Preqin~3 min read

Nobody would staff it, so I made the ask smaller.

An insights feed for institutional investors, built on a backend that could not deliver anything in real time.

  • Role: Product management, design, and research
  • Timeline: Zero to one
  • Team: Cross-functional: a PM, data science, and engineering. Product and design were mine.
  • Impact: 90% faster time-to-insight
  • Platform: Web
0% ↓

Time to insight

0% ↑

Weekly active users

0% ↑

Upsell

Read over the two weeks after launch, the same window as the other Preqin work, and still at that level when I left in 2021. The 90% is measured against the weeks investment professionals were spending to surface an insight by hand, and it is an estimate against that manual workflow rather than instrumented task time. Two weeks is also short for a feed, where the question is whether people keep coming back. Still missing: who measured it, and the user base the 20% is 20% of.

Overview

Leadership wanted the insight problem fixed and had concluded no project could fix it, so nothing was staffed.

Investment professionals spent weeks assembling findings the platform already held. The company owned models that could generate hundreds of findings a day. It had no way to deliver them, so subscribers supplied the labor that turned data into answers.

Worked with: a PM, data science, and engineering; 12 asset managers in research.

What I owned

I owned

  • The reframe that got it staffed at all: converting a commitment nobody would make into a bounded question, and scoping that question so either answer was affordable
  • Design end to end
  • Discovery: 12 interviews with asset managers, and the journey map from them
  • The Insights Feed: the template model and the ranking
  • Pressure-testing every concept against what the models could actually deliver

Shared

  • The delivery model, which the backend’s real-time limits decided as much as I did
  • Which findings counted as actionable: the data science side set what was computable, I set what was legible

Research

The problem was delivery, which made it small enough to ask for.

Interviews with 12 asset managers found three things. Volume made findings impossible to prioritize. Metrics sat in separate dashboards, so patterns showed only to someone already looking. Alerts and anomalies were ignored almost entirely.

Read as an insight problem, it needs better models and funding nobody would give. Read as a delivery problem, the models are a given.

Decision: shrink the ask

I asked leadership for permission to answer a question, with no commitment to a result.

There was no evidence to offer, because producing it required the staffing that was refused. The options were a roadmap commitment, dropping the idea, or a time-boxed experiment with a defined question. I chose the experiment.

No evidence changed anyone’s position. The size of the ask changed. The risk is that an under-resourced experiment can produce a false negative, so I kept the question cheap to answer either way.

Decision: one template

Findings arrive pre-composed in one template, because the backend could not compute them live.

I rejected real-time visualizations and a bespoke layout for each insight type. Neither backend limit would change before ship.

One template makes a feed scannable, and it also makes findings look alike. A daily templated insight can stop registering as fast as a good one does.

Explorations of the insights feed, testing density and ranking.
Testing how much a single insight can carry before it stops being scannable.
A flow headed “Crafting an accelerated user flow”: a Configure panel of metrics and attributes, annotated as informed by the customer’s knowledge model, feeds a Generate Opportunities step into an Explore Insights list of recommended insights, which feeds a Generate Dashboard step into an Insight details view holding a metric-over-time chart, a breakdown table, and a process explorer.
Configure once, against the customer’s own knowledge model. Everything downstream is generated. The 90% improvement is an estimate based on research into the previous manual workflow, not instrumentation of these steps.

Redesign

The feed ranks by actionability and shows its reasoning on each card.

I rejected a chronological feed, which would repeat the volume problem, and user-built filters, which assume people know what they are looking for. Research said they did not.

Each insight unfolds from one metric into the narrative behind it, so a rank can be checked. A rank users can check is a rank the team can retune without re-earning trust.

A focused insight view, showing one metric unfolding into a narrative.
One metric, unfolded into the narrative behind it. This is what makes a rank checkable.
The empty state before any insights have been created.
The empty state. The first thing a new user meets, and the case where ranking has nothing to rank.

Scope

The insights feed shipped, with no new tool to learn.

Shipped: templated findings ranked by actionability, fitting existing workflows. Because findings arrive in slots, a new finding type is a data change rather than an interface project.

Results

Time to insight fell 90%, weekly actives rose 20%, and upsell rose 8%.

An idea leadership had written off shipped and produced numbers.

I left before ranking decay could show up, so I cannot say how long the 90% held

The feed may have worked partly because it was new. Evidence that it works at launch is the easy half.

Shrinking the ask only works when the question is cheap to answer either way. I do not know whether anyone else contested the templating tradeoff.

Sam Cusano