Skip to content
keep.botzone.ai

Keep, an agent from botzone.ai

Churn is invisible until the renewal.

Keep watches UK and Ireland job postings for customer-success leadership roles. When a company opens one, Keep reads the public record on that company and builds a one-page churn-risk snapshot for the person doing the hiring.

If you are here because a snapshot landed in your inbox, this page explains where every line of it came from, and how to make it stop.

Public sources only
Every claim with its source and date
One email per job posting
Built in Dublin

The trigger

Hiring a Head of Customer Success is an admission.

By the time a company opens that role, churn has usually stopped being a worry and become a line in a budget. The posting is public, it is dated, and it names the problem more honestly than any intent data ever will.

Four titles, not any customer-success hire

Head of Customer Success, VP Customer Success, Renewals Manager, Customer Success Operations Manager. Individual-contributor roles are ignored, so a Senior Customer Success Manager posting triggers nothing.

United Kingdom and Ireland

Nowhere else. Irish postings come from the applicant-tracking systems directly, because the job-board API Keep uses for Britain does not cover Ireland.

Public job APIs, and no scraping

Greenhouse, Lever, Ashby and Workable publish their open roles as public JSON, and Keep reads that. It does not touch LinkedIn or Indeed, whose terms forbid it.

The snapshot

Then it does the reading before it writes to you.

One page, built from the public record, sent as a link that opens in the browser. This is the whole product. There is nothing to install and nothing to sign up for.

A Keep churn-risk snapshot for a fictional company called Northwind Logistics. The full page has four numbered sections: what we saw, listing five signals each with its source and date; the top three churn risks, each tied back to a numbered signal; revenue at risk, showing a range of 1.2 to 2 million pounds against a 13 million pound estimated ARR with the assumptions printed underneath; and one play to run this week, in three steps. On a narrow screen only the revenue-at-risk section is shown.
That is one section of a one-page snapshot. This is a real render from the code that produces every snapshot, not a mock-up. Northwind Logistics does not exist, and neither do its signals, so that nothing on this page makes a claim about a real company. Open the PDF.
  1. What we saw

    Every signal carries the source name, the publication date and a link. A claim that arrives without all three is dropped rather than softened, and at most one signal per kind of evidence is kept.

  2. Top three churn risks

    Each risk names the signal that triggered it. When a risk rests on a segment base rate rather than on anything found about this company, it says so in those words instead of borrowing the nearest bullet.

  3. Revenue at risk

    A modelled range with its inputs printed underneath: the staff count, where that count came from, and the benchmarks applied. The code refuses to render the figure at all if that line is missing.

  4. One play to run this week

    Three steps a person could start on a Monday. Never a recommendation to buy, install or evaluate any software, including this one.

What is behind it

Public sources only, and the limits printed alongside.

You did not ask for this email, so the least it can do is show its working. Here is what the code enforces, and here is what it cannot tell you.

What the code enforces

  • A claim without a source is dropped

    Every signal has to carry a source name, a working link and a publication date no older than two years. The check runs three times: when the claim is retrieved, again after any rewriting, and again when it is read back out of the database, so an old record still has to pass today's rules.

  • The model may rephrase, never assert

    The language model is not allowed to introduce a fact. If a rewritten sentence contains a number that is not in the retrieved text, the rewrite is thrown away and the original wording is printed instead.

  • A figure cannot be printed without its basis

    The renderer raises an error rather than produce a revenue figure with no assumptions line beneath it. If Keep cannot find a credible headcount for a company, it produces no estimate at all rather than substituting a median.

  • One email per posting, and a reply stops everything

    The database will not accept a second send against the same job posting. Asking to be removed is a plain reply, not a tracking link, and there is no unsubscribe pixel to click.

What it does not claim

  • The three risks and the play are written by a model

    They are the only sentences on the page that are not a sourced fact, and each one prints what triggered it. Where a risk rests on a segment base rate rather than on anything found about the company, it says so.

  • The revenue range is modelled, not measured

    It comes from published sector benchmarks applied to an estimated headcount. It is a band, not a confidence interval, and it is not a figure anyone has measured for your business. The snapshot says so on its own face.

  • The headcount is an estimate from public text

    Where a source gives a band such as 51 to 200 people, Keep stores the midpoint. The company register in the UK does not publish headcount, and Ireland has no equivalent register to check, so this number is the softest input in the model.

  • Keep is new

    There are no customers to name, no case studies and no numbers to quote, so this page does not invent any. If that matters to you, wait and watch it, or write to me and tell me what would make it useful.

If I wrote to you

How I got your details, and how to make it stop.

I saw your company advertise a customer-success leadership role on a public job board. Your name and work address came from public business sources, and I check that an address is real before sending anything to it. I hold your name, your role, your work email and the company it belongs to. I do not hold anything personal about you beyond that.

The legal basis is legitimate interest under Article 6(1)(f) of the GDPR, for business-to-business direct marketing. You will get one email about one job posting. There is no sequence behind it and no second attempt.

If you would rather I had not, say so and I will delete the record and keep only enough to make sure you are never contacted again. I would rather hear that than keep sending. If the snapshot got something about your company wrong, I would genuinely like to know that too, because it means the sources or the code need fixing.

Where Keep fits

One of a line of agents that each own a deliverable.

Keep owns the snapshot. The others own a meeting document, a fix on your screen, a line-item cost, a signed compliance report. Every one is independent and runs at its own address.

The whole line-up lives at botzone.ai. Built in Dublin.

If the snapshot got something wrong, tell me.

It is built from the public record, and the public record is often out of date. A correction is more useful to me than a polite nothing, and it goes straight to a person.

Email stephen@botzone.ai