Implementing AI in Your Business — How to Turn Aspirations into Reality
By now, it’s safe to assume that everyone here has heard of generative AI. Since you’re reading this on Medium, your newsfeed is probably already inundated with the latest ChatGPT news and opinion. For me the most interesting thing about the explosion of AI innovations is how quickly it’s dominated the cultural conversation. AI is really having a moment here in 2023. Whether it’s people using AI art generators to create fun Instagram profile pics or CEOs answering questions about AI strategy in the board room, everyone is talking AI. However, lost in the chatter and excitement of new AI innovations is the less sexy conversation, how do you actually implement AI in your business at scale? As a product manager, I’ve worked with AI for the past 5 years and have dealt with the excitement and the challenge of incubating new AI models and incorporating them into products. Instead of discussing the merits and risks of AI itself, I thought I’d take a moment to step back and think about how successfully implement AI in your business. Ultimately, it doesn’t matter how good your AI is if you stumble in bringing it to your users. As you look to incorporate more AI into your day-to-day operations, here are my thoughts on how make sure you see the results you want.
Focus on Outcomes
How you define the success of an AI model can have huge implications for your business. Often, companies focus on accuracy metrics, such as false positive rate, recall, and precision. They then measure success as the difference between the AI prediction and the actual result. While model accuracy is important, I would encourage business leaders to consider customer or user outcomes as at least equally as important. An AI model can be degrees off from the source of truth, but still deliver the right consumer result, so while your AI might not be “accurate” in the strict sense, you have a positive experience. Thinking about user outcomes widens your definition of success and will free you to implement AI In production faster. Don’t get bogged down trying to get the model from 98% precision to 99% precision if the user sees a benefit either way. None of this is to say that accuracy isn’t important, we should constantly be researching and refining AI tools, but don’t let an arbitrary accuracy metric prevent you from bringing real-world benefits to your customers. This is especially true as models get more accurate, as incremental improvements get more expensive and time consuming. Ultimately, your customer doesn’t really care how accurate your AI is, they care about the experience it delivers them. By measuring the right metrics and making production decisions based on user outcomes, you’ll see faster implementation of AI and faster realization of value.
Stay on Target
It’s hard not to get excited about the potential for AI to transform your business. Advancements are happening rapidly, and the potential use cases are expanding every day. While thinking big is exciting, the more a business tries to take on at once, the less likely anything is to be successful. Leaders usually starts with a simple problem, such as using an AI chatbot to reduce customer waiting times. As people start developing the implementation plan, they tend to start brainstorming and ideating more problems they can solve. All of sudden you have 30 ideas for how AI can transform your business over the next 24 months and what was a single implementation becomes a multi-project, multi-year roadmap. The project soon gets consumed with requirements and dependencies and 24 months later you still don’t have your AI chatbot.
Instead of trying boil the ocean with AI, force rank your initiatives, and I mean really force rank them, 1 -n. No bucketing 5 items into the top category, there can only be 1 most important project. Look at the potential benefit, the cost to implement, the complexity, and the time to production. I would encourage companies to start with low cost, low complexity, and / or short time projects first, even if they don’t have the biggest payoff. This is because delivering a tangible, measurable result is a huge win. It makes life measurably better for the users and customers and is a morale boost for the AI implementation team. That success will breed future, bigger successes. You’ll also learn from that project what to do and not do next time, increasing the likelihood that your bigger project will be a success. If you start with a huge, multi-dimensional change with no experience to fall back on, you’re likely headed for an expensive disappointment. Nothing drains morale (and money) more than a long project that never quite lives up to expectations. Conversely, don’t underestimate the boost you can get by actually delivering something, however small. Delivering smaller projects has the added benefit of allowing you to change gears and implement newer technologies faster, since your team isn’t consumed with a large, multi-year implementation. Instead of treating AI like a non-stop road trip from LA to New York, plan lots of stops along the way and enjoy every one.
Put the “i” in innovation
I once had a consulting partner who talked about the difference between “little i” innovation and “Big I” Innovation. “Big I” Innovation is the stuff of legend, the technology that changes the world. The best example is the iPhone, no one would argue that the world is completely different pre-iPhone and post-iPhone. “Big I” Innovation is great, it moves the world forward, but it’s really, really hard to do. Even the best companies might only have 1–2 “Big I” Innovations in a decade, most will never even have one. Worse, when you swing for the fences, you might find you strike out. While “Big I” Innovation might get the headlines, the awards, and the VC dollars, companies shouldn’t forget about “little i” innovation. These are the small incremental changes that make life a couple percentage points better at a time. I love “little i” innovation because you can see tangible improvement over time. You can make small changes and slowly but surely make the world a better place. Repeat over a long enough timeframe and you might find you’ve completely changed the game without one big bang Innovation. To carry the baseball metaphor through, if you hit a lot of singles, it may not be as sexy as a grand slam, but you’re going to score runs.
Start by identifying the future state, what you want the product or service to look like. Compare that against the current state and figure out what you need to build (AI and otherwise) to get there. Map out an incremental approach, solving one problem at a time, and making each improvement adds tangible value on its own. Once you start delivering, every small innovation will make your product a little bit better. You might even find that your future state changes along the way, which is great, because nothing is worse than spending months on a project only to find out you had the wrong idea in the first place. Most importantly, you’ll realize the benefits of AI faster and over time, instead of slower and all at once. “Big I” innovation is a lot of fun and should always be a part of any innovation strategy, but placing lots of small bets and winning can be just as transformative for your business and may actually get you where you want to go faster.
Conclusion
Now may be the most exciting time ever to be a part of the AI landscape. The fact that anyone can download a generative AI application and use it is pretty awesome, if a little frightening. But the future of AI for business requires a thoughtful and determined approach to solving problems. The AI itself is rarely the barrier to success, rather the implementation strategy and process. How companies weave AI into their business strategies and approach integrating it into their processes and workflows will determine how successful that AI is.
And to close this post out, I asked ChatGPT to write a haiku about AI Strategy:
“AI learns and grows
Strategy guides its progress
Unleashing its power”
Not bad.
