ChatGPT is not a strategy

How your business should be thinking about Generative AI right now

Dec 13, 2023

I’ve had the same conversation with many leaders in the twelve months since OpenAI’s ChatGPT first burst into public view. It starts with their request to discuss their “AI strategy” or their “ChatGPT strategy.”

Let me be clear --

ChatGPT is not a strategy.

Metaverse is not a strategy.

Web3 is not a strategy.

Blockchain is not a strategy.

Each of these are technologies that you may use in order to pursue specific opportunities for your business to create value in the market. But technologies are not, and should never be, the starting point of strategy.

Instead, we need to take to heart the mantra of startups to “fall in love with the problem and not the solution.”

Before you launch your next ChatGPT pilot project, ask yourself: what problem are we trying to solve?

The Distraction of Technology

The last 12 months have been, admittedly, heady and exciting.

With Generative AI, we are seeing the emergence of an incredibly interesting new technology that will likely reshape our lives and the underlying dynamics of many industries over the next decade or two.

But…

With all its amazing capabilities, GenAI has one obvious downside. Like any emerging technology, it can easily become a distraction, drawing our attention away from questions of strategy.

To avoid this, we must constantly force ourselves to shift our attention away from technology’s shiny new objects and toward a focus on business needs and customer needs.

If you want your people to get past this year’s “yowza!” demonstrations of AI (“ChatGPT turned my email into a PowerPoint… then it summarized my PowerPoint as an email!”), you will need to link Generative AI to strategy.

Taking a New Look at GenAI

So, what should your business be doing with Generative AI in 2024? After all, we are still at a very early stage of this remarkable technology.

I will assume that you are not in the business of building foundation models (LLM’s like ChatGPT) or the chips they run on (in which case, you’re too busy printing money to read this article).

What should you be doing with this new technology?

I recommend to NOT expect a large impact on your business at this stage. Instead, focus on low-risk experiments that are designed for learning, and focus on sharing what you learn across the organization.

This is in sharp contrast to what I see happening in most companies:

A slew of pilot projects are launched, with everyone competing to look like they’re “doing something” with the world’s coolest new technology. In each pilot, the team never clearly defines the problem to solve or what success vs. failure would look like. And there is little to no shared learning.

5 Steps for Every GenAI Experiment

So, how do we do better?

How do you design experiments to minimize risk, maximize learning, and accelerate your path towards getting actual value (repeatable, measurable, scalable) out of GenAI?

I recommend a simple five-step process for every team:

1. Define a problem to solve.

Pick an important problem you think you might solve, or a need you might satisfy, with GenAI tools. (Note: “We could do X!” is not a problem. However, “We could improve how we currently do Z” potentially is.)

2. Find your customer.

Who would use or benefit from your GenAI solution? Who would want your problem solved? Is it an external customer? Or someone inside your organization? This will be the “customer” for your solution.

3. Validate a definition of success.

Go find that customer and meet with them. Solicit their point of view. With their help, you need to validate 3 things: the problem you’re solving (do they see it the same way?), a measurable definition of success (time savings? cost savings? higher quality output?), and guard rails (what metrics would indicate your GenAI solution is not worth using–e.g., measures of reliability or accuracy?).

4. Experiment to see what works.

Drawing on AI tools and supporting experts, try out new solutions to the problem. Have the customer judge the results based on your agreed metrics. Capture qualitative responses too—which may point to a benefit or problem you had not foreseen.

5. Share what you learn.

Share this learning widely, throughout the organization: What did you try? What worked? What didn’t? What else did you learn?

…The above five steps are what I’m seeing in a range of organizations (retail, media, education, financial services) taking a focused, learning-based approach to GenAI.

This post is public so feel free to share it.

Design Your Own GenAI Program

Of course, this doesn’t magically happen on its own. Effective experimentation with GenAI requires that you design your program before you start.

This will entail innovation governance, training and support for your experimenters, reporting mechanisms for shared learning, and a very small amount of money.

In practice, these pieces should include:

Simple training—What tools are out there, and how do they work, at a basic level. Also, basic rules of safety (as one executive put it to me, “Let’s not do anything boneheaded”).
Project greenlighting—To get approval, any pilot must, at minimum, be able to define a problem it aims to solve.
Security oversight—IT must sign off that, yes, you can use this tool with this data in this part of our intranet.
Small budget—For licensing of external tools.
Mandatory reporting—Every participant must report: What did you do with your time and resources? What did you learn?
Distributed learning—Some mechanism (demo days, newsletter, knowledge base) for internal sharing of findings.
Growth board—A group of business and IT leaders that meets every 60-90 days to review the most promising experiments and consider: Is this solution safe to scale further? Is there enough business value to merit the investment?
OPTIONAL: Higher level technical support—e.g., roving “prompt engineers” who can move from project to project (if you can afford it)

Once this framework is in place, all employees should be encouraged to participate in the program. But they must follow the process you’ve thoughtfully designed.

Any business should start this process with MODEST expectations:

Most ideas submitted will not be able to articulate an underlying problem they would solve. These should be rejected at the green-light stage, or shuttered immediately.

Most green-lit experiments will fail to solve their problem or do so with an acceptable error rate.

A few experiments may show results promising enough to validate and refine them further, and to consider scaling across the business.

But the broadest impact will be to dramatically accelerate meaningful learning inside the organization, so you are ready, as GenAI advances rapidly, to unlock real value for your business.

A Goal for 2024

In the next one, two, or three years, GenAI could start to radically change what happens every day inside your organization. It may even upend your current business model. Or help you launch a totally new business model into the market.

But if you’re not building AI chips or foundation models yourself, 2024 should be a year focused on experimentation and learning.

And real learning will only happen if you start, not with the technology, but from specific problems you want to solve for your business or your customers.