TF-017

10x Not 10%

If you needed a 27x improvement, would you still use this approach?

Take your current solution and ask: if I needed this to be 27x better (27x faster, 27x cheaper, 27x simpler, 27x more impactful), would I still use this approach?

A 10% improvement lets you optimize within the current paradigm — tune parameters, shave milliseconds, tighten copy. A 27x improvement forces you to question the paradigm itself. You usually can't get 27x by doing the same thing harder. You have to do a different thing entirely.

You don't need to actually achieve 27x. The point is to discover whether your current approach has a ceiling — and whether that ceiling is lower than you think.

flowchart TD
    A["Your current solution"] --> B["Define the key metric"]
    B --> C["Ask: what would 27x look like?"]
    C --> D{"Could your current approach reach 27x?"}
    D -- Yes --> E["You have headroom — keep going"]
    D -- No --> F["You have a ceiling"]
    F --> G["What paradigm shift would be needed for 27x?"]
    G --> H["Is that shift available to you?"]
    H --> I{"Worth pursuing?"}
    I -- Yes --> J["Pivot approach"]
    I -- No --> K["Continue current approach with eyes open about the ceiling"]

    style C fill:#f39c12,color:#fff
    style F fill:#e74c3c,color:#fff
    style E fill:#2ecc71,color:#fff
    style J fill:#3498db,color:#fff
evaluate / plan · deep · Evaluation · seed: basalt
more

When to Use

  • You're refining and polishing a solution that works but doesn't excite
  • You're in an optimization loop — making things 5%, 8%, 12% better
  • A competitor, new technology, or market shift could make your approach obsolete
  • You want to check whether you're climbing the right hill or just climbing efficiently

Example

Solution: You've built a support chatbot that resolves 30% of tickets automatically. You're tuning prompts and adding FAQ entries to push it to 35%.

The 10x question: What would 300% resolution look like — resolving every ticket automatically?

What it reveals: You can't get there by adding more FAQ entries. The ceiling is the architecture: the bot can only handle questions it's been explicitly programmed for. 10x would require the bot to understand the product deeply enough to diagnose novel problems — a fundamentally different system (retrieval-augmented generation over internal docs, tool-use for account lookups, etc.).

The decision: Maybe you don't build the 10x version today. But now you know your current approach tops out around 40-50%, and you can make an informed choice: keep optimizing for near-term gains while planning the architectural shift, rather than being surprised by the ceiling later.

Watch Out For

  • This move can paralyze if you take it too literally. The goal is awareness, not abandonment. Most days you ship the 10% improvement. But you should know where the ceiling is.
  • "10x" is a thought experiment, not a requirement. If your domain doesn't have a meaningful 10x, pick a different multiplier — 3x or 5x still works.
  • Don't use this to dismiss solid, incremental work. Incremental work compounds. The danger is only when you don't realize you're on a curve that flattens.
  • Sometimes the 10x answer is "do less, not more." Removing steps often outperforms adding features.
pairs with: TF-022, TF-031, TF-029
back next re-roll home

← back · next →