How we count

Every number here, and where it came from.

Every figure on this site sits next to a named client and a link to the system that is running. This page says what each number measures, how it was arrived at, how it was checked, and — where we have not finished checking — exactly what is still open.

19 of the entries below are marked not yet confirmed. They are published that way on purpose. A number whose basis we cannot describe is worth less than no number, and pretending otherwise is the failure this page exists to prevent.

The rules

Three rules, applied to all of them.

  1. 01

    We count hours the client told us, not hours we estimated.

    The starting point for every "saved" figure is the client's own description of the job before we built anything — how many people, how many hours, how often.

    Not yet confirmedWhether this holds for all four, or whether any hour count was ours rather than theirs.
  2. 02

    Salary-equivalent means salary, not "productivity".

    Where a saving is a salary equivalent, it is the hours replaced multiplied by a stated hourly rate. It is not a claim that anyone was made redundant.

    Not yet confirmedThe rate used for each figure. The cost guide's worked example uses £20/hour; whether that was the basis for PlusRooms and Marmadbir is not yet confirmed.
  3. 03

    The client saw the number before we published it.

    No figure goes on the site that the client has not seen attached to their own name.

    Not yet confirmedConfirmation for each of the four.
PlusRooms

Planning-application monitoring across England, replacing a daily manual trawl of council websites.

We say

A full working day, every day, replaced

Not yet confirmedHours per day and days per week, and whether the role was reassigned or the task simply ended.
What was measured
One person's daily manual check of council planning portals across London, before the pipeline existed.
Basis
The client's description of the job: one person, one full working day, every working day.
How it was checked
The manual check stopped when the pipeline went live.

We say

≈ £27,000/yr — a full-time role's worth of checking

Not yet confirmedThe rate and the number of weeks per year used. £27,000 ÷ 46 weeks ÷ 37.5h is about £15.65/h; ÷ 48 weeks is about £15/h. Which one was the basis is not yet settled.
What was measured
The salary equivalent of that working day.
Basis
Full-time hours × an hourly rate.
How it was checked
Arithmetic, from the row above.

We say

97% London borough coverage

Not yet confirmedWhether the figure is 32 of 33, as of what date, and which borough is missing and why.
What was measured
Boroughs whose planning portals the pipeline scrapes successfully, out of 33.
Basis
The scraper's borough list, counted.
How it was checked
Counted from the pipeline configuration.

We say

Council websites checked by hand: none

Not yet confirmedThe date it was last asked.
What was measured
Whether anyone at PlusRooms still opens a portal manually.
Basis
Client statement.
How it was checked
Asked directly.

We say

Refreshed every 24 hours

Confirmed
What was measured
The scrape schedule.
Basis
Scheduler configuration.
How it was checked
Dashboard timestamps.
Marmadbir

WhatsApp dispatch for a field-services platform. Every figure here is the coordinator's, not the field workers'.

We say

Dispatcher time spent chasing workers: eliminated

Not yet confirmedHours per day and how many coordinators, and whether anyone still phones a worker for edge cases.
What was measured
Coordinator hours spent contacting workers one by one, over WhatsApp and phone, to allocate jobs.
Basis
The client's description of the job before the build.
How it was checked
Jobs are broadcast and claimed in the system. There is no manual allocation path left in the code.

We say

Messaging cost down ~65%

Not yet confirmedExactly what was compared — previous call and SMS spend against WhatsApp template cost — and over what period.
What was measured
Per-message cost of coordination, before against after.
Basis
A comparison of messaging spend across the change.
How it was checked
Supplier invoices.

We say

≈ £10,000/yr in coordination time

Not yet confirmedBoth the hours and the rate.
What was measured
The salary equivalent of the coordinator hours removed.
Basis
Hours × rate.
How it was checked
Arithmetic.

We say

Payments reconciled by hand: none

Not yet confirmedWhether a back-office spreadsheet exists outside the system.
What was measured
Whether any payment is matched to a job manually.
Basis
Payment claim is race-safe, and payments release on client confirmation inside the system.
How it was checked
The code path: there is no manual reconciliation step.

We say

New tenant live in under five minutes

Not yet confirmedWhether this was measured once or is a typical figure across tenants.
What was measured
Time from creating a tenant to that tenant sending automated messages.
Basis
Measured during onboarding.
How it was checked
Timed once.
Punthub

Built and owned by The Automation Agency. This is our own product, so it is evidence of what we can run unattended — not a client engagement. We label it that way everywhere it appears.

We say

Staff time to run it each night: none

Not yet confirmedOver what period zero touchpoints was observed.
What was measured
Human involvement in the nightly run.
Basis
The scheduler runs collection, the models, reconciliation and the sync without intervention.
How it was checked
Run logs.

We say

£13,000+/yr in operating costs eliminated

Not yet confirmedWhat the costs actually were — a part-time operator, a data vendor, a hosted service — and the arithmetic behind the figure. This is the least substantiated number on the site.
What was measured
Operating costs that stopped when the pipeline replaced the previous arrangement.
Basis
Not yet written down.
How it was checked
Not yet documented.

We say

6 sources, daily

Confirmed
What was measured
External data sources collected each evening.
Basis
Pipeline configuration.
How it was checked
Counted.

We say

7 prediction models · 21 reconciliation jobs

Not yet confirmedWhether it is still 7 and 21 as of today.
What was measured
Models and scheduled reconciliation jobs in the run.
Basis
Scheduler configuration.
How it was checked
Counted from the codebase.

We say

Figures reconciled by hand: none

Confirmed
What was measured
Whether any nightly figure is corrected manually.
Basis
There is no manual step in the run.
How it was checked
Run logs.
A Story To Tell

A rescue-and-launch engagement rather than an automation build. These numbers are elapsed times, not savings.

We say

Launch-ready inside a week

Not yet confirmedThe start and end dates.
What was measured
Elapsed time from our first commit to the client signing the platform off as launch-ready.
Basis
Commit history and the client's sign-off.
How it was checked
Both are on record.

We say

Previously: 8 months without a shipped change

Not yet confirmedWhether the repository history corroborates it.
What was measured
Time since the last deployed change before we started.
Basis
The client's account of the previous arrangement.
How it was checked
Not independently checked against the repository.

We say

Previously: 2–3 weeks to swap a single image

Not yet confirmedWhether the 2–3 week figure specifically is from the client in writing.
What was measured
Turnaround the client experienced for a trivial content change under the previous arrangement.
Basis
The client's account.
How it was checked
Their Google review describes the same experience in their own words.

We say

Included, unscoped: SEO, AI search, security hardening

Not yet confirmedA line-by-line comparison has not been written up.
What was measured
Work delivered beyond the agreed scope, at no extra charge.
Basis
The scope document against what was delivered.
How it was checked
Both are on record.
What we don't count

Deliberately left out.

Work still in testing is not named or counted anywhere on this site until it is live and the client has agreed to be named.

Ask us about any of it