The #1 Task Your Managers Won't Do.

7,000 hours nobody wants to talk about

Performance management is the third highest-value function an executive performs (McKinsey, Gartner, BCG). It's also the number one task they avoid (Harvard Business Review, AHRI, SafetyCulture).

The answer is the same in every survey: emotional discomfort, fear of legal risk, and the 13 to 24 hours it takes per employee per review cycle. AHRI's benchmark puts it at ~149 hours per people manager annually. For an organisation with 50 managers, that's over 7,000 hours a year on a process most of them are actively avoiding.

In 2021, only 3% of Australian organisations used AI for any part of performance management (AHRI). Five years later, most are still doing it the way they did in 2019.

SafetyCulture found Australian middle managers waste 6.5 weeks per year on low-value tasks. That's $15.5 billion economy-wide, gone to admin that adds nothing. Australia's labour productivity fell 0.6% in 2024-25 (Australian Industry Group).

60% of managers already work remotely (McCrindle), spending 6+ hours per day on mobile devices (Red Search, 2025). The work moved to the phone. The tools haven't.

What four minutes produced

A team member is underperforming. You know it. Your direct reports know it. HR wants documentation before anyone sits down with them. Sitting across your systems: 18 source files. A 121-line performance spreadsheet covering 10 employees across 12 weeks. 14 individual work samples (client reports, coaching notes, incident summaries, escalation logs, risk registers). 25 lines of communication excerpts from Slack and email. 527 lines of operational data total.

That's three weeks of manual assembly. Minimum. Most never finish it. The conversation never happens. The underperformer stays. Performance dips, culture goes out the window...

Until now, AI tools on your phone couldn't touch your local files. Claude Dispatch changes that. It turns your phone into a remote control for your computer. One message from my phone, and Claude opens files on my machine, reads spreadsheets, searches folders, and builds output documents while I'm away from my desk.

I typed one sentence on the couch while my laptop ran the session in the other room. Four minutes later, I had a 5-document evidence pack, a 30-minute discussion plan with scripted talking points, and a 4-week PIP structure. For $30.

Claude cross-referenced the 12-week performance trends against the work samples, flagged the communication breakdowns by date, benchmarked against peers, and built a "what Lucas will say and how to respond" objection matrix.

This has removed every excuse your managers have for avoiding the 'hard' high value work.

What's inside

1. Five Dispatch prompts you paste once and reuse every cycle
2. The cost maths your CFO hasn't seen yet
3. Each prompt searches your local files and creates documents on your desktop
4. The governance line you can't cross

The maths your CFO needs

• Before → Those 149 hours per manager, per year (AHRI) After → 4 minutes with a remote AI agent
• Before → $400-$600 per employee per year in manual HR process costs (AHRI). For 50 people: $20,000-$30,000 in process overhead alone, before salary time. After → $360/year (Claude subscription)
• Before → 3-4 week turnaround from cycle kickoff to completed reviews After → Same afternoon.

Replacing a departing employee costs 100-150% of their annual salary (ForeMind, 2026). For a $100K role, it's $100K-$150K of avoided replacement cost. Your current process now has to justify its cost against that alternative.

Run the numbers on your own business.

Prompt 1: The Evidence Pack

This is the one I built and tested. Paste this into Dispatch from your phone and let your computer do the assembly.

I'm dealing with a performance issue for: [EMPLOYEE NAME] in [TEAM/ROLE].

On my computer, find and open anything relevant from the last [6-12] weeks, including:
- 1:1 notes, coaching notes, performance notes
- weekly updates, status reports, incident summaries, QA notes
- spreadsheets with volume/quality/SLA metrics
- any docs where [EMPLOYEE NAME] is mentioned

Create a folder on my Desktop called "Performance Pack - [EMPLOYEE NAME]".

Create 3 files in that folder:

1) Evidence_Pack.docx
- A dated table: Evidence | What it shows | Source (filename + date)
- Only include claims you can cite to a file.
- If evidence is missing, write "Missing evidence" and list what I should gather.

2) Discussion_Plan.docx
- Fair opening script (60-90 seconds)
- 5 diagnostic questions
- 3 measurable commitments for 30 days
- weekly check-in agenda
- 5 likely objections and calm responses, based only on the evidence you found

3) One_Page_Summary.docx
- What's happening
- What's driving it (top 3)
- What I should do this week (3 actions)

Rules:
- This is workflow support, not HR advice.
- You do not decide outcomes.
- Do not invent facts. Cite sources.

What came back when I tested it: 11 structured evidence files totalling 676 lines of analysis. Every work sample annotated with specific dates and quality flags. A week-by-week tone shift map across Slack and email. An objection-handling script built from the employee's actual language patterns.

Three final deliverables came out formatted and ready to hand to HR.

The process it replaces: HR coordination, calibration sessions, and $20,000-$30,000 per year in overhead alone (AHRI).

