In the wild, survival doesn’t require you to outrun the lion—it just means outrunning the slowest gazelle. Surprisingly, this logic also applies to business strategy.
When we introduce AI into business decision-making, we’re not just dealing with isolated optimization problems—we’re engaging in a complex game, with rivals, competitors, and market players who also make moves. One key trap in this game is assuming that opponents are perfect. That assumption sounds safe—but it can be paralyzing.
Why? Because when your strategy assumes the opponent never makes a mistake, then any deviation on your part may seem useless or even dangerous. You might fail to explore comeback options when you’re behind—or abandon promising but imperfect plans—just because your model doesn’t believe your opponent is human.
Here’s where inspiration from cutting-edge AI game-playing algorithms can reshape our strategic thinking.
Why AlphaZero Was Just the Beginning
AlphaZero famously mastered chess, shogi, and Go from scratch using a self-play reinforcement learning approach and a Monte Carlo Tree Search (MCTS) algorithm called PUCT1. The AI assumed every opponent was perfect, exploring moves with a mix of prior probability and uncertainty bonuses.
Great in theory. But in real markets—or even human games—opponents are rarely perfect.
Enter Search-Contempt: A Smarter Assumption
A recent breakthrough called search-contempt offers a compelling twist: it modifies MCTS by not overestimating the opponent’s intelligence. Instead of endlessly recalculating based on every opponent’s move, it freezes assumptions after a certain point (the Nscl threshold), treating the opponent as fallible and realistically limited.
Here’s what’s fascinating: this shift doesn’t just make AI faster—it makes it more human-aware. Some positions may seem easy for machines to evaluate (because they require brute-force calculations) but are tricky for people due to hidden tactics. Other times, humans may navigate complex strategic ideas that confuse machines. The Nscl threshold mimics these moments of human fallibility, by recognizing where models frequently misjudge—even when those misjudgments are not due to computation limits, but because humans interpret situations with different heuristics, biases, or strategic blind spots.
In game terms, this means exploring positions where the opponent is more likely to mess up—even if those positions wouldn’t be pursued under perfect-play assumptions. The result? Richer training data, faster learning, and better performance with less compute.
In business terms, this means AI systems can start to:
- Model adversaries more realistically (e.g., a slow-moving incumbent, not an all-knowing oracle)
- Focus decision paths on exploitability, not perfection
- Find practical, winnable opportunities rather than ideal but unreachable strategies
Business Decisions Are Competitive Games
Let’s look at three common business contexts—investment, pricing, and product launch—as competitive games.
1. Investment Decision:
- Context: A firm must decide whether to enter a new market (e.g., green energy) where competitors are also evaluating entry.
- Choice Set: Timing of entry, scale of investment, tech stack.
- PUCT-style strategy: Considers all competitors to respond optimally—might delay investment until the market is “perfect”.
- Search-contempt strategy: Assumes competitors may hesitate or misprice risk. Encourages earlier entry and tests adversary inertia.
2. Pricing Strategy:
- Context: Two companies are engaged in price competition in a saturated market.
- Choice Set: Pricing level, discount period, loyalty bundle.
- PUCT-style strategy: Over-anticipates counter-moves; converges toward defensive pricing.
- Search-contempt strategy: Recognizes that competitors may stick with legacy pricing models. Suggests bolder, time-sensitive discounts.
3. Product Launch:
- Context: A startup prepares to launch an AI feature in an ecosystem dominated by incumbents.
- Choice Set: Feature completeness, release timing, market messaging.
- PUCT-style strategy: Delays launch until perfection, fearing incumbents will instantly retaliate.
- Search-contempt strategy: Banks on slower responses; launches MVP to capture early adopters.
Summary Table: Strategic Logic Comparison
| Scenario | PUCT Logic Assumes… | Resulting Behavior | Search-Contempt Assumes… | Resulting Behavior |
|---|---|---|---|---|
| Investment | Rivals respond optimally | Over-cautious delay | Rivals misjudge or delay | Seizes first-mover advantage |
| Pricing | Rivals immediately adjust prices | Defensive pricing equilibrium | Rivals are rigid or anchored | Bold, opportunistic pricing |
| Product Launch | Rivals instantly respond with retaliation | Wait for perfection | Rivals delay or dismiss MVP threats | Launches early, gains traction |
How to Win More with Search-Contempt Logic
Let’s illustrate each case with reasonable, realistic opponent mistakes—and how a search-contempt AI helps you respond.
Investment: Overhyped Sector, Cautious Herd
You’re considering investing in green hydrogen tech. The PUCT logic tells you everyone will rush in soon, so better wait for the perfect data. But search-contempt logic notices:
- Sentiment is too cautious after recent failures.
- Your competitors are waiting for subsidies or ESG certification.
- A large institutional fund recently signaled interest but hasn’t acted.
With this context, your AI proposes:
- Enter now with moderate capital.
- Secure supplier relationships before others react.
- Target early adopter governments who favor first-movers.
Pricing: Competitor Clings to Prestige Pricing
You’re launching a mid-range smartphone. Your biggest rival positions themselves as a premium brand with fixed price anchors.
- PUCT says: Don’t price aggressively—they’ll undercut you.
- Search-contempt realizes: They’ve publicly committed to premium pricing and are slow to discount.
Your AI recommends:
- Launch with a 10% discount in Q1.
- Use trade-in programs to lure their mid-range buyers.
- Run limited-time bundles that exploit the gap.
Product Launch: Incumbent Misses the Signal
You’re shipping a productivity AI tool. The dominant player is focused on large enterprise contracts.
- PUCT: Wait—they’ll copy your product.
- Search-contempt observes:
- They’re in procurement cycles for 9 months.
- Their roadmap doesn’t mention SMEs.
- Their internal dev team is reorganizing.
Your AI suggests:
- Launch MVP targeting freelancers and startups.
- Build a waitlist for an early-access Pro version.
- Push integrations that are hard for the incumbent to copy fast.
In all three cases, your AI is playing not to beat perfection, but to beat what’s likely to be flawed—just like real-world competition.
AI for Advantage, Not Perfection
Search-contempt teaches us that faster progress comes from modeling others as they are—not as they should be. Just like in the wild, you don’t need to beat the best possible version of your rival—you need to outmaneuver the actual one.
At Cognaptus, we believe this insight will shape the next generation of strategic AI tools—leaner, sharper, and grounded in real competitive logic.
So the next time you build your AI strategy, ask yourself:
Are you chasing the lion, or just trying to stay ahead of the herd?
References
Joshi, A. (2025). Search-contempt: a hybrid MCTS algorithm for training AlphaZero-like engines with better computational efficiency. arXiv:2504.07757 [cs.AI]. https://arxiv.org/abs/2504.07757
By Cognaptus Insights. For smarter, faster, more strategic AI decisions.
-
While classic MCTS favors moves based on how often they lead to good results in random simulations, PUCT adds guidance from a trained neural network—combining what the AI already “believes” (prior probabilities) with actual outcomes. It’s like choosing your next move based not just on what worked before, but what your coach suggests might work well. ↩︎