Funnalized
DEENTell me when it starts
Module 19 · OutcomeOS

Prove that it worked — instead of believing it

Set a goal, watch where it stands run along from real records — and test a measure against a control group.

0Capabilities
0Paths through
0Handovers
0Entries in the catalogue

Screens from a demo workspace. Companies, people and figures are fictitious.

What it does

OutcomeOS at work

Four capabilities people actually open OutcomeOS for. Everything else it can do is listed in full in section 04 — this is only what stands on its own.

The load-bearing capability

An impact test on real records

Test two approaches against each other — directly on leads, deals or candidates. Newly created records are assigned to a variant automatically and by weight the moment they are created: per record, once, and unchangeable afterwards. The same record never ends up in two variants.

Never twice

the same record never lands in two variants

The comparison group that makes proof possible in the first place

Alongside control and variant you can form a comparison group: it is allocated but gets no automations — deliberately untreated. Only that comparison separates “it got better” from “it would have got better anyway” — the difference between an observation and a proof.

Observation → proof

only the comparison with the untreated group turns it into proof

A verdict in one sentence instead of pretty bars

For every variant you see the difference in absolute and relative terms, the p-value, the confidence interval and the clear statement of whether the result is dependable — as a sentence: “Variant B: +12 percentage points conversion rate — very unlikely to be chance.” A verdict comes only once every group has reached its minimum size; until then it says how many records each group still needs. If several variants run at once, the threshold gets stricter accordingly.

Minimum sample

stops three lucky hits from passing as a success

Your next step, not only your numbers

Due now

or overdue, paused, or deliberately in the comparison group — kept apart cleanly

“Next steps” works like an inbox: it says where something is needed from you right now — and distinguishes cleanly: due now, overdue after 14 days without activity, in progress, paused by a safeguard, or in the comparison group of an impact test and therefore deliberately without automations.

Handovers

What OutcomeOS receives and passes on

Comes in to OutcomeOS
OutcomeOS
Goes out of OutcomeOS

OutcomeOS proves whether a measure achieved anything. Where the day stands is shown by the dashboard — proof is a different question from status.

Complete

Everything that is in it

0Entries in the catalogue

25 capabilities in 3 groups — complete, not curated.

Honestly

What it does not do

Every limit names the place where it happens instead.

  • No language model.

    The computation is statistics you can read. A model that guesses at an effect would be cheaper and worthless.

  • No control group, no proof.

    Without a marked control group there is no verdict, only a trend. And it says so, instead of claiming success.

  • Thresholds are fixed in advance.

    Significance level and duration are locked to the impact test. Moving them afterwards is not possible — otherwise every experiment wins.

Follow-up

What teams want to know beforehand

How long does an experiment take?

As long as it takes to be meaningful, decided up front — the duration is locked when you create it and not adjusted once you like the result.

Do my people notice anything?

Only the measure itself. Assignment runs in the background, and the records look exactly as they always do.

What if nothing comes out of it?

Then it says nothing came out of it. An experiment without an effect is a result — and cheaper than a measure you keep running for years.

All 25 entries as a list

Goals, the funnel and your next step 13

  • One overview, three questions At the top there is one sentence: whether you are hitting your goals, which records need a step now and whether a change really made a difference. Below it the goals, those at risk first, the five most urgent next steps with the way to the record, and the running impact tests with their verdict. Every section says under “How this is calculated” where its number comes from.
  • Goals that keep track of themselves Set what counts — twenty qualified leads a month, ten hires a quarter — and progress calculates itself from your real records. No maintaining a side spreadsheet that nobody touches after two weeks.
  • Goals per week, month, quarter, year or with a deadline — as a count or, for deals, as a sum, for everyone, one person or a team, with a forecast to the end of the period
  • The first goal in one click from one of four starter templates — the target is suggested from your last period, ten per cent more
  • Your next step, not only your numbers “Next steps” works like an inbox: it says where something is needed from you right now — and distinguishes cleanly: due now, overdue after 14 days without activity, in progress, paused by a safeguard, or in the comparison group of an impact test and therefore deliberately without automations.
  • Early warnings that clear themselves If a goal falls behind plan, at least three records of one kind stay overdue, or the conversion rate of the last four weeks drops demonstrably below that of the weeks before, an early warning reports itself — with the way to exactly those records. Once the reason is gone, it disappears by itself.
  • Every Monday at 7 am the weekly report: where the goals stand, new leads, won deals and hires of the previous week, open early warnings and the five most urgent steps
  • The funnel per record type A funnel view per record type — leads, candidates, deals, campaigns, projects, tasks — shows the run across the stages and where records get stuck, instead of it only showing up in the monthly report.
  • Rules that apply before the action Before a measure is triggered, a policy checks whether it is permitted — in the app it is called a safeguard — and logs the decision along with its reason. That keeps it traceable why somebody was contacted and why somebody else was not.
  • For every case a timeline of the events recorded — and you can add your own
  • What a case is connected to sits beside it as a link, with a weighting for the connection
  • A reported anomaly you tick off as resolved if it is technical — early warnings clear themselves
  • A goal's progress over time — the count reached, the percentage met, the pace along the way and the three previous periods as bars

How an impact test runs 8

  • An impact test on real records Test two approaches against each other — directly on leads, deals or candidates. Newly created records are assigned to a variant automatically and by weight the moment they are created: per record, once, and unchangeable afterwards. The same record never ends up in two variants.
  • The comparison group that makes proof possible in the first place Alongside control and variant you can form a comparison group: it is allocated but gets no automations — deliberately untreated. Only that comparison separates “it got better” from “it would have got better anyway” — the difference between an observation and a proof.
  • A verdict in one sentence instead of pretty bars For every variant you see the difference in absolute and relative terms, the p-value, the confidence interval and the clear statement of whether the result is dependable — as a sentence: “Variant B: +12 percentage points conversion rate — very unlikely to be chance.” A verdict comes only once every group has reached its minimum size; until then it says how many records each group still needs. If several variants run at once, the threshold gets stricter accordingly.
  • Limits written down at the trial An impact test must not ruin your business. For every impact test you record the limits that may not be undercut — they stand at the trial and are visible to everybody who reads the evaluation. And what is not live yet we say openly: automatic monitoring against the running actuals is in the works; today you hold the limits against the result evaluation yourself.
  • Draft: set variants, distribution and limits
  • Running: newly created records are assigned automatically — existing ones not retroactively
  • Paused: assignment stops, what is running stays untouched
  • Finished: evaluation with p-value, confidence interval and a verdict as a sentence

Why no language model does the maths here 4

  • Deterministic — no language model involved OutcomeOS calculates rule-based. The same data yields the same answer, recalculable at any time, with no provider getting to see your records. What is called “at risk” here has a reason you can read.
  • A goal counts only the slice you define — deals from a particular pipeline and stage, say, or leads with a particular status
  • When you create one, the app suggests goals worked out from the real activity of the last thirty days — not from a list of examples
  • A rate from your own closed records For a lead, deal or applicant you see the rate of its stage: how many closed records from that stage you won or hired, with the case count and a 95% range. Below 20 closed cases there is a dash and the reason — no number out of thin air.

One message at launch. After that you'll find me in the top right of the app.

OutcomeOS is one of 21. You get all of them.

Sign up and you'll hear from us the moment we go live. Not a newsletter — one message.

€50€150/ month · permanently

Waiting list

Only your email address. No spam, unsubscribe at any time.

See all modules

€50€150/ month · permanently

50 founder placesTell me when it starts