One hire, start to finish
James says: "Getting Marketing Directors is hard. Find me one." This is what PeopleOS does about it — told as one story, seventeen screens. A person makes every decision; the system does the donkey work in between.
Chapter 1 · Finding herTen working days, one hire
Josh runs it. James decides twice. The AI never decides anything.
1Josh opens PeopleOS and clicks one button to start the search for a Marketing Director in South Africa.
🛒 Bought — this screen is an off-the-shelf ATS (Teamtailor / Recruit CRM, ~£50–150/mo)
Everything about hiring lives on this one screen
peopleos.equalsfive.co.uk
Hiring+ Start a new search
| Looking for | Where we are | Status |
|---|
| Marketing Director (SA) | Just started | Hard to fill |
| SEO Specialist (UK) | 5 people being assessed | Going fine |
2The AI writes the job ad and the candidate questionnaire. Josh fixes one weak question and approves — nothing goes out that a human hasn't signed off.
🛒 Bought — modern ATSs draft job ads with AI; only the E5 question style is ours
Josh's edit — the AI's question was soft, his is sharper
…/marketing-director/check-and-approve
Check what the AI wrote
Written for you — waiting for your OK
Job ad ✓ approved · LinkedIn post ✓ approved · Questionnaire — 1 question changed:
Question 14: "Describe your ideal role" → "Tell us about a campaign you led that missed — and what you changed"
Ready to publish?Approve & publish
3Thandi applies. The system remembers her — she applied two years ago and was rated "strong, but too junior then". Today that history would be lost in an inbox.
🛒 Bought — remembering past candidates is standard in any talent CRM
This is the whole point: nobody good ever gets forgotten again
…/applications/thandi-m
New application — Thandi M.
We know her
Applied Nov 2024 for Marketing Manager · reached first interview · noted "strong — too junior for this role, keep warm". Two more years of experience since.
Connect to her record
4She spends 45 minutes on the questionnaire — 22 questions about what she's actually done. Her answers arrive in the system by themselves.
🔌 Glue — a Zapier webhook from ScoreApp, an afternoon of setup
…/thandi-m/questionnaire
Her answers · arrived 14:0222 of 22 answered
Question 14 — her answer
"The launch missed because we chased reach over intent. I cut the channel mix from six to two, moved budget into 40 named accounts, and we recovered 84% of the pipeline target…"
5The AI reads all 22 answers and gives Josh a one-page summary: what she's proven, what's missing, what doesn't add up, and what to ask her. The hours someone spends reading these today — gone.
🛠 Custom — the E5 assessment layer. This is part of the 15% worth building
It separates what she PROVED from what it's guessing — and says what it can't know
…/thandi-m/summary
One-page read on Thandi
AI summary — for Josh to judge
Proven: 17 years B2B · led a team of 9 · rescued a missed launch with a named-account strategy.
Missing: no revenue number she personally owned.
Doesn't add up: says "hands-on" in one answer, delegates everything in another — worth probing.
Ask her: "Who owned the recovery number — you or the CMO?"
6Josh reads it, disagrees with one point, and decides: interview her. His reasons are recorded. The AI recommends — it never rejects anyone.
🛒 Bought — stage decisions + notes are standard ATS
…/thandi-m/decision
Josh's call
Josh: "Summary right, except the sector note — her pharma work is old. Interview her."NoInterview →
7Interview one: 30 minutes with Josh. The form is ready and the questions are already tailored to her — built from the gaps in her own answers.
🛒 Bought ATS form · the tailored question pack is the custom bit
…/thandi-m/interview-1
Interview 1 · Josh · 30 min
Her questions1. Who owned the recovery number? · 2. The hands-on contradiction · 3. Team retention through the restructure
AvailabilityFrom September, 3 days a weekRate hopeR7,800/day
8Interview two: she presents a case study to James. James scores her 4.5 on commercial, Josh scores 3. The system won't average that away — they have to talk it out, and the resolution is recorded.
🛒 Bought — interview scorecards are standard; the no-averaging rule is configuration
Disagreement is surfaced, not buried in an average
…/thandi-m/interview-2
Interview 2 · case study · James + Josh
| What we scored | James | Josh | |
|---|
| Strategic thinking | 4.5 | 4.0 | |
| Commercial grip | 4.5 | 3.0 | Talk it out before she goes further |
9Interview three: a 90-minute real task. She's allowed to use AI — and how well she uses it is scored. Better than pleading "please don't use ChatGPT".
🛠 Custom — E5’s own task methodology and AI-use scoring. Part of the 15%
…/thandi-m/task
The 90-minute task
How she worked3 sharp questions in the first 12 minutes · finished 4 minutes early
AI useUsed it for market sizing, challenged its numbers, kept her own judgement — 4/5
The workPrioritisation sharp · measurement plan thin — 4/5
10Decision day — nine working days after the ad went up. Leadership puts her in the Core team, with the evidence attached. Her onboarding checklist starts itself.