Prompt 2: The Missing Evidence Finder

This is the one you run when you know the conversation is coming but you haven't been documenting properly.

I'm about to have a performance conversation with [EMPLOYEE NAME]. I'm worried I don't have enough receipts.

Search my computer for anything relevant in the last [6-12] weeks and tell me:
1) What you found (by evidence type)
2) What you did NOT find (missing evidence)
3) The smallest set of documents I should create this week so the next conversation is fair and documented

Output a single file on my Desktop: "Missing_Evidence_Checklist_[EMPLOYEE NAME].docx".

Rules: do not invent facts. Cite filenames for anything you reference.

Prompt 3: The Trend and Driver Pack

For [EMPLOYEE NAME], find the most relevant metrics files on my computer (spreadsheets, exports, CSVs) for the last [6-12] weeks.

Create "Trend_and_Drivers_[EMPLOYEE NAME].docx" on my Desktop with:
- 3-5 key metrics trends (what moved, when, how much)
- likely drivers (ranked), but label speculation clearly
- what I should check next (specific files or questions)

Rules: cite source filenames. Do not guess missing numbers.

Prompt 4: The Comms Pattern Extractor

This one pulls behaviour evidence from the mess of notes and docs already on your machine.

For [EMPLOYEE NAME], search my computer for comms evidence (meeting notes, written updates, emails saved as files, copied Slack excerpts in docs).

Create "Comms_Patterns_[EMPLOYEE NAME].docx" with:
- 10 dated excerpts (short) showing patterns (handover quality, escalation, missed commitments, clarity)
- what each excerpt suggests (1 line)
- a "fairness check": 3 alternative explanations and what evidence would confirm/refute

Rules: cite filename + date. No psychoanalysis. No HR templates.

Prompt 5: The 30-Day Reset Plan

Last one. This is the one you run after the conversation happens and you need a structure for what comes next.

I'm managing [EMPLOYEE NAME] and I want a 30-day performance reset plan that is measurable.

Using any evidence you can find on my computer for the last [6-12] weeks (notes, metrics, work samples), create:
"30_Day_Reset_Plan_[EMPLOYEE NAME].docx"

Include:
- 3 measurable targets for 30 days (inputs + outputs)
- weekly cadence (what happens each week)
- what I will do (manager actions)
- what they will do (employee actions)
- how we will measure progress (exact metrics or artefacts)
- what happens at Day 30 (decision options, written neutrally)

Rules: cite any evidence used. If evidence is missing, list what's needed.

Remember this is hours of sitting at your computer, now all done via 1 msg on your phone!

Where this breaks

Things I found when testing.

• The evidence pack cited a file that didn't exist. It hallucinated a coaching note with a plausible filename and date. Every citation needs a human eye before it leaves your desk.
• The comms pattern extractor over-interpreted tone in two neutral messages. It flagged a standard handover email as "disengaged" when the context showed the employee was on leave the day before.

The outputs weren't perfect. They were 80% there. The remaining 20% is judgement calls a manager should be making anyway. These aren't reasons to avoid the tools - but reasons to keep a human in the loop.

I cover failure modes like these every week on the Applied AI Australia podcast. If you're not subscribed yet, fix that: Apple Podcasts | Spotify

The OAIC's guidance on automated decision-making: if an AI output influences a decision about an individual (performance ratings, terminations, promotions), you need human review, explainability, and a documented process. The Privacy Act requires that personal information processed by AI tools meets the Australian Privacy Principles, including data minimisation and purpose limitation. VAISS (Voluntary AI Safety Standard) isn't mandatory yet. I'm confident it will be within 18 months. Use AI to draft. Use AI to analyse. Use AI to surface patterns. The decision stays with a human.

Before Friday

Everything I built in this article ran through Claude Dispatch. Not a 'chatbot'. An AI agent that runs locally on your computer and takes instructions from your phone.

Here's how to set it up:

1. Update your Claude desktop app and mobile app to the latest version
2. Open the desktop app. Go to Cowork. Click Dispatch in the left panel
3. Scan the QR code with your phone to pair the two devices
4. Grant file access permissions when prompted
5. Turn on "Keep Computer Awake" in settings so Claude can work while you're away from your desk

One persistent conversation across both devices. You type from your phone. Claude works through the files on your computer. Plans start from $30/month. I pay $159 for Claude Max (I am not sponsored by Anthropic)

Paste Prompt 1 tomorrow morning.

If you're not subscribed to the Applied AI Australia podcast yet, now is the time - your career and company cannot afford leaders who don't understand and apply AI.

Apple Podcasts | Spotify

Remember: You can't control the speed of change, but you can control how fast you adapt.

Ramon.

---

This is general information and does not take into account your specific situation, organisation or needs. Before you rely on any ideas or workflows here, get advice tailored to your context.