Fundraisly: Designing and testing an AI product for startup founders

Context

Fundraisly is an AI fundraising platform for startup founders. It matches them with investors and, through Warm intros, finds people in their network who can introduce them. Founders can also run outreach to investors on LinkedIn.

Warm intros in the main product: a Team panel where a teammate has connected Google, Microsoft and LinkedIn accounts, and a ready Warm Intro Report with filters for industries, markets, stages and current investors, a field for target funds and buttons to regenerate or open the report. — In Warm intros, the team connects Google, Microsoft and LinkedIn accounts. The report lists investors the team can introduce the founder to. Current investors are blurred.

Choose campaign type, with two options: Expand network, to reach founders, angels and syndicate leads outside the founder’s network, and Warm intro outreach, to ask founders already in the network for introductions to the funds that backed them. — Each LinkedIn campaign starts with its type. Founders grow their network or ask people they know for introductions.

Create campaign, Audience tab: investor types (angel investors, syndicates, family offices), match score (high, medium, low), the startup’s industry and market, and an empty lead list on the right. — Next, founders choose the audience by investor type and match score. The lead list fills in on the right.

I joined in December 2025 as the only product designer in a small team. Before the launch, I designed the product’s interface, the marketing site in Framer, landing pages, the pitch deck and a brand guide. In June 2026 we launched on Product Hunt with me as one of the makers, and Fundraisly became #1 Product of the Day, Week and Month.

The problem

The sales funnel was long: the homepage, an email form, a sales call, and only then a signed contract. So a test that stops at the first click says little about sales.

The team had recorded over a thousand customer calls, but nobody went back to them.

When the team got smaller in September 2026, I also took over the front end of the AI agent product. It still needed a landing page, sign-up, billing and Warm intros.

What I did

A research base from 1,349 customer calls

I connected an AI agent to the team’s call recorder and pulled all 1,349 calls with their transcripts. I turned sales calls into a guide to how customers talk. From calls with paying clients, I built a searchable base of 1,471 complaints, requests and reasons to leave. The recorder’s AI summaries change people’s words, so I checked every quote against its transcript. Evidence from these calls made the Warm intros redesign a priority.

Homepage tests followed to signed contracts

From August to October 2026 I ran four A/B tests on the homepage and followed each one up to signed contracts. In the main test, an investor calculator competed with a product video at the top of the page. Both brought the same share of visitors to the email form, but the video won at every later step.

The homepage on a phone: the Product Hunt “#1 Product of the Month” badge, “$1.1B raised across 200+ startups”, the headline “Get 10–50 qualified investor meetings on your calendar in 90 days”, a “Start my fundraising” button and the top of the demo video. — Variant A, the video version, is still the homepage today.

Variant B: “Check how many investor calls we can generate for you”, fields for a website, industry, market and stage, a table of matching investors and a button “Generate 47 calls”. — Variant B put an investor calculator at the top, filled with investors from a public demo.

Step	A: video	B: calculator
Visitors	1,849	1,866
Reached the email form	164	168
Booked a call	38	25
Had the call	19	10
Signed a contract	3	0

I stopped the test and kept the video. It brought 52% more booked calls and all three contracts. The sample was small (p ≈ 0.09), so I report this as a direction rather than a proven result.

Catching a fake win

In the first two weeks, a bug marked calculator visitors as video visitors. The data showed 22 emails for the video and 3 for the calculator. I fixed the bug and recovered the real numbers from timestamps. In fact, the video had 10 emails and the calculator 13. Now an automated check clicks every button in both variants and confirms the tracking before each test starts.

Shipping the AI agent’s front end in 16 days

In 16 days I shipped four releases of the agent product with Claude Code: the landing page, sign-up, Pro billing and Warm intros. The agent answers three questions for free before it asks people to sign in. ChatGPT and Perplexity work the same way.

Before review

BriefI write what to build and why
BuildClaude Code writes the code
Browser checkI check the change on the page

Before release

Automated testsThey run on every change
Code reviewBy our back-end developer

Every day, an automated test also clicks through the whole product.

Every change goes through these steps before it reaches users.

The agent’s landing page with a question typed into the box: “Seed-stage funds in the US that back AI”, above “3 free questions · no card needed” and four features: Ask anything, Emails & LinkedIn (Pro), Warm intro paths (Pro) and Advisory call (paid separately). — The landing page puts one question box first. Here a founder asks for seed funds in the US that back AI.

The agent’s answer: 3,228 US funds with at least one seed or angel deal in an AI company, the five with the most such deals in a table with websites and deal counts, a note that the free plan shows part of the result, and the top of a card “What this agent can do”. — The answer leads with a short summary and a five-row table. Our back-end developer built the investor matching behind it.

The sign-in window over the landing page: “Sign into Fundraisly”, buttons for Google and LinkedIn, an email field and “Continue with email”. — Sign-up offers Google, LinkedIn or email after the free questions.

Results

The homepage kept the video, and later an AI-made demo video replaced a co-founder’s recording after its own test.
Every homepage test now starts with an automated check of buttons and tracking.
The agent product went live with its landing page, sign-up, Pro billing and Warm intros.
In the first two and a half weeks, 324 people opened the agent. 75 of them asked a question, and 19 signed in.
The landing page now speaks in customers’ own words.

Who did what

Me: design of Warm intros and LinkedIn outreach, the agent product and the marketing site, A/B tests, research, and the agent’s front end since September 2026.
Back-end developer: the investor matching, the API and the data, and reviews of my code changes.
Founders: the CEO set the product priorities and ran the sales calls. A co-founder recorded the first demo video.
Customer success: the calls with paying clients.

What I learned

I follow every test up to the signed contract. The calculator matched the video at the first step and lost at every step after it.
Three in four visitors left the agent without asking anything, and most of them came from Meta ads. Next, I would test a first screen made for visitors from ads.