Accelerators
Every batch should start where the last one finished.
Apex gives each cohort company the tools to measure its own growth, run a test at the size it actually is, and find out whether the advice worked. Your program gets a permissioned batch view, so mentoring is aimed at the company that needs it and every result the batch produces is still there for the next one.
No credit card to start.Northline Labs is an invented program. The companies named below are familiar examples, not Apex customers, and every figure is made up for this page.
See the batch
Know which company is stuck in week three, not at the demo day rehearsal.
Revenue, retention, customer growth, acquisition cost, and product use land on one screen for every company in the cohort, computed from each company's own events on the same definitions. Office hours go to the company whose numbers stopped moving, rather than to the founder who is best at asking for them.
StripeSpaceXPerplexityAdded up across 4 companies that are reporting
Across every workspace you were given
Weighted by revenue, not an average of rates
Added up across workspaces
Run a test at this size
A company with two hundred users can still learn something true.
Most growth tools will draw a confident line through eleven data points and let a founder build a seed deck on it. Apex will not. It states how many people a result actually covers, it reports that two groups cannot be told apart rather than calling a rounding error a win, and it renders a dash where it cannot stand behind a number. That restraint is the only thing that makes a batch benchmark worth reading.
Reference points
Renewed at full seat count across each arm and the baseline before this experiment. Workspace holdout and industry benchmark live in the portfolio view.
Previous baseline
Before this experiment
Renewal reminder
control
Offer the review
variant
Compound the batch
Mentor advice stops being a war story and starts being a claim with a condition on it.
When a company tests something, the outcome is kept with the conditions that produced it. The next cohort does not inherit the tip. It inherits the tip, the company it worked at, the company it did nothing for, and the difference between them. Companies keep control of their own data, and the program sees only what each one approved.
Reference points
Renewed at full seat count across each arm and the baseline before this experiment. Workspace holdout and industry benchmark live in the portfolio view.
Previous baseline
Before this experiment
Renewal reminder
control
Offer the review
variant
73 out of every 100 accounts renewed with every seat, against 58 on the plain reminder, and 56 before the test existed. Both comparisons point the same way, on 178 accounts.
Reference points
Renewed at full seat count across each arm and the baseline before this experiment. Workspace holdout and industry benchmark live in the portfolio view.
Previous baseline
Before this experiment
Renewal reminder
control
Offer the review
variant
64 against 65, on a starting rate of 64. The two groups can't be told apart, and Apex says so instead of dressing a rounding error up as a small win.
One batch, end to end
Follow one cohort from the weekly update form to a lesson the next batch starts with.
Northline Labs is an invented program running a batch of 5 companies. The company names are familiar, but none of them is an Apex customer and every number is made up. Every product screen below is real Apex. In six steps, follow the batch from what founders report today to a result the program can hand the next cohort.
Every number on this page is made up for this walkthrough. These are familiar companies used as examples, not Apex customers, and none of these figures are their real results.
01 · What founders send today
Every founder reports honestly, and no two of them report the same thing
This is the weekly update form. Nobody is being evasive. One founder counted seats because that is what their billing page shows, one counted money but only above a threshold, and one skipped the question because their product does not have the concept yet. The last response arrived after the batch meeting it was for.
Those figures are invented, and so is every other number on this page. The company names are familiar because you already know what each one sells. None of them is an Apex customer.
The cost of this is not the chasing. It is that the batch cannot be compared to itself, so a mentor has no way to say whether a company is behind or ahead, and every piece of advice has to start from the founder's own account of how things are going.
| Company | You asked for | What they sent | When |
|---|---|---|---|
| Anthropic | Net revenue retention | “Retention: 94%”, which counted seats, not money | day 19 |
| Databricks | Net revenue retention | “NDR 1.02x”, money, but only for accounts over $50k | day 26 |
| Stripe | Net revenue retention | left blank, “we measure churn instead” | day 31 |
| Anthropic | Customers | 412, counting each department on its own | day 19 |
| Databricks | Customers | 1,940, counting every account ever registered | day 26 |
Each company points Apex at its own codebase one time, and from then on a number is never translated on its way to the program.
Settled once, in each company's code02 · The account is theirs
The company grants access for the program, and keeps the account after demo day
A founder joining a program signs a lot of things. This is not one of them. The company creates its own Apex account, on its own plan, and separately grants the program a read-only view of the totals and trends it chooses to share.
That ordering matters more for an accelerator than for any other kind of portfolio, because the relationship ends on a known date. When the batch graduates, nothing is migrated and nothing is lost. The company keeps its events, its definitions, its experiment history, and its account. The program keeps whatever the company left switched on, and the founder can switch it off on the walk home from demo day.
The founder chooses the scope, and the same button that grants it ends it.
Read only, and revocable by the founder03 · The whole batch, overnight
One screen, one definition, and one company that is not pretending to have numbers
This is what replaces the form. Every company in the batch, added up the same way, refreshed overnight, without anybody being asked.
Look at the last card. It has no numbers, and instead of a zero it says why. That distinction is the difference between a batch leaderboard that helps and one that quietly punishes the company that onboarded last. An empty result and a bad result look identical on every dashboard that does not bother to separate them, and a founder on the wrong end of that spends a week defending a number that was never real.
