Most companies start their AI journey in the wrong room.
They start in a vendor demo. Somebody saw a slick product walkthrough, the dashboard lit up, the chatbot answered in plain English, and suddenly there’s a line item in next quarter’s budget. The shopping started before the thinking did.
I’ve watched this pattern play out for twenty years at IBM, at Kyndryl, and now across the businesses Summit AI works with today. The companies that succeed with AI almost never start by shopping. They start by answering a short list of hard questions about themselves. The vendor conversation comes later, and it’s a much better conversation because of it.
This piece is the prequel to one I wrote earlier “9 Questions to Ask an AI Vendor Before You Sign Anything.” That one was about interrogating the people selling to you. This one is about interrogating yourself first. Because if you can’t answer these seven questions, no vendor on earth can save you.
1. What specific problem are we actually trying to solve?
“We want to use AI” is not a problem statement. It’s a wish.
The real question is narrower and more uncomfortable: what is the specific, measurable thing that is costing us money, time, or customers right now? Slow invoice processing. Inventory that’s always wrong. Support tickets that take three days. Name the bleeding before you go shopping for a bandage.
When I work with a new client, I won’t let the conversation move forward until we have a problem written down in one sentence with a number attached. If you can’t put a number on it, you don’t understand it well enough yet to buy anything that fixes it.
2. Where does our data actually live and what shape is it in?
This is the question that separates the projects that ship from the projects that stall at 90%.
AI runs on data. Not the data you wish you had but the data you actually have, sitting in the systems you actually use. So before you shop, find out where it lives. Is it in one system or twelve? Is it clean or is it a graveyard of duplicate records and blank fields? Can the systems even talk to each other?
I wrote a whole article called “You Don’t Have an AI Problem. You Have a Data Problem.” for a reason. The single most common cause of failed AI projects I see isn’t bad AI. It’s good AI pointed at bad data. Answer this honestly before a vendor answers it for you in month four when the invoice is already paid.
3. Do we have the infrastructure to support this or just to demo it?
There’s a canyon between a pilot and a production system.
A demo runs on a clean sample dataset, on somebody’s laptop, with nobody depending on it. Production runs every day, on live data, with real people and real money on the line. The infrastructure those two things require is not the same, not even close.
At one of my manufacturing clients, we designed a 24-month Azure AI modernization roadmap before a single model went into production, because the existing foundation couldn’t carry the weight of what they wanted to build. That wasn’t pessimism. That was the difference between a project that lasts and a science experiment that gets quietly shut off after the budget runs dry.
Ask yourself: can the thing we’re about to buy actually run here, at scale, on a Tuesday, when it matters?
4. Who owns this after the consultants leave?
Every AI initiative needs an owner inside the building. A real one, with a name and a calendar.
I’ve seen brilliant implementations wither because the moment the outside team rolled off, nobody inside the company knew how to maintain it, retrain it, or even explain it to a new hire. The technology didn’t fail. The ownership did.
Before you shop, decide who that person is. If the honest answer is “nobody has time,” that’s not a reason to skip the question. That’s the most important thing you’ve learned all week.
5. What does success look like in a number we’ll actually track?
Vague goals produce vague results, and vague results never survive a budget review.
“Improve efficiency” is not a target. “Cut invoice processing time from three days to four hours” is. When we built the case at one Kyndryl account, we projected a 25% improvement in incident detection and delivered 38% and we delivered $2M+ in documented savings. None of those numbers would have meant anything if we hadn’t defined, up front, exactly what we were measuring and what counted as winning.
Pick your number before you shop. Then you’ll know whether the thing you bought actually worked instead of arguing about it later.
6. What’s our honest appetite for change?
This is the question nobody wants to ask out loud, and it kills more AI projects than any technical gap.
AI changes how people work. It changes workflows, roles, sometimes whole job descriptions. If your team is exhausted, suspicious, or already drowning in three other initiatives, dropping AI on top of that isn’t transformation, it’s a fourth thing to resent.
I’d rather a client tell me “our people aren’t ready” before we start than discover it during rollout, when the tool is technically perfect and nobody will touch it. Adoption is not a phase that happens after implementation. It’s a condition you assess before you spend a dollar.
7. Why now and what happens if we wait six months?
Not every problem needs AI today. Some need a process fix first. Some need clean data first. Some need to wait until the team has the bandwidth to absorb the change.
Asking “why now” forces honesty about urgency. If the answer is “because a competitor has it” or “because the board keeps asking,” that’s worth knowing because it tells you the pressure is external, not operational, and you should plan accordingly. If the answer is “because this problem is costing us $40,000 a month and growing,” then you’ve got your business case writing itself.
There’s no shame in concluding the timing is wrong. There’s a lot of shame in spending six figures to find that out the hard way.
The conversation that actually matters
Here’s what I’ve learned after two decades of this: the vendor selection process is the easy part. The hard part is the self-knowledge that should come before it.
Answer these seven questions honestly and one of two things happens. Either you walk into vendor conversations with a clear problem, clean expectations, and a real shot at success or you realize you’re not ready yet, and you save yourself a very expensive lesson. Both outcomes are wins.
The companies that get AI right aren’t the ones with the biggest budgets or the flashiest tools. They’re the ones that did this thinking first.
Ready to answer these questions with a clear head?
- Start with an AI Infrastructure Assessment — we’ll map your data, infrastructure, and readiness so you know exactly where you stand before you spend. Schedule a conversation →
- Explore our services — from assessment through implementation and change management. See how we work →
- Read more on AI strategy and infrastructure — practical frameworks, no hype. Visit the blog →
Russell Love is the Founder & CEO of Summit AI Business Solutions, based in Browns Summit, NC. With 20+ years of enterprise transformation experience at IBM and Kyndryl, Russell helps businesses build the foundations that make AI actually work.