Kinnect is a family office platform. It gives families, and the advisors who work with them, one place for everything they own, owe and plan: accounts, trusts and companies, documents, estate plans and the tasks that keep it all current. I joined in February 2025 as lead product designer and have shaped every major part of the product since: the navigation, the home page, the data vault, the net worth reports, the ownership map and, above all, Kinnect AI, an agent that lets a family operate the whole platform through conversation.

Role: Lead product designer, partnering with product and engineering
Tools: Figma, FigJam, Claude Code, Figma MCP
Scope: AI agent, home page, data vault, net worth reports, ownership map, navigation, Complete System Design

About Kinnect

Every family runs a family office. Most run it from memory.

The wealthiest families have always had family offices: people who know where everything is, who owns what, and what has to happen next. Most families have the same job and none of the staff. Their picture of what they own lives in a filing cabinet, a spreadsheet and one person's head, and it breaks the moment that person is not there. Kinnect exists to make that picture complete, understandable and actionable, for the family, for their advisors and for whoever has to step in next.

What the Product has to do

• Bring it together: accounts, trusts, companies, documents, people and plans in one record.

• Make it understandable: a family should see what they are worth, how it is held and what changed, without a finance degree.

• Make it actionable: every gap becomes a clear next step, owned by the right person.

• Get families set up: none of this works until the data is in, so the product has to carry people through onboarding.

Where it started

A powerful system that was hard to read.

When I joined, the platform already held a lot of capability, but the interface hid it. The home page looked unfinished, the data page was close to unusable, the net worth charts were dense and hard to read, and the ownership map made it difficult to see who owned what. Kinnect's data model is genuinely complex: households, trusts, LLCs, vaults, role players and documents that link to all of them. The interface was passing that complexity straight on to the family. From the first look I knew it needed a redesign, not a polish.

The real problem was not missing components. The system had no rules and no way to prove anything, so drift was the default. Adding AI would only have made it drift faster.

My role

Leading design across the whole product.

As lead product designer I own design at Kinnect from the first conversation about a problem to the hand-off to engineering. I partner with the head of product and the engineering team to shape each requirement, set the design direction, design the high-fidelity screens and every state behind them, and own the design system that all of it is built from. I learned the domain quickly, from trusts and entities to estate flows, because good design for a family office depends on understanding how families actually hold and pass on wealth.

How I Work

• Learn the domain first: I studied estate planning and how family offices operate, so the interface reflects how wealth is really structured.

• Design against the whole system: every screen is designed with the full data model in mind, never in isolation.

• Revisit what looks finished: when a design worked in demos but not for new families, I redesigned it.

• Build once, reuse everywhere: every new pattern goes into the Kinnect Design System so engineering builds it once.

Product Management

Deciding what gets built, not only how it looks.

From early 2026 I took on product management alongside design. I meet the stakeholders directly, agree which problems are worth solving and what the product should do about them, then write those decisions as sprint requirements on Monday.com and follow them through delivery with engineering.

Results

• 75 percent faster requirement handoff: from two weeks of back-and-forth to about half a week, because the sprint tasks now come straight from the problems agreed with stakeholders.

• 70 percent less rework: requirements arrive complete, so far fewer rounds of revision.

• 11 sprints planned and tracked with engineering, the last 5 on Monday.com.

• Earlier feedback: with Claude Code and the Figma MCP, a 20-screen feature reaches its first draft in one to two days instead of half a week.

Kinnect AI: the hero feature

Not a chatbot. A second way to operate the entire platform.

Kinnect AI was the biggest thing I have designed, close to a product of its own. The brief was never to add a chat window. It was to let a family run a complex system through conversation: find and open any item, read uploaded documents, fill in records, answer questionnaires and act on what it finds, without first learning where everything lives in the navigation. I designed the agent end to end: how it is summoned, how it introduces itself, how it shows its work, and every state in between.

Show the work, not just the answer.

In wealth management a confident wrong answer is worse than no answer. So when the agent reads a document, it does not reply with a paragraph. It returns a structured result: what it checked, what matched, what needs the family's decision and what the documents simply do not say, each line with its source. The family can accept the recommendations in one go or decide line by line. Trust comes from being able to check.

Every answer is a doorway.

Responses link straight to the trusts, accounts, people and documents they mention. The conversation becomes the fastest route to any item in the platform, instead of a dead end that the family has to leave to act.

Context that matches how families work.

Conversations are organised as sessions tied to the project or onboarding step they belong to, so a family can pick up exactly where they left off. An account perspective selector makes it explicit whose household the agent is working on, which matters because advisors look after many families at once.

A guide with a face.

