Diversify Your Bets

What's the most basic financial advice you'd give your kid?

Diversify.

(Seriously. That's it. Don't overthink it.)

You know this. Your financial advisor knows this. Your kid probably knows this. Some investments are safer but generate lower returns. Others are riskier but offer higher upside. The magic happens in the blend—a portfolio that balances risk and reward over time.

Not to humble-brag, but I actually studied finance as an undergrad. And this was pretty much the whole curriculum. (Okay, fine, there was also a lot about discounted cash flows. But diversification was the punchline.)

This isn't controversial. Diversification as a core strategy for risk management is Finance 101. So why does everyone forget it the moment AI comes up?

The Explorer Problem

Try this thought experiment.

You're leading a group of explorers who've landed on an unfamiliar shore. You need to find water to survive. You have no map.

What do you do?

If you're smart, you send scouts in every direction. When they return, the scout who went east and found nothing has contributed valuable information—the whole group now knows not to waste time heading east. When the scout who went south reports finding a river, everyone benefits.

I've taught this at the d.school for over a decade: the group that diversifies its exploration is the group that survives.

We call it the "portfolio approach." Just like a financial portfolio spreads risk across asset classes, an innovation portfolio spreads risk across experiments. As I wrote about recently, research suggests roughly 95% of AI projects fail to deliver on their original objectives. If you're running one AI project and hoping it's the winner, you're not being careful—you're betting your organization's future on a single number at the roulette table.

The math says you need more attempts than you think. Not 10-20% more. Ten to twenty times more.

Rick Rubin Gets It

Rick Rubin operates this way instinctively. When working on an album, he doesn't record one version of each track. He generates options. Lots of them.

For Adele's 21—which won the Grammy for Album of the Year—Rubin didn't fly solo. The final album featured multiple producers, each bringing different sonic perspectives. "Rolling in the Deep" was demoed with Paul Epworth, then re-recorded with Rubin, but the team preferred the original demo. Different inputs. Better outputs.

His philosophy? Deliberately generate more options before making decisions. Wait to prune. (How many organizations prune before anything has a chance to flourish?)

Steve Jobs Commissioned Parallel Teams

Steve Jobs didn't just preach this philosophy—he built Apple around it.

When developing the original iPhone, Jobs didn't put all his chips on one approach. He commissioned two completely separate teams:

P1 was led by Tony Fadell, the iPod's creator. His team built an iPhone based on the iPod's clickwheel interface—essentially an iPod nano with a phone inside.

P2 was led by Scott Forstall, who pursued the riskier multi-touch approach that nobody had ever tried in a consumer device.

Jobs let both teams run in parallel for six months. When the P1 clickwheel prototype proved frustrating, Jobs made the call: "We all know this is the one we want to do. So let's make it work."

The iPhone that changed the world came from P2.

But here's what most people miss: Jobs couldn't have known P2 was the right answer without also commissioning P1. The parallel development wasn't waste—it was risk mitigation.

(I know what you're thinking: "Sure, but Jobs was Jobs. I don't have Apple money." Stay with me.)

Where I Got This Wrong

For years, I preached portfolio thinking while running my own safe-bet-only portfolio.

Weekly blog posts. Podcast episodes. Speaking engagements. All valuable—but all the same asset class. Municipal bonds. Safe, predictable returns. No real downside risk. No real upside either.

Then I almost said no to a book deal.

Here's the thing about a book: it's an 18-month bet. No guaranteed payoff. You're betting time, reputation, and opportunity cost on something that might not work. It's a growth stock in a portfolio that had been all bonds.

I caught myself hemming and hawing—reading my own advice to a CEO client about how "goals achievable for mere humans are too small"—while simultaneously finding reasons to play it safe. The hypocrisy hit me like a gut punch.

So I said yes. And suddenly my portfolio had a Horizon 3 bet in it.

The point isn't the book. The point is that diversification works at every scale—and I wasn't practicing what I'd been preaching for a decade.

The AI Application

If diversification is Finance 101, and it's literally how Rick Rubin and Steve Jobs created some of the most successful creative products in history, why have so few organizations applied it to AI?

I see two blind spots.

Blind Spot #1: No portfolio across risk categories.

Most organizations I encounter have exactly one type of AI project: safe, incremental, low-risk efficiency plays. "We're using AI to summarize meeting notes." "We're piloting AI for customer service transcripts."

