A great brain in a CRM's body.
Credit first: v1 proved the hard parts. One canonical data model keeps every number consistent everywhere — the 4.2% delinquency, the $820K Atlas Freight deal, the 9-day launch buffer appear identically in cards, tables, and chat. Every assistant answer ships source chips. The killer scenes exist and land. Nothing below argues with that content. All of it argues with where the content lives.
4 nav sections + 5 company pages + a sidebar assistant = a CRM in the making. Roughly 12 interactive targets in the chrome before any content renders. The assistant — the only surface that answers cross-cutting questions in one step — gets a 320px rail, visually the twelfth thing on screen, and collapses to a floating button on a laptop.
Seven tabs per company, five companies: 35 views. "Is Northlane on plan?" lives behind tab four, Alignment — if you already know that. Nothing in the UI tells you. At 20 portfolio companies this becomes 140 tab-views.
A destination page for your own warnings. A crossed covenant threshold and a competitor's tent-price change get the same row, the same type size, the same Acknowledge button. The user visits their alerts; the alerts should visit the user.
Navigation-first: you must already know where to look
~40 distinct views, 80+ widgets, ~25 dead-but-live-looking buttons — around 43 honest places an answer could live. Every scripted demo question is one click in the assistant and 2+ clicks plus a scan through navigation. The IA taxes exactly the questions the product exists to answer: the cross-cutting ones.
The assistant is an accessory bolted onto dashboards
~1,100 lines of dashboard components against 149 lines of assistant panel. Voice is a 1.7-second timer that fakes listening. Citation chips open nothing. The single most on-thesis button — Dealflow's "Ask across the vault" — has no onClick. The layout says: dashboard product, assistant attached. The thesis demands the reverse.
Everything is equally loud
The Northlane covenant breach — the most consequential fact in the portfolio — renders six times across the app, and not one placement ranks it above anything else on its screen. The interface presents; the user triages. Attention ranking — the product's actual job — happens entirely in the partner's head.
Proactivity is trapped in one widget
"Since Friday, updated 8:02 AM" — the demo's best idea — is one card among ten, competing for the same glance as an opex chart. "Read briefing," "Open full briefing," "Voice readout: Ready," "Preview tomorrow's briefing" — all dead. The brief should be the home, not widget number seven on it.
Every feature costs a permanent screen
Dealflow became a sixth nav area, alerts became a page, tasks became a kanban plus an Operations page, competitor intel became a tab. Add LP reporting, fund admin, 20 companies, 18 integrations — the nav scrolls, the tabs wrap, and ⌘K becomes survival gear. The fix is a pricing change: a new capability must cost near-zero screens.
One fixed layout for three incompatible jobs
The analyst needs depth on demand. The partner needs five ranked bullets before IC. The portfolio-company CEO — the persona in the owner's brief who installs the product and feeds data upward — has no surface at all. All get the same nine nav items. Only a surface that composes itself to the person and the moment can serve them.