We spoke with Nathan Cheong, Chief Product Officer at online investment platform Havenmark, about the strain of supporting a fast-growing investor base with the wrong tooling — and where the line sits between human judgement and AI-powered speed.
Havenmark gives individuals direct access to private real estate, venture, and credit funds. It serves roughly 1.4 million investors, manages about $2.7 billion in equity capital, and issues more than 900,000 tax documents a year.
After years of what Nathan calls a “shoebox” support stack, the Havenmark team rebuilt the whole thing on Marlowe's AI-first platform — a decision that keeps paying out.
Let's get into it.
What did customer communications look like before Marlowe, and where did they break down?
For years we ran email and phone only — no chat at all — routing every message straight to our Investor Relations team. The questions ranged from “what is Havenmark, exactly?” to password resets, all the way through to underwriting detail on a specific fund.
The complicated ones were worth our time. The repeat ones — two-factor lockouts, where to find a quarterly statement — quietly ate the week.
The real problem was answering everybody quickly while still holding back enough capacity for the questions that deserved a careful reply.
“Every answer to our most common questions was already written down somewhere in the product. People just couldn't find it. Ora was a way of handing that library back to them.”
What did you need to solve those challenges?
We had to pull the low-complexity cases off the Investor Relations queue. Without that, we were choosing between three bad options:
- Get back to people more slowly
- Lower the quality of every reply
- Hire ahead of plan just to keep up
We brought in Marlowe specifically for Ora, its AI agent, to absorb that low-complexity tier. All of the answers already lived in our help center and investor updates — they simply weren't findable.
We ran a short proof of concept, trained Ora on the help center, our quarterly investor letters, and the marketing site, then widened the audience test by test.
Hi there — I'm the beta of Havenmark's AI assistant.
I can answer general questions about:
- Fund strategies and allocations
- Adding or withdrawing funds
- Account settings (auto-invest, dividend reinvestment, etc.)
What can I help you with today?
Watching the success metrics as we widened the rollout is what gave us the confidence to move fast. It also proved the answers coming back were actually right — the goal was a better experience for investors, not just a lighter queue for us.
“Eleven weeks after launching Ora, it was closing more than 60% of our total support volume.”
What results have you seen since implementing Marlowe? Any hard numbers you can share?
The results beat our forecast by a wide margin. Within eleven weeks, Ora was handling more than 60% of total support cases.
Interaction volume always spikes at the start of the year — seasonality plus tax documents. We saw the same spike this time, and Ora fielded nearly all of it, landing us about 47% below the same weeks a year earlier.
Ora has also reshaped the hiring plan for Investor Relations, letting us be deliberate about how and when the team grows instead of hiring because the queue demanded it.
What do you think is the biggest benefit of Marlowe?
First and foremost, it clears the low-complexity cases. Beyond that, it lets people who don't write code shape and improve the experience over time.
Since launch, our Product and Investor Relations teams have taken Ora's answer accuracy from 81% to just past 94% — with no engineering time required at all. That's a genuine unlock.
What advice would you give to others considering Marlowe?
Start with Ora.
