Product strategy for AI-assisted decision-making

Leading product strategy, research and design for an AI research platform, translating the needs of professionals in regulated industries into a product built around trusted evidence, governed access and human judgement.

I initially joined KiteEdge as a UX consultant. Like many early-stage companies, team size, resources and priorities changed quickly, and people routinely worked beyond the boundaries of their job titles. My role grew with the company: I became a trusted pair of hands across research, product strategy and design, eventually joining the leadership team as CXO and then CPO.

Over that time, I helped move KiteEdge from its roots in enterprise search towards a more distinctive proposition: using AI to support research and decision-making in industries where information is valuable, access is controlled and decisions need to be accountable. Contours became the first expression of that strategy, initially focused on investment professionals and the licensed research they use to make decisions.

I spoke with investment professionals, research providers and others across the industry to understand how research was actually carried out: how information was found and evaluated, where existing workflows created friction, and where AI might genuinely help. This sat alongside conversations with prospective customers, partners and investors, balancing what we were learning from users with technical possibilities and commercial opportunities.

A central focus of my role was shaping the product vision for Contours with the wider team, and translating that vision into product strategy and design. We wanted AI to operate as a trusted research assistant rather than an authoritative answer machine. In a regulated environment, a plausible response was not enough: professionals needed to understand what information the AI could use, where an answer came from and what evidence supported it.

I worked across strategy and implementation to turn that vision into a product, helping define the first MVP and scope our first paid proof of concept. I remained hands-on throughout: mapping research workflows, developing product concepts, designing in Figma, building the design system, prototyping and testing ideas. More recently, that has extended into working React interfaces using Claude Code and Storybook.

Define → Reason → Commit → Act gave us a model for the relationship between researcher and AI. The researcher defines the boundaries; AI helps reason across the evidence; the researcher decides what to commit to the record; and action remains outside the system. Importantly, Commit allowed researchers to select individual passages or sections of reasoned output to construct an evidence case, retaining provenance while making human judgement an explicit part of the process.

Scope, permissions and marketplace connected the user experience to the wider research ecosystem. Contours needed high-quality content, while providers needed confidence that licensing and permissions would be respected. We developed entitlement-aware access and controls over the research universe available to the AI, making scope and changes to it visible and auditable. We extended this into the Contours Marketplace, enabling independent research providers to make their work discoverable through AI while retaining control over their intellectual property.

Citation and verification came directly from early research with investment professionals. The ability to move from an AI-generated claim to the supporting passage in the original document was identified as a significant potential time-saver. We made that movement part of the core research experience.

Multi-modal deep research grew from a problem identified during a paid proof of concept: Senior researchers were spending significant time turning existing research into weekly and monthly client reports. We developed MMDR to reproduce their structure and tone with full citation and traceability, reducing substantial manual work to minutes. The initial response was: “If you can solve this problem, you have a sale.”

But delivering it revealed another problem. The team was concerned that clients knowing AI had been involved could undermine the perceived value of their research. It was a useful product lesson: technical capability and measurable efficiency were not enough on their own; adoption also depended on trust, professional identity and how AI use would be perceived by their clients.

We subsequently tested the capability in another sector, using PubMed data to produce comprehensive thematic reports while significantly reducing the cost of report production.

Contours progressed from an early product concept into commercial use with professional investment teams, growing to more than 50 active users and supporting over 10,000 documents from 60 research providers.