Plain English · the long version

The five patterns in full.

The short version is on the home page. This is the whole thing: what each pattern is, what it changes in a business, where it falls over, and what it costs to keep running once it is in. Written for an owner who has used ChatGPT twice and been told nothing since.

← Back to anum.com.au
01
Assistants

Assistants your team already has

Microsoft 365 Copilot, ChatGPT, Claude. A chat window that drafts, summarises, rewrites and explains.

What it is

General purpose assistants, sold per person per month. Most businesses your size already pay for at least one of them, usually because it arrived inside a licence bundle nobody read.

What it changes in the business

On its own, close to nothing, which is why so many owners conclude AI is overrated. It changes the work once someone sets it up against your work: a prompt library for the jobs your team repeats every week, a written rule about what can go near client information, and training done at people's own desks on their own tasks. Then the quote, the file note, the report and the client email stop starting from a blank page.

What it costs to run

A per person monthly licence you are probably already paying, and nothing further to operate. The spend is the setup and the training, once. This is the cheapest and the fastest of the five, and it is where we start with most businesses.

02
Retrieval augmented generation

Answers from your own documents

Retrieval augmented generation, usually shortened to RAG. The plain version: instead of the AI answering from what it read on the internet, it searches your own documents first and answers from those, quoting the source. The retrieval is the search step. The generation is the writing step. The order matters, because the search is what stops it making things up.

What it is

Your contracts, procedures, policies, manuals, past correspondence and file notes are indexed so they can be searched by meaning rather than by keyword. A question goes in, the system finds the relevant passages, and the answer is written from those passages with a pointer back to where each part came from.

What it changes in the business

Your staff stop hunting through folders, shared drives and old inboxes for an answer somebody already wrote down. The answer arrives with the clause or the page attached, so it can be checked rather than trusted. A new starter reaches competence in weeks instead of months, because the knowledge stops living in the head of whoever has been there longest. And when the answer genuinely is not in your documents, a system built properly says so instead of inventing something that sounds right.

The honest limit

It is only ever as good as the documents you feed it. Give it procedures that are three years out of date and it will quote three year old procedures with complete confidence. Someone has to decide what goes in, keep it current, and set who is allowed to see what.

What it costs to run

Ongoing but small: storage for the indexed documents, plus a usage charge per question asked. The effort sits at the front, in choosing and preparing the material.

Worked example · Alma

Alma answers Victorian owners corporation questions grounded in the Owners Corporations Act 2006, and cites the section it used. A manager can check the answer against the Act instead of taking it on faith. That is the whole point of the pattern.

03
Tool use

AI that does the task, not just the chat

The AI is given tools and permission to use them, so it acts inside a system rather than talking about it.

What it is

It reads the receipt, pulls out the date and the amount, codes the transaction, drafts the entry and files the record in the system that holds it. A person checks the result and approves it. The AI is doing the handling, and your staff are doing the deciding.

What it changes in the business

Friction and cost. Work that used to need a person at a keyboard for twenty minutes becomes one review click. The saving shows up twice: once in the time, and again in the errors, because the repetitive step stops depending on how tired the person doing it is at five o'clock on a Friday.

What it costs to run

A small charge per document or per transaction processed. It scales with your volume, so it stays cheap while you are small and stays proportionate as you grow. There is no seat licence to buy for people who barely touch it.

Worked example · Summed

Summed is bookkeeping software for Australian sole traders and small teams, and the entire design is easy tool use. The tools do the reading, coding and filing. The owner reviews. Less friction in the week, and the cost of keeping the books stays down.

04
Agent orchestration

Automating a whole workflow with agents under supervision

Several AI workers, each with one defined job, handing off to each other down a chain, with a person approving the parts that matter.

What it is

One agent drafts. One checks it against your rules. One files it and notifies the person who needs to know. Every run is logged, so you can see what ran, what it used and what it decided, and a human sits at the approval points you choose.

What it changes in the business

Throughput on repeatable multi step work. The kind of job with six steps, four systems and one person who knows the whole sequence. That sequence gets written down and run the same way every time, so it stops falling over when that person takes leave.

The honest limit

This is the most involved of the five and the easiest to get wrong. It only makes sense once the process itself is stable. Automate a broken process and you get the same mess, faster and harder to unpick.

What it costs to run

The most of the five, because several models work on every job and a person still spends time approving. It earns that back on high volume repeatable work and rarely earns it anywhere else. We will tell you which one you have.

Worked example · Nexus

Nexus is the platform Anum runs its own delivery on. Specialist agents with scoped tasks, a human in the loop on anything that matters, and every run logged. We recommend this pattern to a business only after we have lived with it ourselves.

05
Content operations

Producing consistent output at volume

Briefing, drafting, brand checking and scheduling handled as a pipeline, rather than as one person writing everything when they get a free afternoon.

What it is

Your tone of voice, your claims and your rules are written down once, in a form the system can check against. Briefs go in one end. Drafts come out the other, already checked against those rules, ready for a person to edit and approve.

What it changes in the business

Marketing output stops depending on who has a spare afternoon. What goes out sounds the same whoever produced it, because the voice is written down and checked instead of carried around in one person's head. Volume goes up without a hire.

What it costs to run

A usage charge per piece produced, plus the time of whoever edits and signs it off. It does not remove the editor. It removes the blank page and the drift in tone.

Worked example · A Melbourne consulting firm

A content platform we built for a Melbourne consulting firm, running daily. Voice profiles so the output sounds like them rather than like a model, image generation, and industry research feeding the brief. The client is named on request with their permission.

Five people seated around a boardroom table with laptops open while one person stands at the head of the table explaining something.

Most businesses need one or two of these, not five. Which ones, and in what order, is what Discovery is for.

If you are earlier than that, what this work looks like in a Melbourne business and what it costs are the two to read first.

Questions on any of the above, or want to know which of the five your work actually points at? Email hello@anum.com.au.

Start a conversation