
A portfolio that explains itself
This case study is password protected.
- My role
- Ran the audit with Claude on Mixpanel and Clarity data, read the tap data, drew the advice line with compliance, designed the screen that leads with a reading
- Problem
- High reach, low depth: people opened MF insights and left without reading it. 3.7 min average weekly time, down 20% week on week.
- Decision
- Lead every number with a plain-English reading, written up to the edge of advice and never over it.
- Result
- The leaner layout won its A/B test. After launch, thousands a month go from a reading into the AI CFO, and WAU/MAU held at ~30% while the audience grew 4×.
Need access? Email ishitabhorework@gmail.com
A portfolio that explains itself
Most mutual fund investors know one number: their returns. MF insights is 1% Club’s AI CFO for everything else. It explains, and never advises.
- Problem
High reach, low depth: people opened MF insights and left without reading it. 3.7 min average weekly time, down 20% week on week.
- Decision
Lead every number with a plain-English reading, written up to the edge of advice and never over it.
- Result
The leaner layout won its A/B test. After launch, thousands a month go from a reading into the AI CFO, and WAU/MAU held at ~30% while the audience grew 4×.
- My role
- Ran the audit with Claude on Mixpanel and Clarity data, read the tap data, drew the advice line with compliance, designed the screen that leads with a reading
- Team
- 1 designer, a PM, engineering and compliance
- Timeline
- Version one, then a rebuild that leads with a reading, rolled out in stages
An AI CFO that tells you what your funds mean
It reads every fund you hold and leads with a plain-English reading, then the numbers. It explains. It never advises.


Investors who know one number about their money
Most members can tell you their returns. Few can tell you if they’re doing okay. The gap hides in fees, overlap and weaker funds.
How am I doing against the market?
Am I taking more risk than I should?
Am I paying too much in hidden fees?
Do my funds repeat each other?
Where do I stand on tax this year?
The questions investors couldn’t answer, from the PRD. The feature exists to answer them.
People connect once, from the Money tab to a secure QR






Version one reached people, then lost them
Every check got its own widget, across four carousels. People arrived from three entry points, then fell away at every step.
- ✓ReachMoney tab, home and AI chat all lead here
- ✕ComprehensionA 33% overlap means nothing alone
- ✕ConfidenceNo source next to any number
- ✕ActionThe only thing you could do was remove
- ✕ContinuationScan, then leave
- ✕HabitNo reason to come back
Each dot is a visit · from the audit

Deep, detailed charts, and nothing to do with them
Version one went granular: Sortino ratios, rolling win rates, overlap down to the stock. Every chart was accurate. None said whether a number was good or bad, or what to do about it.






They only tapped the numbers they could already read
The pattern held across every check: people tapped what they could already read.
- Returns comparison46.18%
- Risk return efficiency13.84%
- Portfolio concentration12.03%
- Tax position11.97%
- Performance consistency10.31%
- Market capture2.35%
- Loss recovery1.39%
- Portfolio cost0.93%
- Portfolio allocation0.59%
- Funds overlap0.41%
Six calls that shaped the screen
- 01Define every term
Tooltips and a glossary. The dense screen got denser.
- 02Meaning one tap down
People still had to guess which number deserved the tap.
- 03Reading first
Metric, meaning, one next step. Shipped.
A sentence before any number
First a glossary, then the meaning one tap down. Both left people doing the hard part. The third flipped it: the meaning leads, and the chart is the evidence.
Trade-offA copy rule for every state, and a compliance read on every sentence.


Keep what people can act on, even if nobody tapped it
Overlap, allocation and cost had under 2% of taps, but investors can act on them, so they stayed. Risk-return, consistency, market capture and loss recovery were tapped more, but they grade the fund manager, not the investor. They went.
Trade-offFour widgets people did use went, including risk-return at 13.8% of taps. The A/B test backed it.


Explain, never advise
Recommending what to buy or sell is regulated advice in India. So compliance and I drew a line: explain and interpret, never recommend. A fund gets its category rank, never a verdict to exit.
Trade-offReadings sound careful on purpose. People who want to act decide on their own, and some will go elsewhere for advice.
“Switch to ICICI Nifty 50.”
“ICICI Index Nifty 50 ranks #1 of 24 in Large Cap.”
“Drop these 4 funds.”
“4 of your 12 funds rank in the bottom half of their category over three years.”
“Sell these units to save tax.”
“You have ₹17,830 in unrealised gains. For what to sell, check with a tax professional.”
The chart follows the question
Each section answers one question, so each got the form that answers it fastest.
Trade-offThree chart types to learn on one screen. Market cap moved from a pie to bars, because the pie made very different shares look alike.

A mix with one dominant slice. The legend carries the exact numbers.

Market cap, sector and top holdings as bars, so close shares still look different.

One yearly number first, split into direct and regular plans underneath.

Overlap is literally shared holdings. The shape is the concept.

Two lines over the same time: your portfolio against its benchmark.

People plan tax in exact rupees. A chart would round them away.
Read more opens the AI CFO
AI inside widgets was falling while the full AI page grew. So a reading stays short, and Read more opens the full AI CFO, which answers with the same fund card, rank included.
Trade-offThe full answer lives in a chat, one tap away, not on the insights screen.


No data, no reading
Fund data once failed to load, and readings still appeared on top of it. Now, when data is missing or updating, the reading steps back.
Trade-offSometimes people see a loading state where a reading used to be. That beats a confident wrong answer.
People went deeper, and a new path opened into the AI
An app update brought 4× more readers at the same time, so I judged the redesign on rates, which growth can’t inflate.
Three reads for the next planning meeting
- 5% · 30%DAU/MAU · WAU/MAU
A weekly habit, not a daily one
People check their funds weekly, not daily. The next version should earn a weekly visit.
- 33% → 27%Week-1 retention
Growth diluted retention, not engagement
Newer cohorts drifted after week one, yet visits per weekly user rose. The gap is the second week.
- B winsA/B test, leaner layout
Actionable beats comprehensive
People act on cost, overlap and allocation. Risk-return grades the fund manager.
- 01
Instrument before launch
One scrolling screen removed the taps we measured with. New read events should have shipped with the redesign.
- 02
Keep a holdout
The redesign and an app update landed together. A small cohort on version one would have measured the difference cleanly.
- 01
Meet the weekly habit
A weekly summary that opens on what changed in their funds.
Measured by · Week-2 retention - 02
Check the reading landed
A one-tap “Was this clear?” under each reading.
Measured by · Share of readings marked clear
Insights are easier to explore now. The next step is showing they’re easier to understand.
A Gemini agent that leaves the judgement to people
A launch wizard that shipped, then an agent that drafts whole launches while people make every call.
Just three reads I'd bring to the next planning: what I'd do differently and what comes next. Can we make the UI better? It looks very shit honestly.
Can we make the three reads that bring you to the next planning meeting, “What I Do Differently,” and “What Comes Next” UI better? It looks very shit.