If something that used to cost us $100 now costs $1, what newly becomes worth doing? Where does that change what we offer or how we operate?
What's our checklist for green-lighting an AI project? What passes, what doesn't, and who gets to decide?
How do we let employees try AI tools without exposing client data, IP, or putting compliance at risk? Where are the safe sandboxes?
How do the people doing the actual work tell us where AI could help, and how do we decide which of their ideas to pursue?
How do executives get hands-on enough with AI tools to credibly lead the rest of the company through this — instead of delegating it?
How do we get our data progressively ready for AI without freezing all AI work for two years waiting for a perfect data project?
Of everything we could do, where do we start in the next 30, 60, 90 days, and why that work and not something else?
How do we build AI tools so we can swap models or vendors when the market shifts, instead of being stuck rebuilding from scratch?
How do we test for AI curiosity, comfort, and basic literacy in interviews — including for non-engineering roles?
How do we turn IT, legal, and compliance into partners who help us move faster, instead of the people whose default is to say no?
When we say AI will free people up to do higher-value work, how true is that really? What if higher-value work isn't actually waiting for them?
How do we shift culture when most employees are either afraid AI will replace them or just disengaged from learning it at all?
Should one central team own the AI strategy, or should it live in every function? What are the trade-offs we're choosing between?
What are the real risks of investing aggressively in AI right now, and what are the real risks of waiting? Both lists need to be on the table.
How do we find the people inside our company already using AI well, and turn what they do into a standard others can learn from?
Which companies have used AI well, which have failed publicly, and what specifically can we apply or avoid from each?
If a brand-new competitor launched tomorrow and built everything around AI from day one, what would they do differently — and why aren't we?
What happens to our cost structure if the AI vendors we depend on raise prices significantly? How do we hedge that exposure?
A predictable rhythm: each mental model gets stated, shown in both cases, then handed to you to apply on a real open question.
Pick three from my 25-question card. We'll come back to them.
An aircraft leasing deal model and Republic's RFC Drilling AP report.
Understand the work first. Fix what goes in. Match the rules to the task.
The real questions I'm asking my clients this week. How would you decide?
Closing thought experiment, then back to whatever the room ranked highest.
Both come from Republic Financial work — specialty aircraft finance on one side, RFC Drilling AP on the other. A spreadsheet that reformats data from a source system, layered with judgment, maintained by an unsung operator who has been doing this for years. This is the universal corporate document — and where most "AI strategy" conversations actually live.
A Boeing 767 cargo plane on a six-year lease to a freight airline. The seller hands over their version of a spec sheet — an Excel file with 49 photos of contract pages pasted into it. The analyst spends hours retyping numbers into Republic's deal model.
A drilling company runs Microsoft Dynamics GP. Operator (Dana) builds a payment-plan summary every two weeks: pivot, categorize, layer in actual and projected payments, ship to leadership.
Your senior people have decades of judgment they've never written down. Before I recommend any tool — AI, software, anything — I'm pulling that out: into questions, then answers, then documented steps. Most of my first month on a project is email.
Sounds simple. Bob's reply was partial. If yes, a simple script can match them. If no, we need a more complex tool that can handle near-matches — different cost, different risk, different approval process.
one yes-or-no changes the whole projectGP has a Vendor Class field on every record. Every single row in the export reads "DEFAULT." The real categories live in Dana's head and in a side spreadsheet she keeps updated by hand.
the field exists. nobody is using it.Bob: "I think that happens manually." Even Bob doesn't fully know. Dana's process has never been written down because nobody asked. That's not a Republic problem — that's every company.
knowledge that lives in one person's headThe answer tells us whether projecting payments is a clear rule we can code, a guess we can train software to make, or human judgment that should stay with a person.
rule, guess, or judgment?Each one filters out a category of bad AI projects before they get funded. Each one shows up in both Republic cases. Each one changes the question you ask before you pick a tool.
