I've had variations of this conversation many times—so many that I figured it would be easier to write it down and point people here rather than explain it again and again.
We’ve all read about how GenAI is going to change the world. But the question I get most often is: How do we actually take advantage of it in a large corporate setting?
I have a two-step plan to get you to self-sufficiency. Here it is:
Step 1: Put a general-purpose "Chat" model in front of every employee—and monitor usage. Employees who use GenAI will be more productive than those who don’t. Encourage non-users to engage. If they refuse, consider what’s involved in replacing them with someone who will.
Step 2: Get your tech team to build a basic internal Question Answering (QA) service. Start with something like internal IT support or HR queries.
Once your company isn’t afraid of the technology—and crucially, has some appreciation for how statistical models work—you can start brainstorming your own “killer app.”
That’s it. Just two steps. Don’t skip them. Complete both, then bring together the users from Step 1 and the dev team from Step 2 to evaluate everything else you think you should do. It’ll save you a lot of time and money. I promise.
Starting at the Beginning
If you're in management and don’t have a clear plan to roll out GenAI, don’t worry—you’re not alone. GenAI is revolutionary, and like any revolutionary tool, it requires a fundamental shift in how we think.
When I talk about GenAI’s likely near-term impact at corporations, I use the analogy of the spreadsheet. Think back to the pre-Excel days—could you have predicted how profoundly it would change work? Probably not. That’s because you don’t know the intricacies of everyone’s job.
Spreadsheets were revolutionary. They made users significantly more productive.
Two key takeaways from this analogy:
- Spreadsheets help users get to an answer. You don’t deploy a spreadsheet to a server in the dungeon and let it run. A person uses it. The same applies to GenAI. It helps a user get to an answer more quickly. And if the answer is wrong (which will happen from time to time), it’s the user’s responsibility—just as it would be if there were a bug in a spreadsheet.
- Spreadsheets didn’t reduce costs or headcount or create new revenue lines—they enabled users to do a lot more. With GenAI, expect productivity, headcount, and revenue to increase. If your mindset is “How many people can this replace?” or “What new revenue will this create?”—you’re on the wrong path.
To torture the analogy further: if you were hiring a graduate for a financial modelling role and they said they’d love the job but wouldn’t use Excel, would you hire them? Probably not.
If someone needs to write documents, presentations, or code, and they refuse to use a tool that would make them significantly more efficient—you should consider replacing them with someone who will. The productivity gains are immediate.
If you're in the knowledge or service sector, roll out a GenAI tool to all employees. Monitor usage. You’ll quickly notice the difference in productivity between users and non-users. How you manage the non-users (and the errors!) is between you and HR—but you already know what the answer is.
Don't let people keep using log tables and slide rules when there's Excel.
Also: don’t be overly prescriptive. If enough people are using the tool, novel and valuable uses will emerge.
As management, you should:
- Monitor usage
- Encourage sharing of experience
- Support non-users as they onboard
There aren’t clear KRAs —but hey, nobody said management came with clarity.
Step 2: Build Internal Competence
This step is for your technical team—but note: the end goal is for them to collaborate with business users to identify killer apps. I recommend multiple teams undertake this step to build experience across the organization.
In my experience running technical teams, I’ve found that regardless of background, there are “ah-ha” moments everyone needs to feel. They need to internalize these—not just read about them. In GenAi, there are two "ah-ha" moments:
Principle 1: Statistical models cannot always be correct. Make it a requirement for every team member to solve the MNIST digit recognition problem (https://www.kaggle.com/competitions/digit-recognizer) from scratch. That means writing out the vectors and matrices, deriving the gradient descent update, and coding it manually.
Yes, really. This isn’t overkill. It’s first-year university math. If someone can’t do this, they should be a user, not a builder.
At the end of this, they’ll understand why a statistical model cannot always be accurate. Sometimes, you genuinely can’t tell if a rushed “1” is a “7,” or whether a smudged digit is a “6” or an “8.” Life is messy. Statistical models help us navigate the mess—but they sometimes make mistakes. Make sure you really understand what that means.
Principle 2: How do you know if a change is an improvement? After building a basic neural network, the next project should be practical. I suggest using an existing internal FAQ with answers to build a simple question-answering service.
The goal: when someone asks a question, the service answers it correctly or fails gracefully.
This is not about the front end. It doesn’t even have to be deployed.
The real purpose is to expose your team to Retrieval-Augmented Generation (RAG) and get them thinking deeply about evaluation. The project is “done” when they can confidently answer this question:
If a third-party tool or library that you used changes, will the change improve your service?
Until they can explain this to a non-technical stakeholder, don’t proceed.
Bringing It All Together
Now you have:
- Users who understand how GenAI helps them in their daily work
- Technical teams who understand what it takes to build and evaluate models
You’re ready to brainstorm truly valuable internal applications—your killer apps.
Until then:
- Don’t spend money on commercial tools
- Don’t hire consultants
- Don’t put GenAI into production
- Don’t promise headcount reduction
- Don’t promise revenue growth
- And don’t believe what vendors are telling you
Get tools into users’ hands. Get your tech team solving real problems. Then pair them up and define your GenAI strategy together.
It might not sound revolutionary—but trust me, it is.
Just do it. Today.