A one-to-two-week engagement to figure out exactly how AI should fit into your product. We map where AI earns its place, which models and tooling to use, how it integrates with your stack, and what it costs to build, ship, and run.

Traditional discovery scopes a product. AI Discovery scopes the AI inside one. We get into the specifics: which features benefit from AI, which models to use, where copilots earn their cost, and which AI bets to make first.

Because AI features carry their own economics: token costs, latency budgets, model selection, fine-tuning, RAG infrastructure. Treating AI like any other feature is how AI projects blow up. Discovery scopes those questions before the build.

Both. Most clients fall into one of two paths: a live product looking to add AI features, or a new AI-first product looking to scope it from scratch. The discovery shape is the same; the inputs differ.

One to two weeks from kick-off call to final document. A few hours of working sessions across the engagement; the rest is analysis and recommendation.

Usually within a week of the first intro call. We block the senior engineer and designer, prep the kick-off, and schedule the working sessions around your team.

Skipping discovery is the most common reason AI projects fail. Token costs balloon, models get picked wrong, RAG architecture turns brittle, and compliance arrives after the architecture is already locked. Discovery surfaces those decisions before you commit budget to a build.

A single planning document covering: AI features mapped against business value, recommended models and tooling, RAG or fine-tuning approach where relevant, integration architecture, time and cost estimates, a risk register, and a prioritized roadmap.

Detailed enough for any senior developer to start building. Architecture diagrams, model selection logic, integration patterns, data flow, and the recommended stack. Every page earns its place in the document.

Yes. AI features cost money to build and money to run. We model both so the roadmap reflects total cost of ownership: token usage at scale, infrastructure, and ongoing fine-tuning or retraining where relevant.

You do. The plan is yours, the recommendations are yours, the document is yours. Build it with us, with your in-house team, or with another agency entirely.

This is one of the most common reasons companies come to us. Discovery turns a vague AI mandate into a concrete plan: where AI fits in your product, what it costs to build, what it returns, and what to ship first.

Yes. Discovery assesses what to keep, what to rebuild, and what's missing for production: real architecture, proper test coverage, a scalable data model, a deployment pipeline. The output is a plan to take a vibe-coded prototype into a real AI product.

Often the best use of Discovery. Many in-house teams use it to validate AI architecture choices, stress-test model selection, or get a second opinion before committing to an expensive AI build. Your CTO or lead engineers can join the working sessions.

Yes. We audit the existing product and surface where AI features earn their cost, then produce a roadmap for adding AI in stages while keeping the existing product stable as you go.

Yes. AI Discovery is built for exactly this: turning an early concept into a clear, realistic plan with the AI architecture, scope, and roadmap to take it forward.

Any product with workflows or data that AI can meaningfully change: SaaS, fintech, healthcare, analytics, marketplaces, enterprise platforms, AI-first products. The framework adapts; the output stays concrete.

Balanced. Deep enough for engineers, clear enough for product leads or non-technical readers. The goal is alignment across the people who matter, with plain-language framing wherever possible.

A senior engineer and a senior product designer run the engagement. A project manager joins when scope or coordination calls for it. Every recommendation comes from people who have shipped AI products in production.

Yes. Most engagements are fully remote, with sessions recorded and decisions documented in real time so the whole team can stay aligned.

Miro or FigJam for mapping, Figma for wireframes, Notion or your tool of choice for the document, and our internal AI tooling to support stack analysis where it earns its place.

You walk away with the document and the plan. The recommendations are vendor-neutral, the architecture is the architecture, and any senior team can build from the output. About a third of Discovery clients hand the document to an in-house or other team and that's the right outcome for them. Both paths work.

Yes. The same team can move straight into AI Design Team, AI Dev Team, AI Product Team, or AI Agents engagements with onboarding and scope translation already done. Or you can hand the document to your own team.

We use enterprise AI tooling with private workspaces, with client data kept out of training and signed agreements covering anything sensitive. Recommendations on your AI stack include data residency, retention, and model isolation choices appropriate to your industry.

Speed and substance. We've shipped AI features in production, so the recommendations come from real build experience. Senior engineers and designers run the sessions, AI tooling supports the analysis where it earns its place, and the output is a document engineering can actually build from.