Case studies · Zeroth

Engineering consultancy · Defence & asset-intensive

The whole back office, absorbed into the tools they already had

Every invoice meant typing the same hours twice. Asset Management Associates are reliability, ILS and RAM engineers, a small technical team whose billable hours are the product. So the admin around those hours was expensive in the only currency that matters: senior engineering time spent not engineering.

Timesheets lived in Harvest, a separate subscription with a separate login that nobody opened until Friday. Getting from recorded time to a sent invoice meant re-keying the same numbers in a second place, with a third chance to get them wrong. Everything else the firm ran on was already Microsoft 365.

We put the AI OS in the middle of it.

Discuss a build like this

The system

Time captured conversationally, against real projects. No separate app, no Friday reconstruction
Harvest retired. Timesheets moved into the same system as everything else, so there is one record of an hour instead of two that have to agree
Approved hours draft straight into an invoice, and those hours are stamped as billed in the same transaction. Nothing is invoiced twice, nothing is quietly missed
Wired into the Microsoft 365 they already ran: same sign-in, same mailbox, same calendar, no new platform to adopt
Nothing irreversible happens on a single call. An invoice going to a client is staged and released by a person, after a fresh sign-in

Results

One subscription gone. And with it, an entire re-keying step
Time to invoice is now a conversation, not a Friday afternoon. Recorded hours to drafted invoice is one reviewed step
Every action against their accounting system is on an audit trail nobody can edit, including us
No new software on any engineer's machine. The OS reaches them through the tools already open

Management consulting · Federal government

AI-native from the first week, before the habits set in

Every process a firm runs on gets invented once, by hand, usually by the founder who can least spare the hour. 41 South Consulting is a Canberra-based consulting firm specialising in P3M and change management, serving government clients. Founded in 2025, and just getting started as we built this.

That last part is what's unusual. Most firms come to us carrying a decade of habits: processes invented under pressure, held in two people's heads, and expensive to unpick. A firm this new has none of that yet, and none of the processes either.

So the OS went in before the headcount did.

Discuss a build like this

The system

The full OS from day one. Five role groups configured and ready before there were five people to hold them
Time, projects, clients, invoicing and the document store in one system, before they could become five subscriptions accumulated over three years
Running on their own Microsoft 365 and Teams, with accounting wired through the gateway
Client prospecting and recruitment prospecting built as workflows on top, because for a firm this size those two activities are the growth plan
Their business context captured as the firm's own documents: how they write, how they work, who they serve. The system speaks in their voice

Results

Live in their own Microsoft tenant inside a week of the build starting. Signed in, with email, calendar and Xero connected
The firm's operating processes exist as a system before the second hire. Nothing important lives only in the founder's head
Adding a person means provisioning an account. There is no fortnight of shadowing, because the role already exists and already knows the firm
Prospecting for clients and for people runs on the same platform as delivery, so nothing has to be joined up later
Growth doesn't mean re-platforming. Automations get added one at a time, as the work justifies them

Real estate · Property management

Finding the revenue that walks out with lost managements

One agency's year of lost managements came to ~$1.1M in rent-roll value and commission. Every agency loses them, quarter after quarter: a landlord sells, an owner moves back in, or a competitor quietly takes the file. Almost nobody adds up what walked out the door, because working out where each property went means looking up every address by hand. So it never gets done.

We built a quarterly automation that does the looking up.

Discuss a build like this

The system

Ingests the agency's list of ended managements, in whatever layout it exports
Resolves each property on realestate.com.au and classifies the outcome: sold, re-leased by another agency, or returned to the owner
Attributes each loss to the competitor now holding the file, and applies the agency's own fee and commission rates
Delivers a branded quarterly report: losses by manager, financial impact, competitor attribution, quarter-on-quarter trend

Results

A pilot run classified 167 ended managements with no manual lookups
Surfaced ~$1.1M in rent-roll value and commission across one agency's year of lost managements
Every dollar figure derives from the agency's own rates. The automation estimates nothing

Real estate · Lead generation

Outreach drafted by the system, approved by a human

Prospecting properly means reading each profile, judging fit against your criteria, and writing an email that sounds like you. Hundreds of times. Generic mail-merge gets ignored, and doing it by hand doesn't scale. The usual answer is to automate the sending and hope. That's how firms end up with a reputation problem.

So we built the drafting, and kept a human on the send button.

Discuss a build like this

The system

Reads the target agent profiles from realestate.com.au and evaluates each against the user's own criteria, written in plain English
Drafts an individual email per qualified agent, in the user's voice and style
Then it pauses: the drafts arrive by email, and nothing sends until the user replies with a yes
Reply "approve, but make it shorter" and it redrafts before sending. Every step lands in an audit trail

Results

One approval reply per batch replaces hours of reading and writing
Zero emails sent without an explicit human yes, because the system stops and waits for one
While it waits for approval, it costs nothing to run

Systems we've built, and what they changed.

Where the manual work went away.

The whole back office, absorbed into the tools they already had

The system

Results

AI-native from the first week, before the habits set in

The system

Results

Finding the revenue that walks out with lost managements

The system

Results

Outreach drafted by the system, approved by a human

The system

Results

Want your operation on this page?