For businesses that know AI should be helping by now but can't see where to start: two weeks of measurement inside your business, a plan with every recommended build priced in writing, and no obligation to build anything.
State one
Competitors mention AI in their pitches. Your accountant asks about it. Someone on your team uses ChatGPT for drafts and nobody else knows what it's good for. The gap is real — but nobody can say, specifically, where it is in your business.
State two
You're sure some of the week could run itself. The monthly reports, the re-keying, the chasing. But "somewhere" isn't a business case, and you're not going to bet real money on a hunch about your own operation.
State three
Every week there's a new tool, a new agency, a new promise. They can't all be right, and none of them knows how your business actually runs. So nothing gets picked — which quietly costs more than picking wrong.
All three are the same problem: nobody has measured where your time actually goes, so every AI decision so far has been a guess. The Systems Diagnostic replaces the guess with evidence — measured inside your business, priced in writing. Diagnosis before prescription. It's how every other serious profession works.
One on-site day builds your endorsed function diagrams. The interviews and Zed then fill in the detail while we analyse — and your Diagnostics Report lands 48 hours before the Diagnostics Review. Click any step for the detail.
The interviews check what the room said. Zed checks what the interviews estimated. And where the evidence disagrees about the same piece of work, the disagreement is usually the finding.
An education session for your team first — what AI can and can't do, and how we decide whether something is worth automating at all. Then the workshop: your people build out the diagrams of the key business functions under acquisition, delivery and operations — how the business actually makes money and runs itself, drawn by the people who do it.
Those diagrams get endorsed by the business before the day ends. That matters, because they become the reference for everything after: every workflow we assess later is judged for criticality against the endorsed picture of how your business works — not against what happens to annoy people.
Where two groups contradict each other about who does what, we record it rather than resolve it in the room. Those contradictions are usually where the real findings are.
After the on-site day we sit down one-on-one with your people. The staff questionnaire is conducted through these interviews — some people simply complete it in their own time instead, where that suits them better. Named, not anonymous, and confidential: themes travel into the Diagnostics Report, names don't.
During and after the interviews is where each workflow gets described in detail: what actually happens step by step, the edge cases, what goes wrong, what tools are used — and the workarounds and exceptions nobody writes on a form or says in front of colleagues.
We are not asking what should be automated — that's our job. We are asking about the work.
Everything from the interviews feeds into Zed, our time-study agent, working in your Teams. Zed fills the gaps in what we know about each workflow, and gives your people somewhere to drop extra information and issues as they arise through the week. And it runs the time study: checking in during working hours — what are you working on, how long has it taken, were you interrupted — measuring how long each element of each task actually takes.
A questionnaire returns "about 30 minutes". A time study returns "between 6 and 52 minutes across 41 occurrences". The first is a number; the second is a finding.
Afternoon — what did the last hour or so go to?
Monthly client report for Harper & Co. Got pulled into two calls partway through.
Logged — monthly client report · ~40 min hands-on · interrupted twice. Was this a normal run, or an exception?
Everything gets coded and cross-checked against your endorsed function diagrams: every workflow tied to the part of the business it serves, assessed for criticality against the picture your own senior people signed off, and credited by exactly one of the four ROI methods below. What survives gets priced — and drafted into your Diagnostics Report, sent 48 hours before the Diagnostics Review so you meet every number and every price at your desk, not in a meeting.
Want the long version — the day block by block, every rule Zed runs under, and the report section by section? Read the Diagnostic in detail.
Four ways a workflow can pay for itself. Each candidate is credited by exactly one — and picking which one is itself a finding about the work.
Method one
Before automation
Post automation
Presented as a reduction in time and cost, or as a saving per period of time. The simplest and most credible method when a real person really was doing the whole thing on a known cadence. Not used when the automation runs far more often than a human ever could have.
Method two
Before automation
Post automation
Presented as an increase in capacity, or the equivalent value created. Used when the new capacity can actually be used by increasing throughput — often where the process was a bottleneck — and when the automation doesn't reduce human input time but removes waiting.
Method three
Post automation
Post automation · repeating
Presented as an equivalent value per time period. Used when value is realised by having access to live or recently updated outputs. Labour-savings arithmetic loses credibility the more often the automation runs — or where the task was deliberately done rarely because of other business constraints.
Method four
Before automation
Post automation
Presented as a portion of reduced failures, or the increase in success rate — a rate, not a dollar. A workflow may be cheap to do manually, so time and cost savings miss the real value: more customers reactivated, less rework, less production downside.
Worked examples — illustrative figures, not client results. Your Diagnostics Report runs this arithmetic on your own hours and rates, with the working shown.
Fifteen to twenty-five pages, built from your own team's answers. Printed and bound if we're on site, and sent 48 hours before the Diagnostics Review — so every number and every price is met at your desk, not in a meeting.
It reads like a finding, not a pitch. What your team actually said, verbatim and unspun. Where the time goes, measured where we could measure it and honestly estimated where we couldn't. Every automation candidate we found — including the ones we don't recommend building, and why.
Every recommended build is individually priced. The plan is the price list: each item carries its own figure next to what it fixes, prices hold for twelve months, and there's no bundle — take one item, take three, take none.
And it's yours either way. If you never build a thing with us, you still own a measured map of your business's operations, a sequenced plan, and a priced list any builder could work from. The full contents, section by section, are in the Diagnostic in detail.
Executive summary — key findings, one of them usually a surprise
What your team told us — verbatim, anonymised, unspun
The friction map — where the week actually goes
The full candidate inventory — everything we found, ranked, incl. what we don't recommend
Tools and subscriptions — the one hard-dollar page: what overlaps, what renews when, what we deliberately don't touch
The roadmap — now, next cycle, later · sequenced over 12–24 months
What runs first, priced — each item with its own figure · prices hold 12 months
The build order — what you can buy today, in writing
What we left out, and why — so nothing is quietly forgotten
Appendices — per-workflow briefs · your stack and subscription ledger · raw-data summary · the Baseline Record, frozen
Priced by team size, counted on the call and locked at booking. All figures AUD, ex GST.
The front door
Fixed price · invoiced at booking · on-site in one to two weeks
The deliverables
Two weeks, a fixed price, and a plan you keep either way. If your business isn't a fit, we'll tell you on the first call.