We publish what would prove us wrong.
Arvyn is building a revenue system around one question — which twenty companies should we contact today. Below is what we believe, what we refuse to build, and the specific observation that would end each bet.
- Stage
- Pre-revenue
- Customers
- None yet — see below
- Open bets
- 6
Proofboth columns are ours
The column on the left is empty, and we are printing it anyway.
A company whose whole argument is that claims must carry sources cannot open with a wall of logos it does not have. So here is the borrowed proof we have — none — beside the things we have actually published and you can check today.
Borrowed proof
- Customer logos
- 0
- Testimonials
- 0
- Case studies
- 0
- Conversion statistics
- 0
- Certifications held
- 0
Every figure in this column stays at zero until a customer puts their own name to something. We will not borrow one in the meantime.
Ours, and checkable
- Falsifiable bets, published
- 6
- Security invariants, published
- 7
- Pipeline stages, documented
- 11
- Refusals we will not trade away
- 3
- Sources required per claim
- 1
each with the observation that would end it
named mechanism per rule, on /security
what each one emits, on /how-it-works
no invented reasons, no manufactured volume, no borrowed proof
a URL and a date, or the claim is dropped
Both columns are counts of our own artefacts. Neither contains a customer result, because there are no customers to have produced one.
The biggest problem isn't volume. The biggest problem is targeting.
Outbound got faster every year and got worse at the same time. More inboxes, more automation, more volume — all of it engineering poured into the one part of the problem that was never the constraint.
Thousands of companies receive a cold email every day that is irrelevant for a single reason: they are not currently buying anything. The tooling around that email is excellent. The decision to send it was never made.
So we build the decision. Traditional outbound software provides tools. The Revenue System provides decisions — and every decision it makes has to show its work.
Positioning
It is easier to say what this is by saying what it isn't.
Not because those categories are bad, but because being mistaken for one of them means the actual product never gets explained.
We are not
- Email software
- Lead generation software
- A CRM
- Marketing automation
- AI copywriting software
- Another Instantly, Apollo, Smartlead or Lemlist
We are
An AI revenue operating system.
A system that continuously discovers companies with a real probability of buying, researches them properly, works out why they should buy, writes the outreach, launches it, reads what came back, and improves — without asking the customer to operate every step by hand.
Email is one execution channel underneath that, not the product. Others can be added without changing anything above them.
Six bets, and each one has a test that would kill it.
The product is not a feature list. It is a machine for testing six falsifiable claims, and each claim is encoded in a specific component. If a bet dies, the component that encodes it dies with it — that is the design working, not the design failing.
These are our own tests, run against our own system. They are not customer results, and nothing on this page is one.
B1
A buying window is visible from the outside.
A company that is about to spend money leaves traces in public before it talks to anyone: it hires, it migrates, it starts or stops advertising, it changes its own pages, it files something. If that is true, no proprietary database is required to find the companies worth contacting — only the discipline to read what is already published and to store where it came from.
Falsified ifAfter four weeks of harvesting a market, fewer than a third of it carries any citable evidence at all.
B2
Timing beats fit.
Everyone in this category sells fit — the right industry, the right size, the right stack. We are betting that a message citing a dated, company-specific trigger outperforms a perfectly-fitted message with nothing happening behind it, and that fit alone should never be enough to earn a send. That bet is why the qualification gate refuses companies that match beautifully but show no event.
Falsified ifTrigger-backed sends do not outperform fit-only sends across several hundred sends.
B3
Evidence discipline wins.
Every score and every sentence has to cite a source that a person can open. Not because it sounds rigorous, but because the alternative is a system that generates plausible reasons — and the graveyard of AI sales tools is full of products that generated plausible reasons and lost their customers within a quarter. Citations are enforced in code, and a claim whose evidence does not survive review is deleted rather than softened.
Falsified ifA weekly twenty-row human audit finds more than one in ten evidence rows false, three weeks running.
B4
Independent signals stack; identical ones do not.
Two signals from genuinely separate sources should predict better than one signal counted three ways. Most scoring systems quietly do the second thing — a single hiring event detected by three different collectors becomes three points — and the score inflates while the accuracy stays where it was. Our gate caps per category and requires independence before it will call a window open.
Falsified ifCompanies backed by multiple independent signals do not out-reply companies backed by one.
B5
A system this small can still learn.
Twenty decisions a day is not enough data to train a per-customer model on, and pretending otherwise is how products in this space end up with a learning loop that is really a random number generator. So learning happens globally first, with per-customer adjustment switched on only after there are enough outcomes to mean something — and three of the twenty daily slots are permanently reserved for companies the ranking is least sure about.
Falsified ifThe confident slots stop out-performing the deliberately uncertain ones once there are enough sends to tell.
B6
Scarcity reads as rigor.
We show a short day rather than a full one. Nine of twenty slots cleared the evidence bar today is a more honest sentence than twenty companies, and we believe it builds more trust than padding the list would. A hard quota is the root cause of every quality-degradation mode we could find in this category, so the daily number is elastic by design.
Falsified ifCustomers leave specifically over short days, while the replies from those short days stay good.
Operating rules
Four rules that decide arguments internally.
Not values on a wall. These are the four sentences that settle a disagreement about what to build, and each of them costs us something specific.
The system never invents a reason
Every recommendation references observable evidence. If a claim loses the evidence it cited, the claim is removed and the score is recomputed without it. There is no fallback to a plausible sentence.
We do not manufacture volume
The strong tier is capped at what a market honestly yields and the daily list is elastic. A quota would force exactly the relabelling this product exists to replace.
We do not borrow proof
No customer logos we do not have, no testimonials we did not receive, no metrics we did not measure. A company that asks you to trust its citations cannot fake its own.
The uncomfortable number is shown
Short days, capped tiers, per-tier reply rates including the bad ones. If the tiering stops working, the customer sees it in the same report we do.
What we refuse to count, and what we count instead.
Every number in the left column can be increased by trying harder at the thing that was never the constraint. None of them has ever closed a deal.
Not this
- Emails sent
- Leads exported
- Contacts stored
- Inboxes connected
This
- Qualified conversations created
- Meetings booked
- Pipeline generated
- Revenue influenced
- Time given back to the customer
The customer should not feel like they are operating software. They should feel like they hired an outbound team.
We are early, and we are not going to hide it.
What we will tell you is what exists, what is being built, and which of our own bets is currently under test. When a bet fails we will say so here, with the date.
- NexusThe console, campaign intelligence and the sending engine, running in production.
- ScoutResearch intelligence: cited dossiers in case-file format.
- SonarMeeting intelligence: the practice room and live call coaching.
- QualificationThe evidence harvest, the gate and the deliberation — in build, and being run against its own falsification tests before it is run against anyone's market.
Talk to the people building it.
At this size, the person who answers is the person who wrote the thing you are asking about.
General
hello@arvyn.ioQuestions about the product, pricing, or whether your market is one the engine can read.
Security
security@arvyn.ioVulnerability reports and data-handling questions. See the security page first.
Faster than email
Connect a website in the console. The compiler sizes your market and shows you the evidence it can find before anybody has to get on a call about it.
The whole argument, in nine stages.
If you want to judge whether any of this is real, the fastest route is the page that shows the machine deciding — including the parts where it argues with itself.