Practice 02 · Building AI
You're building an AI product, and you don't yet know if people will trust it, adopt it, or pay for it. We find out before you build, and while you do.
An AI product can work perfectly and still fail, because people do not trust it, do not understand what it is doing, or will not pay for it. Engineering teams rarely have the research capacity to find that out early. Second Mind provides the user experience research layer for AI-driven products and services: validating the concept before a line of code, testing prototypes, and checking that a model's assumptions about people actually hold. It works whether your team is building, a development partner is, or we are.
Using AI inside your firm instead? See AI Workflow Redesign →
Four Engagements
Each engagement answers the question your product team faces at that stage. Use one, or run them in sequence as the product matures.
3–4 Weeks
Should we build this, and for whom?
Surveys and in-depth interviews, integrated in one analysis, to test your core hypotheses before development money is committed.
Used to test an AI wallet concept for a global payments group before MVP build.
3–4 Weeks
Does it work for real people, and what goes in V1?
Participants work through your prototype's actual scenarios while we capture what they do, not just what they say about it.
Shifted a voice-first banking concept to a multi-modal design with voice as an optional layer.
4–6 Weeks
Are the model's assumptions about people true?
For products that score, segment, recommend, or personalize. We test the behavioral logic underneath the algorithm with the people it is meant to describe.
Used to calibrate an insurance propensity model for a new market, 20 in-depth interviews.
Monthly · Aligned to build cycles
Who keeps the user in the room?
A senior researcher attached to your product team or development partner, running short research cycles timed to your sprints.
Scoped per month, with a defined number of research cycles.
Analysis runs through custom-built qualitative coding tools with manual spot-coding for quality control. You get the speed of AI-assisted synthesis without trusting it blindly.
Interview findings are cross-referenced against survey and usage data throughout, so every recommendation says whether it rests on numbers, on people's stories, or on both.
Every insight is mapped to a product decision: what to build, what to cut, what to say, and what to test next. Written for the development, design, and leadership teams who have to act on it.
Built for
How an Engagement Runs
We agree the decisions the research has to inform, the hypotheses to test, and what counts as a clear answer.
Screeners, discussion guides, prototype scenarios, and survey instruments, built around your product and your users.
Interviews and tests with real users, coded with AI-assisted analysis and checked by hand, then joined to your quantitative data.
A decision-ready report and a live session with product, design, engineering, and leadership, with every finding tied to an action.
Recent Work
Fintech · Concept + prototype testing
For a global payments group building a white-label AI banking app. Scenario-based prototype sessions, integrated with survey data, reframed the product from voice-first to multi-modal and set the MVP priorities.
Read the case study →Insurance · Behavioral model validation
For an insurance engagement studio and its reinsurer client. Twenty in-depth interviews recalibrated the risk personas behind a behavioral scoring tool and mapped each profile to products and messaging.
Read the case study →The leadership of Second Mind led user research at Pennsylvania's digital services agency, including a 12-month enterprise AI study with OpenAI, and ran qualitative research for clients including Google and YouTube.
Common Questions
Both practices answer the same question from different sides: will people actually use AI well in this situation? User research is where the leadership of Second Mind started, at Pennsylvania's digital services agency and in qualitative work for clients including Google and YouTube. Product teams building with AI face the same trust, judgment, and adoption problems that firms using AI face inside their workflows, so the methods carry directly across.
Yes, and most engagements work this way. We work alongside your in-house team or your development partner, and we can work white-label under an agency or development shop. If you need the build as well, Build & Implementation can take it on with research included.
It depends on your audience. Many clients recruit through their own customer base or a panel such as Prolific, and we design the screener and handle scheduling. For harder-to-reach audiences we can arrange recruitment. Participant incentives are typically covered by the client.
Transcripts are coded with custom-built qualitative analysis tools, then spot-coded by hand for quality control. That speeds up synthesis without handing the judgment to a model. Participant data stays confidential and is handled under the terms of your agreement.
Yes. Much of this work involves unreleased products, so confidentiality is the default. Case studies are only published in anonymized form or with the client's permission.
Start the Conversation
A 30-minute call to talk through your product, where it is in the build, and what you most need to know before the next decision. No pitch, no pressure.
Book a Product Research Call