🛒 Bought — categorisation + onboarding checklists: ATS + Notion/BambooHR tier
She can't be sent to a client until every box is ticked
…/thandi-m/welcome
Thandi M. — decision9 days, start to finish
Joining as CORE TEAM · evidence attached · James signed it offStart onboarding
Onboarding — running by itself
Handbook sent ✓ · contract details requested ✓ · two intro meetings booked · chat with James pending · Not ready for clients yet
Chapter 2 · Keeping herThe relationship doesn't run on memory any more
Three months later. Thandi is on a client project, billing 4 days a week.
11The system taps James on the shoulder: "You haven't spoken to Thandi in 7 weeks" — and briefs him before the call: she's been working over her agreed days.
🛠 Custom — the join: this nudge reads her record AND her timesheets. Part of the 15%
Today this depends on someone remembering. Now nothing slips
…/people/thandi-m/catch-up
Time to talk to ThandiBook 15 minutes
Before you call — 20-second brief
Last chat 10 Jun: workload "heavy" · Since then: two weeks at 5 days when her agreement says 4 · her project ends 28 Nov, nothing booked after. Open with the workload.
12The catch-up is 15 minutes. James records a voice note afterwards — the AI types it up, updates her record, and James approves what it changed. Nobody writes up meeting notes.
🛠 Custom — voice note → approved record update. Part of the 15%
…/thandi-m/catch-up/notes
What changes on her record
From James's voice note
Days from October: 4/week 3/week · + interested in strategy-day work · mood: good, but "at the edge" on workload · working over her agreement · nothing planned after November
James approves what goes on the recordApprove
13The two problems don't sit in anyone's head — each becomes a job with a name and a date on it, chased until it's closed.
🔌 Config — an action list; any task tool does this
This list is the Friday meeting
…/people/needs-attention
Needs attention
| Problem | What we're doing | Who | By when |
|---|
| Thandi over her days | Re-agree the deal or add support | James | 28 Aug |
| Nothing after Nov | 2 matching projects found — offer one | Josh | 15 Sep |
Chapter 3 · Paying her rightThe bill always matches the deal
Her agreement: 4 days a week, £620 a day, never more than £43,400 total.
14Thandi logs her week in two minutes. She types a 5th day — the system flags it instantly, because her agreement says four.
🛒 Bought — this is a PSA tool (Productive / Scoro, ~£10–50/user/mo)
Caught the moment it's typed — not three weeks later in a spreadsheet
thandi's view …/my-week
Week of 17 Aug
| Mon–Thu | Client project | 4 days | ✓ fine |
| Fri | Extra workshop | 1 day | ⚠ your agreement says 4 days |
Send 4 days now, flag the 5th for a decision?Send it
15Friday afternoon: the project lead approves the week in one look — with a note back to Thandi about the extra day.
🛒 Bought — weekly approval is standard PSA
…/approve-the-week
Approve this week
Thandi · 4 days ✓ + 1 flagged. Note: "Approving 4 — the extra day needs the new agreement James owes you."Approve 4
16Her invoice arrives at month end. The system checks it against the deal and the approved days — and catches £310 too much: she billed the day that was never approved.
🛒 Bought at volume (Ramp / Precoro, ~£3/invoice) — at today’s volume, our AI rules-check does it
This check is what stops quiet overbilling — automatically, every invoice
…/money/thandi-invoice-aug
Thandi's August invoiceDoesn't match
| The deal | Days approved | Her invoice |
|---|
| Days | up to 16 | 12 | 12.5 |
| Money | £620/day | £7,440 | £7,750 — £310 over |
Why
The invoice includes the flagged 5th day that was never approved.
17One click asks her for a credit note. When every invoice matches, August locks — nothing can quietly change afterwards. A person clicked approve on every pound.
🛒 Bought — month locking is standard PSA / accounting
…/money/close-august
Close AugustEverything matches
All weeks approved ✓ · all invoices match ✓ · Thandi's credit note received ✓ · sent to accounts: £11,940
Lock the month?Lock 🔒
The honest tallyOf these 17 screens: 11 bought, 2 glue, 4 built
The bought screens already exist — see them live: Teamtailor or Recruit CRM (hiring, screens 1–3, 6–8, 10) · Productive or Scoro (time & money, screens 14–15, 17) · Ramp / Precoro (invoice matching at volume, screen 16). The four magenta screens — the assessment layer, the task methodology, the shoulder-tap that joins people to money, the voice-note concierge — are the only part nobody sells. That 15% is what Tim builds.
That's the whole system. Find people and never forget them. Decide with evidence, in days not months. Keep the relationship warm without relying on memory. And never pay a pound that wasn't agreed. The AI does the reading, writing, reminding and checking — a person makes every single decision.