Every GP shop runs on the same patched-together stack: an investor portal, a CRM like HubSpot, a coordinator like Monday.com, a separate underwriting spreadsheet, and a bank login you refresh forty times a day during a raise. Cashflow Portal wants to collapse all of it into one login. Founder Perry Zhang walked us through the entire platform on the podcast this week, so here’s the honest breakdown: what it does, what stood out, where it stops, and who it’s actually for.
First, what we saw and what we didn’t
This was a founder-led demo, not our own hands-on test. Zhang drove the screen and we pushed on the numbers. So treat the impressive moments as shown and credible, worth verifying on your own files, rather than as a verdict we earned by running fifty deals through it. We test underwriting tools constantly at the Collective. Last week we put the same deal through Claude, ChatGPT, and Perplexity and got returns that landed wildly far apart, and that’s the lens we watched this through. A clean demo and a tool you’d stake an LP presentation on are two different things, and we’ll say where the gap is.
What Cashflow Portal actually is
Zhang’s pitch is “Toast for commercial real estate.” Toast started as a restaurant POS and ended up the best lender to restaurants, because it already knew their financials. Cashflow Portal wants to be that connective layer for GPs: the system that owns the investor relationship and the money that moves through it, so it can eventually do more with that data than any point tool could.
The avatar is a real estate private equity firm with at least $10M in equity under management, an established GP with a couple of syndications or a fund or two behind them. It’s not built for brokers or lenders yet. The company says it has 800 paying customers and 150,000 LPs on the platform, with more than $1B in distributions run through its waterfall engine. Zhang is an ex-Amazon, Twitter, and Lyft engineer who syndicated more than 1,000 units himself before building it, and the company went through Y Combinator. (His origin story is every syndicator’s nightmare: 70 investors’ social security numbers sitting in a Google spreadsheet he updated by hand.)
The all-in-one, module by module
The investor portal is the core. LPs get a deal room, invest from their phone, and a questionnaire that pre-populates the legal docs so there’s almost nothing left to sign. Distributions are calculated and sent in the same place. The GP sets a waterfall per share class, the platform computes each LP’s number, and one click sends the ACH from the GP’s bank straight to the LP’s. You can export the math behind any LP’s distribution, which is the kind of thing that defuses a “where’s my money” email fast.
Around that sit the rest: a fund admin layer that reconciles asset-level P&L against the investor cap table and drafts K1s (self-serve, or a concierge plan where their team keeps your books), a CRM with automations tied to investment status (if a wire hasn’t landed in five days, fire a reminder text), a marketing module that drag-and-drops your KPIs into a monthly update and drafts the summary paragraph with AI, and an embedded bank through Core Bank that opens an account in minutes and clears ACH next-day. That bank pays a 1.5% annualized cash reward on raised capital sitting in it. On a $10M raise, the company frames that as roughly $150K a year.
The underwriting is the real hook
Cashflow Portal is both a spreadsheet and an AI tool, and the way it splits those two is the smartest part of the demo. The AI parses the rent roll, the T12, and the OM. The math, meaning waterfalls, distributions, and sensitivity, runs on a deterministic model they built, not on a language model.
Zhang is blunt about why. To him, AI is pattern matching rather than precision. It’s strong at reading documents and surfacing insights, and it’s slightly different every time, which is fine for inspiration and fatal for a number you hand an LP. So they keep the model deterministic and let the AI do the reading.
The payoff shows up in one feature. Click any number in the pro forma and it traces back to the exact line item on the T12 it came from. That citation trail is exactly what we keep telling people to demand from any AI extraction tool: if a number can’t show you its source, you have to re-verify it by hand. Most tools rebuild the model from scratch each run and can’t point back to the source line. This one can. The AI assistant on top lets you ask “what’s the IRR at $11.5M versus $12M” in plain English and watch the model update (Claude and ChatGPT in the backend, depending on the task).
Where it stops
The honest limits matter as much as the demo. It can’t underwrite complex multi-tenant office or retail, anything with shared, pro-rata utility math, and Zhang says so plainly: Argus still wins there. Straight leases only. Moving onto it means moving your investor portal, which he compares to moving houses, so most GPs come in through the underwriting tool first and migrate the rest later. And by design, they keep AI out of double-entry accounting. It can classify expenses, but a person still owns the ledger, because a bad journal entry from a model is nearly impossible to trace back.
Who it’s for
If you’re an established GP tired of stitching a portal to HubSpot to Monday to a spreadsheet, and you care about the LP-side experience, this is worth a serious look. The investor experience was the most differentiated part of the whole platform, and that’s usually the part GPs neglect. If you’re a broker, a lender, underwriting multi-tenant office, or a brand-new sponsor with one deal, it’s not aimed at you yet.
On a founder demo, the integration breadth is real, the underwriting traceability is the standout, and the open questions are price and how that deterministic model holds up on a genuinely messy T12. Zhang’s own closing bet is that technology gets commoditized and relationships win, so the tools worth paying for are the ones that buy back your time instead of eating it in coordination. A single login that underwrites a deal and raises the money for it is at least pointed at that.
We break down a tool like this every week and actually run them on real CRE deals. If you want the running list of what’s worth paying for, that’s what the AI for CRE Collective is built around: 750+ operators testing and sharing what holds up. Join here: https://www.skool.com/ai-for-cre-collective/about
