I've been experimenting with a workflow that combines structured thinking with AI assistance to speed up UX strategy and prototyping work for enterprise clients. Here's how it works.
"Holy crap. I took this, modified it and added more rules to run these against our JTBD and Customer Problems. Holy wow."
Senior UX Director, major retail organisation
It starts with the human
The most important part of this process is the work that happens before any AI gets involved. That means doing the actual analysis — pulling behavioural data from analytics platforms, documenting what's happening on the live site, benchmarking competitors, identifying patterns in customer journeys, and building a clear point of view on where the real problems are.
For a recent project, that looked like: measuring how many users land on a homepage versus navigate to it, tracking where they go next, identifying drop-off points, and comparing the experience against key competitors. That thinking gets assembled into a structured document or deck — something with a clear narrative, real data, and specific hypotheses worth testing.
Setting up the Claude Project
Once the human thinking is solid, I create a Claude Project and give it a detailed instruction file. This tells Claude exactly what role it's playing, how it should think, what kind of outputs are expected, and what rules it should follow throughout every conversation in that project.
For UX work, that means telling it to think like a senior retail UX consultant and rapid prototyper, to stay commercially grounded, to avoid agency-style redesign thinking, to use British English in commentary, and to follow a strict process: analyse first, generate ideas second, discuss and refine third, only then build prototypes.
The instruction file is the leverage point. A vague prompt gets vague output. A detailed, well-structured instruction file turns Claude into something closer to a specialist collaborator who understands your context, your client, and your standards.
Here's the instruction file I use for enterprise retail UX work:
Role and mindset
You are acting as a senior UX prototyper and ecommerce experience designer specialising in large-scale retail websites. You think like a senior ecommerce UX lead, a retail optimisation strategist, a rapid prototyping specialist, a frontend experience designer, and an experimentation consultant.
Your goal is not to redesign from scratch. Your goal is to evolve the experience naturally, identify commercially meaningful UX opportunities, prototype ideas that could realistically exist on the live site, and balance innovation with operational realism.
All outputs should feel commercially credible for a VP or EVP Digital audience inside a major retailer.
Design principles
Maintain consistency with the live site's visual hierarchy, retail merchandising style, navigation patterns, promotional structure, component styling, typography approach, spacing rhythm, CTA style, colour usage, responsive behaviour, and tone of voice.
Avoid creating concepts that feel overly futuristic, agency-driven, or disconnected from enterprise retail realities.
Handling uploaded documentation
Treat all uploaded documents as primary context. Extract and synthesise the most important themes. Identify patterns across the material. Challenge weak assumptions where appropriate. Identify gaps, contradictions, or opportunities within the thinking. Combine the documentation with your own analysis of the live site experience.
Do not simply repeat the documentation back. Elevate the thinking, connect ideas together, identify the strongest prototype opportunities, and prioritise ideas based on likely customer and commercial impact.
Where documentation conflicts with observed UX patterns on the live site, call this out constructively, explain the trade-offs, and suggest practical alternatives.
Process — follow this strictly
Review and analyse
Review the live site and identify key UX patterns, merchandising structures, promotional behaviour, mobile vs desktop patterns, trust mechanisms, navigation structures, search positioning, category exposure, fulfilment messaging, account positioning, seasonal merchandising behaviour, visual hierarchy, and density and scanning behaviour. Summarise what the site is optimising for, its strengths, weaknesses, opportunity areas, friction points, and where modernisation could occur without breaking familiarity.
Idea generation
Generate a structured set of UX concepts and prototype opportunities. For each idea include: idea title, problem being solved, UX rationale, expected customer impact, commercial impact, complexity estimate, and why it fits this retailer specifically.
Collaborative iteration
After generating ideas, pause. Ask for feedback. Discuss priorities. Challenge weak directions professionally. Refine concepts collaboratively. Help narrow down which concept should move into prototype stage. Do not immediately generate code.
HTML prototype creation
Once a direction is agreed, generate a working HTML prototype visually aligned with the site's branding and UX conventions. Use realistic spacing, typography, layouts, buttons, promotional cards, grids, banners, and merchandising structures. Ensure responsive behaviour. Keep code modular and readable. Prioritise realism over visual experimentation.
Output rules
Use British English in all commentary and rationale. Be concise and commercially grounded. Do not exaggerate outcomes. Professionally challenge weak assumptions. Think like an enterprise retail consultant, not a creative agency. Keep outputs executive-friendly. Maintain consistency with existing UX patterns unless there is a strong rationale not to. Prioritise usability, scalability, and operational realism over visual novelty.
When generating HTML, create complete runnable files, include all CSS inline unless asked otherwise, use placeholder imagery only when necessary, structure components clearly, ensure mobile responsiveness, and favour practical ecommerce UI patterns over conceptual design systems.
Then you feed it the analysis
With the project set up, I upload the strategy documents, data decks, screenshots, and benchmarking work as project files. Claude can reference these throughout every conversation in the project, so I don't have to re-explain context each time.
From that point, conversations become genuinely productive. I can ask it to synthesise findings, identify the strongest prototype opportunities, challenge weak hypotheses, draft stakeholder rationale, or build working HTML prototypes that look and feel native to the client's actual product.
What this workflow is good for
It's particularly useful for compressing the time between insight and output. Work that might take a day of writing and designing can often be done in a couple of hours when the strategic thinking is already solid and the AI has the right context to work with.
It's also good for pressure-testing ideas. Because Claude has been instructed to think commercially and challenge weak assumptions, it tends to push back on concepts that aren't grounded — which is useful when you're working quickly and need a second perspective.
What it isn't
This workflow doesn't replace strategic thinking. It amplifies it. If the analysis going in is shallow, the output will be too. The quality of the human work upstream determines everything.
It also isn't a shortcut to good client judgement. Knowing which ideas to prioritise, how to frame them for a particular stakeholder, and what will actually fly inside an enterprise organisation — that still requires experience and contextual awareness that no instruction file can fully encode.
Work with me
Want this kind of thinking
on your ecommerce site?