StripeSpaceXPerplexityAdded up across 4 companies that are reporting
Across every workspace you were given
Weighted by revenue, not an average of rates
Added up across workspaces
The company at the bottom approved two days ago and says so, rather than reporting a zero that would read as a fact about the business.
No update form04 · Compared to the batch
The right benchmark is the eleven companies sitting in the same room
Published startup benchmarks describe companies at a stage the batch has not reached, measured by people who did not say how, and averaged across businesses that make money in completely different ways. A founder who lands below one of those is told very little.
The batch is a better reference and it costs nothing to assemble. Same program, same weeks, same amount of help, all measured the same way. When Anthropic is losing customers faster than Databricks is, that gap is real information about the business rather than about the market, and it is specific enough for a mentor to work from on Tuesday.
Same companies, one line each, sorted by the metric under discussion, every figure produced one way.
Your cohort is the benchmark05 · Small numbers, honest answers
The test covers a hundred and seventy-eight accounts, and the page says so
A mentor suggests a change. Rather than being delivered as advice and reported back as done, it ships as a test: two versions, real customers, a fair split, and a group left on the old behaviour so there is something honest to measure against.
The part that matters at this stage is what the screen refuses to do. It reports how many accounts the result actually covers. It reads the outcome against the rate the company was already achieving, and against a group that received nothing at all, because a company growing quickly will improve on almost any week you pick. When two groups cannot be told apart, it says that, rather than promoting a rounding error into a slide.
An early-stage company can survive being told a test was inconclusive. What it cannot survive is raising on a number that was never there, and then building the next eighteen months around it.
Reference points
Renewed at full seat count across each arm and the baseline before this experiment. Workspace holdout and industry benchmark live in the portfolio view.
Previous baseline
Before this experiment
Renewal reminder
control
Offer the review
variant
Anything Apex cannot stand behind at this sample size renders as a dash, and it names which comparison each figure came from.
A dash rather than a guess06 · What the next cohort inherits
It worked at one company and did nothing at the other, and that pair is the asset
The same change ran at two companies in the batch. It moved renewals at one and did nothing at the other, where customers pay for usage and there are no seats to keep.
A program that records only its wins produces a playbook that gets less true every cohort, because the conditions that made a play work are exactly the part nobody writes down. What the next batch should inherit is not offer a review before renewal. It is offer a review before renewal when the customer pays per seat, with both companies and both outcomes still attached. Six months from now a founder can read that, decide it applies to them, and go and check.
Reference points
Renewed at full seat count across each arm and the baseline before this experiment. Workspace holdout and industry benchmark live in the portfolio view.
Previous baseline
Before this experiment
Renewal reminder
control
Offer the review
variant
73 out of every 100 accounts renewed with every seat, against 58 on the plain reminder, and 56 before the test existed. Both comparisons point the same way, on 178 accounts.
Reference points
Renewed at full seat count across each arm and the baseline before this experiment. Workspace holdout and industry benchmark live in the portfolio view.
Previous baseline
Before this experiment
Renewal reminder
control
Offer the review
variant
64 against 65, on a starting rate of 64. The two groups can't be told apart, and Apex says so instead of dressing a rounding error up as a small win.
When a team pays per seat, a drop in weekly use predicts lost seats at renewal better than support tickets do.
last moved by the first renewal test
Offering a quiet account a review keeps seats, when the customer pays per seat.
last moved by the second renewal test
The same offer does nothing when the customer pays for what they use instead of per seat.
last moved by the second renewal test
Paid social brings in teams that don't stay long enough to pay back what they cost.
last moved by the channel review
The flat result only means something because the other one exists. Kept together, the two become a claim the next batch can test.
A claim with a conditionWhat the program keeps
The batch graduated. What it learned did not leave with it.
The companies keep their accounts and their data. The program keeps a growing set of claims, each one carrying the company it worked at, the company it did not, and the conditions that separated them. The list below is also what Apex will not tell you, because a program that oversells its own measurement is the one thing worse than not measuring.
What the product reportsApex measures this
Revenue, retention, first value, use, and what each channel costs and returns. All of it computed from the events each company's own code sends.
Whether a change workedApex measures this
Because the change runs as a test with a control group, the rate before it started, and a holdout. It isn't claimed after the fact.
The company's booksApex does not
Bookings, deferred revenue, cost of goods, headcount, cash. Apex never sees the accounting system and never checks its numbers against it.
DiligenceApex does not
Contracts, cap table, customer references, quality of earnings. Nothing here replaces the work you do before you wire money.
- definition of a metric, shared by the whole batch
- 1definition of a metric, shared by the whole batch
- steps in the example below
- 6steps in the example below
- reference points behind every result
- 4reference points behind every result
- accounts the program takes with it at graduation
- 0accounts the program takes with it at graduation
Your move
Make the next batch start ahead of this one.
Measure every company the same way. Let each one test at the size it actually is. Keep the result with the condition that produced it, and hand the pair to the cohort behind them.