When I left Amazon, the thing I missed most wasn't the scale, the resources, or the two-pizza teams. It was Working Backwards.
Not as a concept — plenty of people know the concept. What I missed was the discipline of it. The forcing function that made every new initiative start with the customer and work backwards to the solution. The PR/FAQ document that crystallized vision and exposed weak thinking before a single line of code was written.
Over the years since, I've taught Working Backwards to dozens of organizations. I've watched teams transform their approach to innovation by adopting this methodology. And over the past year, I've watched AI transform the methodology itself.
Here's my five-step Working Backwards framework, and how AI is changing each step.
Step 1: Listen — Gather Evidence, Not Opinions
The first step has always been the hardest to do well. You need evidence of customer struggles — real data, real patterns, real pain points. Not opinions from a conference room. This aligns with two Amazon leadership principles: Customer Obsession and Dive Deep.
Traditionally, this meant weeks of customer interviews, survey design, data analysis, and synthesis. Most teams either rushed through it (and built the wrong thing) or got stuck in it (and never built anything).
AI has fundamentally compressed this step. Platforms like Synthetic Users use multi-agent architectures with personality profiles to simulate customer research. AI can analyze customer feedback, support tickets, reviews, and social media at scale to surface patterns that would take a human team weeks to find.
The parity scores between synthetic and real user research now reach 85-92% depending on audience type. That's not a replacement for real research — Nielsen Norman Group is right that synthetic users should complement, not replace, real customers. But it means you can generate and test hypotheses in hours instead of weeks, then validate with real customers from a position of informed curiosity rather than blank-slate discovery.
Step 2: Define — The Five Questions That Change Everything
The Define step centers on five questions: Who is the customer? What is the customer's problem or opportunity? What's the most important customer benefit? How do you know what customers need or want? What does the customer experience look like?
These questions sound simple. They are brutally difficult to answer well.
AI acts as a thinking partner here — not to answer the questions for you, but to stress-test your answers. When you say "small business owners" are your customer, AI pushes back: which small business owners? In which industry? At what stage? When you claim to know what customers want, AI asks for evidence.
I use AI to simulate ten different customer personas answering the five questions. Then I compare the AI-generated answers with real customer data from Step 1. The gaps between what AI predicts and what real customers say — that's where the insights live.
The key principle in this step: think big about problems, not solutions. AI helps you resist the gravitational pull toward premature solutioning by keeping the focus on the customer's world.
Step 3: Invent — Structured Creativity, Not Brainstorming
At Amazon, there is no set method for the Invent step. Innovation can come from anywhere. But in my experience, "innovation can come from anywhere" often means "innovation comes from the loudest person in the room."
That's why I bring systematic inventive methods to this step: SIT (Systematic Inventive Thinking), SCAMPER, and Six Thinking Hats. These frameworks provide structure for creativity — they don't constrain it, they channel it.
AI has supercharged these methods. SCAMPER has seven dimensions of creative exploration — AI systematically works through all of them, generating variations that humans typically miss. Six Thinking Hats requires role-playing different perspectives — AI does this in minutes rather than the hour-long workshops it used to require. SIT's five patterns (subtraction, division, multiplication, task unification, attribute dependency) can be applied to any product or service with AI assistance.
Early data suggests significant impact: organizations using AI for ideation report saving 2-3 hours daily on brainstorming and holding 41% fewer meetings as AI handles divergent thinking asynchronously. The human team's job shifts from "generate ideas" to "evaluate and build on the best ideas."
I've taken this further by creating and open-sourcing an Innovation Skills Suite for Claude Code — ten skills covering SCAMPER, customer discovery, PRFAQ writing, and other systematic innovation methods. Anyone can install these and run structured innovation sessions with AI. It's available on GitHub.
Step 4: Refine — The PR/FAQ as Truth-Seeking Document
The PR/FAQ is the heart of Working Backwards. It's a press release written from the future — describing the product as if it's already launched — followed by frequently asked questions that stress-test every assumption.
AI compresses the drafting cycle dramatically. It structures the narrative, identifies logic gaps, and generates FAQ responses from different stakeholder perspectives. What used to take weeks of iteration can now happen in days.
But here's where I need to share an important tension.
Marcelo Calbucci, author of The PRFAQ Framework, deliberately wrote his book without AI assistance. His argument: "The problem with generative AI is that it takes away from your thinking." AI doesn't help you find the truth in an opportunity — it helps you articulate a version of truth that sounds convincing.
He's not wrong. The PR/FAQ is meant to be a truth-seeking document, not a truth-generating one. The conviction behind the vision must be human. If you use AI to write a PR/FAQ that sounds compelling but doesn't reflect deep personal understanding of the customer problem, you've created a beautiful fiction.
My position: Use AI to sharpen the blade. The force behind the cut has to be yours.
AI should draft, structure, stress-test, and challenge. But the founder, the product leader, the person who will drive this initiative through organizational resistance — that person needs to own the vision at a visceral level. AI can help you articulate what you believe. It cannot believe it for you.
I built OutcomeHack (outcomehack.com) — an AI-guided Working Backwards tool on Base44 — specifically to walk this line. It guides users through the five questions, generates PR/FAQ drafts, and designs validation experiments. But it positions itself as a guide, not an author. The human stays in the driver's seat.
Step 5: Test and Iterate — When Experimentation Costs Approach Zero
The final step is where AI's impact is most dramatic. As I discussed in Issue #21, experimentation costs have collapsed.
Startups using AI during the MVP phase are 40% more likely to find product-market fit and iterate 60% faster. Functional prototypes can be generated in under ten minutes using tools like Claude Code. No-code platforms democratize experimentation for non-technical founders. Time-to-market reductions of 20-50% are now routine for companies effectively using AI.
After the PR/FAQ is refined, you can use AI to generate a functional prototype — a landing page, a workflow simulation, an interactive demo — within hours. Test with real users before writing production code. Iterate based on feedback in days, not months.
This is where Working Backwards and Bias for Action converge. The cost of testing an idea has dropped so dramatically that the biggest risk is no longer building the wrong thing — it's not building at all.
The Through-Line
Working Backwards was always about discipline — the discipline to start with the customer, to resist premature solutions, to write down your thinking, to invite critique, and to test before you scale.
AI doesn't change any of that. It compresses, augments, and accelerates every step. But the discipline — the willingness to be honest about what you don't know, to let customer evidence override your assumptions, to kill ideas that don't survive scrutiny — that remains irreducibly human.
The organizations that will win with AI-enhanced innovation aren't the ones using AI to generate more ideas faster. They're the ones using AI to find truth faster — and having the courage to act on what they find.
Your action step
Pick one initiative or product idea you're currently working on. Run it through the five Working Backwards questions with AI as your thinking partner: (1) Who is the customer? (2) What is their problem? (3) What's the most important benefit? (4) How do you know? (5) What does the experience look like? Don't accept your first answers. Use AI to challenge each response from three different customer perspectives. The gaps and contradictions that emerge will tell you where your thinking needs work.