Strategy Experiment Canvas

Convert an experimental idea — however rough — into the Strategy Experiment Canvas, a structured one-pager for pitching a strategy/behavioral experiment to a business or academic audience. The canvas forces clarity on what problem exists, what you think the fix is, how you'll test it, and what you'll compare it against.

The canvas has eight boxes, always filled in this structure and order:

1. Setup / Motivation for Strategy Audience — What is the phenomenon, and why is it interesting and important?
2. Setting + Subjects — Who/what are the subjects? Where do they operate? What is their production function? How many are there?
3. Biz Challenge / Friction — What challenge or problem do they face in improving their performance?
4. Your Insight — What is the unique insight about how the problem/challenge can be addressed? (This box spans the full width — it's the hinge between the problem framing above and the solution below.)
5. Solution (Your Intervention) — How will the solution (i.e., the treatment) be designed to solve the problem?
6. Why + When Will It Work? — What is the mechanism through which the solution solves the problem described above?
7. Null (What's the Right Comparison?) — What is the solution being compared to, and why is that a credible comparison?
8. Impact on the Business — If the solution works, what impact will be seen on the behavior and performance of the subjects, and how will that be measured?

(Attribution when presenting the canvas: © Sharique Hasan, Hyunjin Kim and Rembrand Koning.)

Core workflow

1. Extract what's already there. Read the user's prompt (and any uploaded materials — protocols, IRB docs, prior canvases, past chats) and map whatever is available onto the 8 boxes. Don't force-fit information into the wrong box.

2. Identify gaps and ask before writing. Never fabricate a box's content or paper over a genuine gap with a vague, generic sentence. If a box can't be filled from what's given, ask a short, specific question — grouped together rather than one at a time. Good gap-filling questions map directly to the box's prompt, e.g.:

   • Box 2: "Who exactly are the subjects — a specific population, firm, or platform? Roughly how many, and how are they currently making the decision in question?"
   • Box 6: "What's the underlying mechanism — is this an information/attention effect, an incentive effect, a social/normative effect, a friction-removal effect, something else?"
   • Box 7: "What's the comparison arm — pure control (no intervention), the status-quo process, an alternative treatment, or a placebo/attention-matched condition? Why is that the right counterfactual?"

   Use ask_user_input_v0 when the missing pieces reduce to a short set of concrete choices (e.g., comparison arm type, unit of randomization). Ask in plain conversational questions when the gap is more open-ended (e.g., "why is this phenomenon important?").

3. Flag internal inconsistencies. If the stated insight (Box 4) doesn't actually motivate the proposed solution (Box 5), or the proposed comparison (Box 7) wouldn't isolate the mechanism claimed (Box 6), say so directly and ask the user to reconcile it rather than silently smoothing it over. The canvas is a diagnostic tool — its value comes from surfacing exactly these mismatches.

4. Draft the canvas. Once the essential boxes have real content (a box can stay a placeholder like "[Not yet specified — need X]" only if the user explicitly says they want to leave it for later), write out the full canvas.

5. Offer a round of tightening. Business/strategy audiences reward brevity — after drafting, offer to tighten any box that reads as more than 2–3 sentences, and flag if "Your Insight" is merely restating the challenge rather than adding a genuinely new idea (this is the single most common failure mode of this canvas: an insight box that just repeats the friction).

Output format

Default to a clean markdown rendering of the eight boxes with their italicized guiding question retained under each header, e.g.:

## Setup / Motivation for Strategy Audience
*What is the phenomenon and why is it interesting and important?*

[content]

## Setting + Subjects
*Who/What are your subjects? Where do they operate? What is their production function? How many?*

[content]

...continuing through all 8 boxes in the order above. This is inline conversational content, not a standalone file — don't create a markdown artifact/file unless the user asks to save, download, or share it (e.g., "make this a doc I can send to my co-author"), in which case follow standard file-creation practice (a .md file is sufficient for most cases; use the docx skill only if the user wants a formatted Word document, or pptx if they want a slide version for a talk).

Notes on tone

Push back gently but genuinely — a canvas with soft, hand-wavy language in the Insight or Mechanism boxes is a canvas that hasn't done its job. Better to leave a box marked incomplete and ask one more sharp question than to fill it with filler that reads well but says nothing.