We’ve run dozens of AI experiments at DOXA Talent® over the past two years. Many of them didn’t produce repeatable results and we’ve since moved on. That might sound like a bad track record, but I’d argue it’s the whole point.
The conversation I keep having with CEOs about AI usually starts in the same place: they want to move, but they’re afraid of getting it wrong. I think most leaders are framing the decision incorrectly. They see a choice between speed and safety, when the real risk is standing still. If you delay experimenting, you delay learning, and learning is what creates competitive advantage. You move fast by running small, contained experiments that teach you something useful, even when that something is “this didn’t work.”
Give Experimentation Its Own Space
We created DOXA Labs® because it was nearly impossible to experiment inside normal operations. When teams are focused on delivery, client work, and hitting quarterly targets, experimentation gets deprioritized. It usually disappears when capacity gets tight, which is exactly when you need it most.
DOXA Labs® gives us a structured place to explore questions that matter to the business. Can AI support our recruiting teams? Can we accelerate onboarding for remote workers? How do we help employees become more productive without burning them out? Instead of debating these ideas endlessly, the Labs environment lets us test them quickly. Some experiments fail. Some lead to small improvements. Occasionally we discover something that changes how we operate entirely. The real value of a dedicated space is the learning muscle your organization builds by making experimentation a permanent part of how you operate.
Scope, Data, and Time
We think about experiments in three layers: scope, data, and time.
Scope comes first. We start with very narrow use cases like drafting internal documentation, summarizing meetings, or assisting with research, which keeps the impact contained while still producing something you can evaluate.
Data is the second layer, and this one matters more than people realize. The instinct is to avoid sensitive data altogether, but many of the most valuable AI use cases require working with real data early. The principle should be controlled usage, not avoidance. That means operating in secure environments, minimizing data exposure, enforcing access controls and auditability, and being thoughtful about model selection – knowing the difference between closed systems that keep your data contained and open ones that may not. PII handling, vendor risk, and broader data governance should be part of the conversation before an experiment launches, not after something goes wrong.
The third layer is time. Every experiment has a clear window, usually a few weeks, and at the end we ask three questions: What worked? What didn’t? What did we learn?
Beyond structure, teams need guardrails that let them explore without second-guessing every move. At DOXA Talent®, we keep those guardrails practical. Data governance standards are set before experiments begin, not bolted on afterward. AI-generated outputs always get reviewed by a person before being used in any decision. And every experiment produces documented learning that gets shared with the broader team. When people know the boundaries are clear and the environment is secure, they feel safe enough to push into unfamiliar territory.
Not every idea is worth testing, and not every test is worth scaling. So DOXA evaluates any experiment through a simple lens. We prioritize based on the potential impact, ease of implementation, and how quickly we can learn something useful. The goal isn’t just to explore. It’s to identify what can move the business forward in a measurable way. Most experiments won’t create any meaningful value, but a few will, and those are the ones that compound over time.
Protect the Time
If experimentation depends on spare time, it will never happen. We treat it like any other operational priority. Teams dedicate small windows, sometimes just an hour or two, to test a new tool or workflow. The key isn’t the size of the experiment. The key is consistency.
A company that runs 50 small experiments over a year will learn far more than one that runs a single large AI initiative every two years. Progress compounds when learning becomes routine, and the only way to make it routine is to protect the time for it.
One Experiment
One recent experiment focused on candidate research during recruiting. Our team tested whether AI could summarize candidate profiles, highlight relevant experience, and prepare interview briefs for hiring managers.
AI didn’t replace the recruiter’s judgment, but it significantly reduced the time required to prepare for interviews. Work that used to take 20 minutes per candidate – reviewing resumes, pulling together notes, organizing details for the hiring manager – was compressed into a fraction of that, freeing recruiters to focus on higher-value evaluation and relationship-building. The lesson was that AI amplifies expertise by removing repetitive preparation work, and that’s a pattern we’ve seen repeated across multiple experiments since.
Start Now
The companies that figure this out early will develop instincts their competitors won’t have: the ability to identify the right use cases, ask the right questions, and course correct quickly when something isn’t working. That instinct comes from the accumulated experience of dozens of small experiments where your teams learned what works, what fails, and what to try next.
At DOXA Talent®, we manage teams across six countries, and one thing I’ve learned is that the organizations with the strongest experimentation habits also tend to adapt fastest to everything else. The muscle you build by testing AI in a structured way carries over into how you onboard people, how you enter new markets, how you make decisions under uncertainty. It becomes part of how your company thinks.
I won’t pretend there’s a clean playbook for any of this. We’re all writing it as we go. But the companies that start writing it now, even with imperfect drafts, will be miles ahead of the ones still waiting for someone else to figure it out first.
David Nilssen is the CEO of DOXA Talent® which helps businesses build and scale-up high-performing, borderless teams leveraging talent from across the world. He has over 800 team members and ZERO office space. He is also the Co-founder of Guidant Financial which has helped 30,000 entrepreneurs to secure $7B to start or buy a business in each of the 50 states. David serves on the Board of Directors for the AI Officer Institute, helping leaders evolve how they lead in the era of AI.
