In my previous article, we discussed the dangers of over-optimization — how an obsession with incremental improvements can lead to stagnation and irrelevance. Now, we’re going to take that idea further, using a powerful metaphor from the world of machine learning: the gradient ascent algorithm.
In our relentless pursuit of perfection, have we forgotten the art of exploration? As we dive deeper into the age of AI and machine learning, it’s time we learn that we should not just use algorithms for solving problems, but apply their core principles to business itself. Today, let’s talk about gradient ascent and its surprising lesson for innovation in business.
Gradient Ascent
For those unfamiliar, gradient ascent is an optimization algorithm commonly used in AI and machine learning. Its goal? To find the highest point (or global maximum) in a complex landscape of possibilities.
Gradient ascent is like climbing a mountain in fog — you can’t see the summit, but you can repeat steps always in the direction that goes up the steepest, until you can’t go any higher.
The challenge? You might reach a small peak (local maximum) thinking it’s the mountain top, when the actual summit (global maximum) is hidden in the fog. In business terms, you might optimize your current strategy to its peak, missing a completely different, more successful approach, by being too focused on always moving uphill to find it. Sound familiar? It should — it’s not unlike a company searching for the optimal business strategy.
The Power of Suboptimal Steps
Companies often fall into the trap of a local maximum. They optimize and optimize, always chasing the next small improvement, the next profit-boosting measure, the next minor feature addition. They’re stuck in a local maximum, convinced they’re at the peak of success. But what if the true peak, the global maximum of potential, lies beyond a valley of temporary setbacks and risks?
Here’s where it gets interesting. To escape local maxima, advanced versions of gradient ascent sometimes do something counterintuitive: they accept steps that initially seem worse. They’re willing to descend, to temporarily move away from the apparent optimum, in the hopes of finding something better.
This is the leap of faith that businesses need to embrace.
These are some of the approaches the algorithms use:
-
Random Restarts: Occasionally start over from different random points. In business, this is like trying entirely new strategies or markets.
-
Simulated Annealing: Start “hot,” accepting many downhill moves. Gradually “cool,” becoming less likely to accept worse positions. Early on, it’s like a flexible startup open to drastic changes. As it “cools,” it’s more like an established company, still open to occasional risks but increasingly focused on incremental improvements. This approach allows thorough exploration early, settling into fine-tuning later.
-
Stochastic Gradient Ascent: Use random subsets of data for each step, introducing some noise. For businesses, this could mean making decisions based on varied, sometimes unconventional data sources.
-
Learning Rate Adjustment: Dynamically change the size of steps. In business terms, this is about being flexible in how drastically you change your strategy based on current results and market conditions.
These methods all introduce elements of exploration and randomness, helping to avoid getting stuck in suboptimal solutions. In a business context, they represent different ways of balancing exploitation of known strategies with exploration of new possibilities.
Origination: Your Business’s Escape Mechanism
In the world of business, these exploratory, “suboptimal steps” are what we call origination — the pursuit of fresh ideas and novel approaches that might not immediately show positive results on standard metrics.
It’s the electric light bulb in a world of optimized candles. It’s the automobile in a landscape of faster horses. It’s the smartphone in an era of increasingly efficient landlines.
Cultivating Your Company’s Gradient Escape
So, how do we apply this lesson? Here are a few strategies to consider:
-
Embrace Controlled Chaos: Set aside resources — time, money, talent — for projects that don’t need to immediately prove their worth. Create a playground for your originators.
-
Redefine Success Metrics: Short-term ROI isn’t everything. Develop new ways to evaluate innovative ideas that might not show immediate returns.
-
Celebrate “Intelligent Failures”: When a risky venture doesn’t pan out, treat it as a learning experience. What did it teach you about the landscape of possibilities?
-
Cross-Pollinate Ideas: Encourage collaboration between different departments, or even with other industries. Fresh perspectives can help you see the valleys worth traversing.
-
Train for Exploration: Develop your team’s creative muscles. Teach them to question assumptions, to look beyond the immediate horizon.
The Future Belongs to the Explorers
In our previous discussion, we talked about the coming economy of creativity. Now, we see it’s not just about creativity — it’s about the courage to explore, to take those seemingly suboptimal steps in pursuit of greater heights.
The companies that will thrive in the future aren’t just the ones with the best optimization algorithms. They’re the ones willing to occasionally descend, to traverse difficult terrain, in the relentless pursuit of true innovation. What could your company achieve if it were willing to take more of these “suboptimal” steps? How might your industry landscape change if more businesses adopted this exploratory mindset within it?
So, to the business leaders reading this: Are you stuck optimizing candles, or are you ready to illuminate new worlds? The choice is yours, but remember — in the grand algorithm of business, sometimes the best move is the one that doesn’t immediately look optimal.
Let’s escape our local maxima. Let’s embrace origination. The global maximum awaits. What will be your first step towards finding your company’s true peak?