Those aren't bad projects. But they're the municipal bonds of AI strategy. Safe. Predictable. Barely beating inflation. (And everyone's doing them, so where's your competitive advantage?)

Meanwhile, almost nobody has a balanced portfolio that includes higher-risk, higher-reward possibilities. Nobody's commissioning the equivalent of P1 and P2.

Playing it safe isn't actually safe—it's the riskiest strategy of all, because you're guaranteeing you'll never discover what you're capable of.

Blind Spot #2: No duplication of effort.

This one makes executives squirm. Conventional corporate wisdom says: one team, one project, no duplication. Redundancy is inefficiency.

But Wade Foster, CEO of Zapier, has a different take. When Zapier declared "Code Red" in 2023 and transformed into an AI-first company, they didn't try to eliminate duplication. They embraced it.

"Tolerance for duplication," they called it. "We didn't try to clean everything up in real-time. Some AI experiments... didn't take off. But we let them run."

Wade told us directly on Beyond the Prompt: "The first order thing for a leader is you have to know, as a leader, duplication of effort is required. Then the question becomes, how do I manage people's psychology who feel like this is wasted effort."

When I first heard this, I pushed back hard: "But what about focus? What about resource constraints?"

His response shifted something for me: "Those are real concerns—but they're not solved by running one experiment. They're solved by knowing which experiments to stop. You need parallel paths to have something to choose between."

Duplication isn't waste if it's how you learn which path to take.

The Case Against This Advice

A skeptical reader might say: "This is advice for Apple and Zapier. I have 50 employees and one AI budget line."

Fair point. Let me steel-man the counterargument.

There are absolutely times when focus beats diversification. SpaceX bet everything on a single rocket architecture. Amazon poured resources into AWS before it was profitable. Sometimes, when you have genuine conviction and proprietary insight, concentration wins.

But here's the difference: those companies had deep domain expertise that gave them conviction. They knew something the market didn't.

Do you know which AI approach will work for your organization? Really know?

If the honest answer is "no"—and for most of us, it is—then diversification isn't a failure of strategic nerve. It's the only rational response to genuine uncertainty.

The trap is pretending you have conviction when you don't, then betting everything on one approach and calling it "focus."

What a Diversified AI Portfolio Actually Looks Like

Here's a simple framework. Aim for something like:

60% Learning Experiments: Low-risk, personal or small-team experiments designed to build capability. Meeting summaries, email drafts, research synthesis. Municipal bonds.

30% Capability Experiments: Moderate-risk projects that could change how your team works if they succeed. Workflow automation, customer service augmentation, internal tools. Corporate bonds.

10% Moonshots: High-risk experiments that would transform your business if they worked. New products, new business models, things that make your CFO nervous. Growth stocks.

The specific percentages matter less than the principle: you need all three categories.

And you don't have to commission two full teams on day one. Start with two small experiments. Let the winners earn expanded investment. Staged duplication is still duplication.

The Hard Part: Knowing When to Converge

I'd be lying if I said generating experiments is the whole game. The hard part is knowing when to converge—which experiments to scale, which to kill, and which to let run longer.

A few principles I've learned (often the hard way):

Time-box everything. Before you start, define how long you'll run parallel experiments. Jobs gave his teams six months. You might need six weeks, or six days. The point is to decide before you start, not when you're emotionally invested.

Define success upfront. What would "working" look like for each experiment? If you can't articulate it before you begin, you'll rationalize any outcome as success.

Make decisions at decision points. Build in explicit moments to evaluate and choose. Without them, experiments drift into zombie projects that never die and never scale.

(Real talk? This is genuinely hard. But the answer isn't to avoid diversification—it's to diversify with discipline.)

The Question

A friend asked me recently whether their company's AI strategy was too aggressive. Too many experiments. Too much uncertainty about which would work.

I told them: that's exactly right.

The explorers who survived weren't the ones who picked one direction and hoped. They were the ones who sent scouts everywhere.

Pull up your AI initiative list. Now be honest: is that a diversified portfolio, or did you just bet everything on municipal bonds and call it strategy?

Related: Create A Portfolio
Related: Host A Shoot Out (Like Rick Rubin)
Related: Stop Being A Hypocrite
Related: Beyond the Prompt: Wade Foster, CEO of Zapier
Related: Update Your Priors
Related: Stop Operating, Start Orchestrating

Join over 27,147 creators & leaders who read Methods of the Masters each week.