Most companies use AI to do the same things they’ve always done, just a little bit faster. Summarizing meeting notes in half the time. Drafting emails in seconds instead of minutes. Generating first versions of reports (that still need extensive editing).
All of that is useful, but incremental. It takes the existing workflow and makes it marginally more efficient. If that’s where your AI strategy stops, you’re missing a bigger opportunity.
At Luxury Presence, we’ve spent the last year experimenting with AI in ways that go beyond speed. We don’t just ask how AI can help us do our jobs faster, but how AI can help us do our jobs differently—and in some cases, do things that weren’t possible before.
How we built a focus group with AI
Inside our company, we recently built synthetic customers to serve as our on-demand focus group. Because traditionally, product development was always constrained by insufficient time with customers to validate all your ideas.
You might have 20 concepts you want to test: Three different versions of a landing page, five variations of a feature, or ten ways to position your product. But you can only realistically get a handful of customers on calls each week. Most of your ideas go untested, and you’re making decisions based on gut instinct rather than real feedback.
We trained an AI on more than 50,000 Gong call transcripts from our platform to analyze conversations and identify the six most common customer personas.
Not generic archetypes, but detailed profiles based on actual behavior and language patterns. We then trained six separate AI agents, each modeling a specific persona. These agents could respond to questions, react to product demos, and provide feedback similar to real-world counterparts.
Now, when we want to test a new feature, we don’t wait two weeks to schedule customer calls. We run it by our synthetic customers first, and they provide feedback within minutes. We can show them wireframes, marketing copy, pricing structures, or new branding concepts, and they’ll tell us what works and what doesn’t.
Of course, this doesn’t replace real customer validation entirely, but it gets us about 90% of the way there. We use our limited time with actual customers to validate final details, not to determine whether the entire concept is worth pursuing.
Our next step is even more ambitious. We plan to give these synthetic customers access to real prototypes and ask where they get frustrated. We’ll have synthetic user data from the actual product.
Why most teams don’t think this way
When I discuss our approach with other founders, the most common reaction is surprise. Not because the technology is complicated—it’s not—but because solving problems this way requires a mindset shift.
For the last 20 years, I’ve been told that focus is the most important quality for a startup founder. Don’t try to do too many things at once. Pick one thing and do it well.
That advice made sense in a world where capacity was limited by how many people you could hire and how many hours they could work. AI changes that equation.
Now, our best engineers spin up multiple AI agents simultaneously. They prompt the first one and send it off to work for a few minutes. While that runs, they start the next agent on a different problem. They’re managing eight tasks in parallel, like a chess grandmaster playing 12 games at once.
That’s not how leaders were trained. We’ve been advised to avoid context switching, protect our focus, and do deep work on one thing at a time. But the people getting the most out of AI are those willing to rethink those assumptions.
How to think differently
To move beyond incremental improvements, start by identifying your biggest bottlenecks. Not necessarily the tasks that take the most time, but constraints that hold you back from doing better work. Then ask yourself: Is this a high-stakes task or a low-stakes task?
High-stakes tasks are those where the cost of getting something wrong is very high. You’ll likely want a human-in-the-loop for those. Focus on building AI tools that make humans faster and more effective, not on replacing them entirely.
Low-stakes tasks involve a minimal cost for getting something wrong. You can afford to let AI take the first pass, even if the output isn’t perfect. If one out of ten outputs is inaccurate, you’re still getting valuable insights because you’re generating so much volume.
Customer feedback is a perfect example of a low-stakes task. If a synthetic customer gives you feedback that’s slightly off, it’s okay. With hundreds of these interactions, emerging patterns remain useful. And when the stakes arehigher, you can always prioritize feedback from real-world customers.
Once you’ve identified where you can experiment, use the tools and build something.
What this unlocks
When you stop thinking of AI as a way to do your current job faster and as a way to do your job differently, you open up possibilities that didn’t exist before. That’s the difference between incremental improvement and transformation. The opportunity exists for anyone willing to rethink the work instead of just speeding it up.
Malte Kramer is the founder and CEO of Luxury Presence.