The agent appears as Joe, a friendly avatar that waits on every page as a small Talk to Joe launcher. On a first visit Joe welcomes the family and offers a short introduction to Kinnect, which they can start or skip. Returning users are greeted with where they left off. A familiar character made a serious financial tool feel approachable.

Designed for every state, not just the happy path.

I designed the full set: an input that grows from one line to a maximum height for long questions, attachments and uploads, stop generation, copy and feedback on each response, questionnaires inside the chat, completed badges, loading and error states, and minimised and full-screen views. Each became a reusable component in the design system, so engineering could build the agent once and use it everywhere.

Outcome

Kinnect AI is now the conversational layer across the platform. The same agent welcomes new families, guides them through onboarding, reviews documents against the records they describe and answers questions about the family's data, from any page. For a typical family, building their data from documents used to take 3-4 hours of manual entry. With Kinnect AI it is under an hour of review, and about 10 minutes for smaller families.

Home page: the second hero

Three versions of the home page, each one earned.

The home page is where a family decides whether Kinnect is worth their time. I redesigned it twice, and each version solved the problem the previous one revealed.

From a list of links to a picture of the family's wealth.

The original home page looked basic and incomplete, and I knew at first glance it had to change. As part of a complete overhaul of the platform, including a new sidebar navigation, I rebuilt it as a widget dashboard: the family's network, net worth, cash flow, data status, recent templates and financial accounts in one view. For the first time a family could understand their data and their wealth from a single screen.

A dashboard is only as good as the data behind it.

The widget home worked beautifully for families whose data was already in, and poorly for everyone else: new families landed on empty widgets with no clear way forward. So in August 2026 I put a new focus area above the widgets: a guided onboarding module with overall progress, project milestones, a single Continue button, and the next step and the one after it spelled out. A short AI-written summary tells the family what they have done and what comes next.

Keep the payoff in view.

I kept the widgets directly below onboarding, so each finished step makes the family's own data appear underneath: income and expenses, tasks, document links, AI suggestions and what they share with family members. Progress is never abstract; the family can see the picture filling in.

Outcome

The onboarding-first home page changed how quickly families get going. Setting up a family's data used to take one to two weeks. With the guided flow it now takes one to two days once their documents are ready, about 85 percent faster.

Data vault

From a confusing file list to the backbone of the product.

A family's documents are the evidence behind everything else in Kinnect: the trust agreement behind the trust, the statement behind the account. The old data page showed them as a flat list inside groups, with little sense of what each file was, who it belonged to or where it lived, and families found it confusing. I rebuilt it around how families think about their records. Categories run down the side with their own icons: entities, financial accounts, physical assets, investments, estate plan, insurance, identification and more. The document list shows each file's type, tags, links, last update and vault at a glance. Folders, filters and counts sit on top, and AI suggests clean names for messy files. The vault redesign became a foundation for how Kinnect organizes and presents data across the product.

Net worth and reports

Charts that answer a question.

Net worth is the number families care about most, and the original report buried it under dense, hard-to-read charts. My first redesign made the charts cleaner and more legible, and added input tables so families could manage their assets and liabilities in the same place they read them. The latest version, simplified with the product team, goes further: each card is titled with the question it answers. What are we worth? What changed? Where is it? How much could we use right now? A family can read the report top to bottom like a conversation.

Ownership map

Color that makes ownership traceable.

The ownership map shows how a family's people, trusts and companies own one another. The original was hard to understand: the structure broke down and it was difficult to trace an owner (the parent) to what it owns (the child). I redesigned the map from the structure up and added one idea of my own. Every owner card gets its own stroke colour, and the lines flowing down from it carry the same colour. Following an owner to its holdings, or a holding back to its owner, becomes a matter of following a colour, which is exactly the question the map exists to answer.

Impact

A platform families can finally read.

• Every major part of the product redesigned: navigation, home, Kinnect AI, data vault, net worth reports and the ownership map.

• Kinnect AI designed end-to-end and shipped as the conversational layer across the platform.

• The data vault redesign became a foundation for how Kinnect organizes and presents data.

• Onboarding moved to the centre of the home page, giving new families a clear path from sign-up to a complete picture.

• A design system behind all of it, so engineering builds each pattern once (see my Kinnect Design System case study).

Reflection

Clarity is the product.

The families using Kinnect are not short of data. They are short of a picture they can understand and act on. The best decisions in this project came from asking what a family needs to do next, not what the system is able to show. I also learned to keep revisiting work that looks finished: the home page I was proud of in 2025 had to change once I saw where new families were getting stuck. That is the habit I bring to every team now: learn the domain deeply, design against the whole system, and let real use decide when a design is done.