Help every portfolio company grow without a giant operating team.

You send a founder a request, and this is the screen they receive. They choose which workspaces you can read, and whether the breakdown names each one or leaves the names off. Then they approve, or they decline. The choice keeps belonging to them.

Reading across company lines without permission is not a gap in process, it is a breach. So the limits sit on the grant itself. The access reads and never writes, it covers this company and nobody else, and individual customer records stay out of it. Everything the firm will see is written above the buttons before anyone agrees, including how to end it. The founder keeps the same view you have, with a button that stops it.

Somebody will hand you an industry average. It describes companies you have never seen, measured by people who did not say how, and it dies at the first follow-up question, which is always the same one. Compared to what?

Your own portfolio answers that better. Same companies, one line each, sorted by the metric under discussion, every figure produced one way. These are businesses you selected, at stages you understand. Read across the row and you already have what the company earns, how fast that is moving, how many people use it, and whether a customer is worth more than they cost.

Anthropic has the largest recurring revenue on the list and the weakest retention of the two companies that report it. Databricks is a smaller business holding one hundred and twelve percent, so this is not the market and it is not the stage. That combination is what makes Anthropic worth a month of an operating partner's time instead of a line in a memo.

A single month is an anecdote and everybody in the room knows it. The useful fact about Anthropic is not today's eighty-eight percent. Retention was one hundred and two percent in February and has slipped every month since, which turns a rough quarter into something that changed in the spring and went uncaught. That is a different board conversation, and a much shorter one.

Some of these numbers are levels and some are totals, so the tiles chart them differently on purpose. Recurring revenue and customer count are read at the end of each week. Revenue and new customers are added up across it. Reverse those and you produce a chart that looks fine, is wrong, and survives three board meetings before anyone notices.

A dashboard that stops at the table hands the thinking back to the company, and the thinking is the exact work your operating team was hoping to contribute. These three cards say what to do and show the arithmetic that produced them. Put more behind Google. Cut Meta back. Repair the LinkedIn connection before trusting another number from it.

Disagreeing takes about ten seconds, because the reason is printed on the card instead of living in somebody's head. That is the difference between a Monday meeting that ends in three decisions and one that ends in a request for more analysis.

A new message that beats the old message is not enough. Both groups could be sitting below where the company was six months ago, and the lift chart would look excellent the entire time.

So every result is read against four things. The control is the old way. The baseline is the rate Anthropic was already achieving before anybody touched anything. The holdout is a slice of accounts left alone on purpose, so there is always a group that received nothing. The portfolio benchmark is the other companies you hold, and only the ones that opted their numbers in when they approved.

The holdout is the one that protects you. A gain that appears against the control and vanishes against the holdout is a message that moved people who were going to renew anyway. That is the difference between a real $71,400 and a number you have to walk back. Apex also states where each comparison lives rather than pooling them into one confident figure. Anything it cannot stand behind renders as a dash instead of a guess.

Ridgeway Group is an invented customer with twelve seats across four teams. Two of those teams stopped opening the product in June, which is the exact condition the journey watches for. The timeline is every event in order, from the Google search that brought them in to the day they renewed with all twelve seats. Below it sits every experiment this account was ever in, which group it landed in, and the message that group actually sent.

One of those rows is still open. The account is in a second experiment whose window has not closed, so the row shows the date it will be judged on and states that the goal event has not fired yet. An assignment is recorded on the account the day it is made, which is how you know the measurement was set up before the result rather than assembled after it. It is also the fastest answer to the question every operator eventually asks. Did any of this reach a real customer?

Anthropic crosses back above one hundred percent. Net revenue retention is one hundred and three percent, the share of customers lost is half what it was, and recurring revenue is up $320,000 without one extra dollar of ad spend. The renewal experiment explains $71,400 of that, the seats it kept from being dropped. Everything else came from new customers and from expansions the experiment had nothing to do with. Claiming the whole $320,000 is the exact move this page argues against.

The amber flag is gone for a reason somebody can name, on a date somebody can point at, with the experiment that caused it still attached to the card. The portfolio figure moved with it, from ninety-seven percent to one hundred and six. Anthropic is about three-fifths of the revenue that figure is computed over, which is the other argument for weighting instead of averaging. An average would have shuffled two points and told you nothing about where the money came from.