Preqin~4 min read
The form was never the reason fund managers stopped sharing.
Six options were on the table, and the three easiest to build solved a problem nobody had.
- Role: Product strategy and design
- Timeline: 3 weeks
- Team: 1 PM, 1 engineer, head of engineering, me
- Impact: 5% → 15% retention
- Platform: Web
Impact
Retention
up from 5%
Due diligence
down from 10 months
Both figures held through 2021, when I left Preqin, and each was measured a different way. Retention comes from a two-week test with 20% of traffic diverted to the experiment and the rest left on the existing form, run alongside usability sessions on the same build. Due diligence was not inferred from better coverage: I shadowed partners through a cycle, and the team read deal closure for that same 20% off an internal platform rather than off anyone’s recollection. They are still not equivalent evidence. Retention is a controlled comparison against traffic held back. Diligence is a before-and-after on the experiment cohort with no control arm beside it, so it records that cycles ran shorter for the people who saw this, not that nothing else shortened them. Two weeks is also a short window for a behavior partners repeat quarterly.
The data stopped coming in
Preqin is only as valuable as its data, and fund managers had stopped contributing performance figures. Everyone paying for the platform got less back.
Research showed that collection wasn’t the problem. Partners weren’t withholding numbers because submission was tedious. They were withholding numbers they were embarrassed by, and better upload tooling wouldn’t change that. The work had to move from collection to exchange.
My first experiment, a leaderboard, backfired. The version that worked kept the comparison and removed the identity.
What I owned
I owned
- Research with partners at large and mid-market private equity funds
- Reframing the problem from collection to exchange
- Both experiments: the leaderboard that backfired and the quartile ranking that replaced it
- Keeping the form to what someone could complete in one sitting, despite a push for 32 fields
Decided with others
- Getting Legal and Sales to the same answer: I brokered it; they made the call
- The cohort floor beneath anonymization, with the head of engineering
The problem
The platform depended on data that was going stale.
Fund managers weren’t contributing performance figures, reducing the platform’s value for every subscriber who relied on it. Revenue and renewals were exposed.
The obvious diagnosis was friction. Submission was tedious, so make it easier. Research pointed somewhere else.
That last question changed the brief: what does a fund manager get for sharing a number that might embarrass them?
We considered six options. Three tackled collection, an automated scraper, upload flow, and parser, and would have been easier to engineer. Three tackled exchange: a leaderboard, trust signals, and an inline form. The collection options solved a problem nobody had.
Competition was the obvious lever
I tried a leaderboard. It was wrong.
I considered a public leaderboard, a private benchmark against peers, and no comparison at all. Competition seemed like the obvious lever when people wouldn’t contribute, and I was confident enough to ship a named ranking with prompts to submit benchmark data first. I didn’t think it needed mitigation, although a product asking for trust was about to publish the information partners guarded most closely.
Partners saw exposure rather than motivation: “This doesn’t reflect who we are.” We removed it.
Legal and Sales, both right
Could comparison work without naming names?
Legal wouldn’t accept anything identifiable. Sales wouldn’t accept something a partner couldn’t use for comparison. That ruled out a named ranking, and it ruled out dropping comparison altogether, which left an anonymized quartile position. Both positions were reasonable but incompatible as stated.
We used an anonymized quartile position in an inline field. Legal could accept it because nobody was identifiable, and Sales could sell it because partners still had something to measure against. Anonymity holds only while the peer set is large enough to hide in: slicing by fund type, geography, and vintage can leave four funds in a quartile, allowing everyone in it to work out who is bottom. The feature then stops protecting people at exactly the specificity that makes the data useful. A minimum cohort size, rather than anonymization alone, is the control, and someone must keep deciding where that floor sits as the dataset grows. Both sides said yes, and the timeline went from five sprints to two.
Thirty-two fields, refused
Would anyone finish 32 fields?
Stakeholders had asked for 32 fields. The alternatives were a phased multi-session form or a short inline one. Every field had an internal advocate and a report it supported; the issue was their cumulative cost.
We used a short inline form with trust signals beside it, limited to what someone could finish in one sitting. The data model stayed thinner than the business wanted, and the argument returned every quarter. We would watch what people completed and extend the form from evidence rather than requests. We never tested the 32-field version, so this choice rests on reasoning rather than comparison.
The obvious lever made people feel exposed
Competition was the obvious lever, right up until a product asking for trust made its partners visible. I built the leaderboard first because people wouldn’t contribute, and I was confident it would work. It backfired.
The quartile ranking that worked used the same idea without identity. That fix was available before the leaderboard: research had already said, “if my numbers aren’t great, I don’t want anyone to know.” I shipped it anyway.
The retention figure doesn’t separate recovery from damage, so I don’t know what the leaderboard cost. Did the later result overcome the harm, or just cover it up? The number I have can’t tell me.