The same job often has parts that need AI, parts that need a simple script, and parts that should stay with a person. You can't tell which is which until the work is documented and the questions are answered.
questions before toolsWhen the source is yours — your CRM, your accounting system — fix it there. When it isn't — sellers, vendors, regulators — build a clean intake on your side so the mess gets caught at the door instead of spreading through everything downstream.
your side, their sideSome tasks have one right answer; others involve judgment. They need different review, different approvals, different oversight. Don't apply one risk policy to everything called "AI."
match the rules to the taskWhen AI gets pitched, notice the assumption that the whole job needs the same answer. It doesn't. Some pieces benefit from AI, some don't. You only know which is which after you've understood the work.
The two cases sit on opposite sides of this. RFC Drilling is Republic's own system — you can fix it. Aircraft sellers will hand over whatever they have, and you can't change that. The mental move is different in each, and getting that right matters.
Matching vendor names across reports is the cleanest example we have. The same business need leads to a simple script or to AI software, depending on the answer to one question — and they need very different oversight.
For each, I'll walk through how the three lenses guide the decision. Then I'll ask the room how you'd decide. None of these have a "right" answer — they have defensible answers, and walking through the trade-offs out loud is the work.
The engine maker publishes a price list every year. Three real options: keep using the prices in our template, build and maintain our own list, or buy a data feed. The "obviously automate it" answer might actually be the wrong one.
It looks productive. It would add zero analytical value, and create real maintenance work because formulas reference sheet names. Sometimes the right answer to "should we automate this?" is just no.
Notice these aren't AI questions or automation questions. They're work questions about how the job actually gets done. Get them answered and the tool choice becomes obvious. Skip them and you'll automate the wrong thing.
If yes, a simple script can match them. Compliance signs off in ten minutes. If no, we need software that handles near-matches, with confidence scores and human review — and that's a much bigger procurement and oversight conversation.
The field exists. Every record reads "DEFAULT." Two paths: automate Dana's side spreadsheet, or use the field GP already has so the category lives on the vendor record forever. The second one is the answer that lasts.
Once you see this pattern, you'll see it everywhere in Republic. Probably in every business meeting you have for the next month. This is the most useful thing you'll take from today.
| What goes wrong | Aircraft leasing | RFC Drilling AP |
|---|---|---|
| How the work gets done isn't written down | How the analyst figures out how much money changes hands at lease end — only in his head. | How Dana decides which vendors get paid this Friday — never documented. |
| Data the system already has, retyped by hand | Monthly maintenance fees that already live as clean data in Salesforce — retyped into Excel. | GP's Vendor Class field set to "DEFAULT" on every record; the real categories sit in a side spreadsheet. |
| Fix what you control; translate what you don't | Sellers will keep sending photo-laden Excels. Build a one-time intake on Republic's side that catches the mess at the door. | RFC's GP system is ours — fix the categorization at the source, in the field GP already has. |
| Judgment sitting on top of clean data | Pricing the deal, estimating resale value, structuring the debt, settling at lease end. | Which vendors get paid in full, partial, or deferred to next cycle. |
| Setup work paid every cycle | About 3 hours per deal of typing before any real analysis happens. | Several hours every two weeks, forever, just to produce the report. |
| Errors and outdated parts nobody fixes | Broken cell references in the valuation section. A label on the model that names a completely different engine — the consequence of copying the file from another deal years ago and never cleaning up. | Broken cell references in the totals row of the summary tab. |
Not because they're better at AI. Because they understood the work first, then picked the right tool for each piece — a parser at the door for the seller's mess, a clean field in GP for vendor categories, a person on purpose for the judgment calls. The tool followed the work.
The question isn't whether they exist yet. It's what's stopping you from being them.
They'll be the ones who got the questions out of their senior people's heads, wrote them down, and picked the right tool for each piece — sometimes AI, sometimes a simple script, sometimes leaving a person in charge on purpose. That's the work.
Show of hands — what did you circle on the card? I'll spend our remaining time on whichever question the room is loudest on, then we'll wrap.
