Access to information has always been the edge for business. Firms that pull insights from the data first could beat great products, edge out superior talent, and even neutralize competitive advantages. But over the last few years, artificial intelligence has changed this dynamic. With just a few button clicks, anyone can access some form of analysis almost instantly, and at very little cost. Information has become a commodity.
But when analysis is commoditized, the advantage moves away from access and toward interpretation. I like to think of AI as a very smart individual contributor with little business experience. It can certainly draw interesting conclusions, but they may not be relevant to the work at hand. The most important question for business leaders today is less “do we have the right data,” and more “have we built the capabilities to use it?”
As a leader of a global growth operation at Donorbox, I’ve had to contend with this exact dilemma. It’s forced me to develop a framework for using AI that enables my team to get the most out of these powerful tools while keeping our workflows human-centered and rooted in hard-earned experience. There’s no manual for this yet. You’ll have to develop your own.
Pillar no. 1: Use discretion with AI tools
The first rule I emphasize to my team is discretion, which begins before AI produces anything. Every input should be treated as a human judgment call and a potential governance risk.
I recently used an AI tool to analyze Donorbox’s top 200 nonprofit partners. I could have tackled this project in many different ways. The easiest would have been to upload a file with all our customers’ information, which would have required less work on the front end and would have delivered useful results.
However, our customer lists include sensitive information, such as emails, phone numbers and revenue data. I don’t know where that information ends up once it leaves my system. So, before uploading it to a third-party tool, I stripped it of any individual identifiers and kept only generic information, such as organization type and feature usage. The useful signals went in; the sensitive information stayed out.
For our organization, the hard line is contact information, unique identifiers and revenue figures. We wouldn’t give AI access to any of that on any given day. For other organizations, it could be patient data, proprietary company information or even material governed by non-disclosure agreements.
Understanding where this line lives is a learned skill. A junior analyst under deadline pressure will paste the whole spreadsheet in the chat window without realizing they may have just jeopardized the entire organization. Developing this instinct, or at least teaching your team to recognize when they need to start asking questions, must be a learning and development priority in the AI age.
Pillar no. 2: Build a method for working with AI
AI tools don’t know what they don’t know. For these tools to be useful, users must provide the right context and routinely question the tool’s outputs. Unfortunately, AI speaks convincingly even when it shouldn’t. Consequently, inexperienced team members might take an output that is factually incorrect or strategically misaligned at face value, creating unintended problems.
I’ve developed my own method for closing this context gap: I give the model a specific goal, like growing the business by 20 percent, hand it relevant data I’ve carefully screened and ask it for three to five options I can act on. Then, I run those options against what I already know about our business, internal goals and existing team commitments before making any decisions.
Here’s a more personal example: I once used AI to segment our customer list to see if it could find similarities I hadn’t thought of on my own. Its output lumped all Christian organizations into one bucket, which makes sense at a high level. However, my experience told me that a Christian media publication, a church and a ministry all operate very differently, which impacts how we interact with them. I pushed back on the results by explaining my reasoning and asking for another iteration.
The lesson here is to never accept the first output as the gospel truth. It often takes five or six rounds of prompting to get something genuinely useful. As we train our teams, we must reinforce that the people who will get the most out of AI aren’t the best prompters. Instead, they’re the ones who can provide the right context, recognize flawed outputs and ask for something better by providing specific feedback.
This is a workflow every team should begin creating today.
Pillar no. 3: Adopt a mindset of continuous learning
Organizations have traditionally been limited by their internal capabilities, and they set goals and made plans accordingly. If a company wanted to take on a new project or expand into new areas, it needed to either train existing employees or hire new ones with the skills it needed. But that’s changed with AI. An organization’s capabilities are no longer static, and the way it plans shouldn’t be either.
Every new AI release has the potential to transform what an organization can accomplish overnight. I could start building a Donorbox subproduct today and have something finished in a week or two. Before AI, that same project might have taken my entire team two quarters to finish.
As tools and capabilities keep evolving, we need to change how we deploy our teams and plan for the future. We must adopt a posture of continuous learning just to stay current. Otherwise, our people could be working toward a version of the company that will be made obsolete with the next model release.
One effective way to implement this within an organization is to reduce your operational timelines. Instead of planning annually, my team now sets monthly goals and runs two-week sprints. This forces us to regularly reassess what we’ve accomplished and what’s possible and adjust our longer-term plans accordingly.
In the old days, continuous upskilling was the dream only the best organizations could realize. Today, it has to be the standard so employees can get the most out of a new class of constantly evolving tools.
Pillar no. 4: Make mistakes and learn from them
For all the capabilities AI offers, it cannot give users judgment. That is the residue of previous mistakes and only comes through experience.
If I asked AI to create a social strategy for my organization, it might tell me to build an audience on TikTok and create content there. That advice would be right for many brands. However, I know my ideal customer is on LinkedIn, and I haven’t seen any of my competitors use TikTok successfully. The model I’m working with can’t know that information. That’s my own experience talking.
Unfortunately, the wrong AI answers are often the most exciting and the ones the models deliver most confidently. The judgment to disregard a bad suggestion comes only from having been wrong before and learning from those mistakes.
If judgment is a byproduct of experience, then organizations can’t afford to insulate their people from making decisions. An AI that does all the deciding prevents people from building the experience to override it when necessary. To escape this trap, you must allow your people to make calls, be wrong and learn from those mistakes.
The L&D leader’s new responsibility
Information is no longer a competitive advantage. AI now makes analysis free and available to everyone. But it can’t deliver discretion, disciplined workflows, adaptability or judgment. These are all human capabilities that must be developed.
Within this framework, competing in the AI era becomes less about developing technology and more about developing humans. Learning will be the new edge. The companies that succeed over the next decade won’t be the ones that turn their decision-making over to machines. Instead, they’ll be the ones that teach their employees how to think alongside them.
