What does an AI consultant actually do?
People type "what does an AI consultant do" into Google right after a colleague mentions Claude in a meeting and nobody in the room can explain what happens next. An AI consultant chooses the right AI tool for a business, trains the people who will use it on their own work, and manages the licences afterwards so the habit survives past the first month. That is the whole job in one sentence, and everything below breaks it into the weeks, the deliverables and the price tag that make it real for an Australian business weighing Claude in 2026.
The confusion is fair, because the title covers three genuinely different jobs depending on who is hiring. A logistics firm wants someone to teach staff how to turn a two-hour report into a twenty-minute one. Over at a law firm, the ask is different: connect Claude to the software that stores client files, without breaking confidentiality along the way. Both call the person an AI consultant, and both mean something a little different by it.
What does an AI consultant do day to day?
An AI consultant spends most of a working week on three things: picking the tool, teaching the team and keeping the platform running once the excitement fades. Picking the tool means recommending Claude, Anthropic's AI assistant, over a dozen near-identical options, based on what the team actually needs rather than which one has the loudest marketing. Teaching the team means sitting people down with their own invoices, contracts or reports, not a generic slide deck. Keeping the platform running means owning licences, permissions and the questions that arrive three weeks after launch, once the first genuine problem shows up.
The mix shifts depending on where a business sits. Anthropic's own research on Australian usage found that 46 percent of Claude conversations in Australia are for work, ahead of the 7 percent used for coursework and roughly level with personal use1. A consultant spends the bulk of a working week training and managing rather than building, because most Australian teams are already inside a Claude conversation. The job is closer to a coach who shows up with a plan than a builder who disappears for six months.
How does a Claude rollout actually run, week by week?
Most Australian businesses picture an AI consultant showing up once for a workshop and leaving a slide deck behind. A real rollout runs longer than that and follows a fixed order, because skipping a step is exactly where adoption quietly dies.
satori-launch: scoping call to skills going live
The first call sizes the team and picks which five processes are worth automating first, a conversation that takes under an hour and needs no preparation from the business. A two-hour webinar follows for the whole company, giving everyone the same starting point on what Claude is and is not. Then comes a full-day workshop at the business's own office, built entirely around real documents pulled from that morning's inbox rather than a demo dataset. By the end of that day, up to five ready-made skills go live across the team, each one a saved Claude prompt set up for one specific recurring task. A 30-day check-in gets booked before anyone leaves the room.
The 30-day check-in catches the skills nobody opened and the ones running twice a day without anyone mentioning it. A second check-in at 90 days confirms whether the habit survived the first quarter or quietly slipped back to the old way of doing things. Not every skill survives that check. A typical 90-day result keeps three of the five skills in daily use and drops the two that turn out to duplicate work already handled inside existing systems, an overlap the workshop cannot always see coming.
What survives after the consultant leaves?
The honest test of an AI consultant is what remains three months after the invoice is paid, once nobody is checking in weekly anymore. Four things should still be standing: people who can use Claude without a manual, and prompts or skills that keep producing the same output on demand. Two others matter just as much: licences that someone still actively manages, and any connections into internal systems that keep working unattended.
satori-claude: the licences someone still manages
Trained people are the least visible deliverable and the one that fails first if it is skipped, because a licence with nobody confident enough to use it is money spent for nothing. Working skills are prompts saved once and reused, the difference between typing the same instructions every Monday and clicking a saved one. Licences that someone still manages cover renewals, permissions and the support ticket that shows up when a new starter joins. satori-claude handles all three for A$850 to set up, then A$15 per person a month, on top of the Claude licence itself.
MCP: the connector that outlives the consultant
The fourth deliverable, system connections, relies on MCP, short for Model Context Protocol, an open standard that lets Claude read and write inside a business's databases and tools instead of stopping at a chat window2. Once an MCP connector into a CRM or accounting system is built and tested, it keeps running long after the consultant has moved to the next client. That is the whole point of building it in the first place.
What does an AI consultant not do, and when do you skip one?
An AI consultant does not write the strategy document nobody reads, and does not sit in the business indefinitely as a substitute for someone on the team learning the tool properly. The job ends at training people and managing what was built, not owning decisions that belong to the business itself.
Two situations rarely justify hiring one. A team of two or three people running free-tier experiments on Claude has no handover problem yet, because everyone already sees what everyone else is doing without a formal process. The second is a business with no real appetite from leadership to change how a process gets done, because a workshop only sticks when someone senior insists the new way replaces the old one. Nothing about the tool fixes that on its own.
What does an AI consultant cost in Australia?
Australian pricing for this work splits into two camps: an hourly rate with no ceiling, and a fixed number published before the first call. XLev, a transparent AI advisory firm in Australia, lists senior advisory time at up to A$599 an hour3, a figure that turns a single afternoon of questions into thousands of dollars before any training happens. satori.'s satori-launch runs A$990 per person, once, from a minimum of 8 users, which puts a full launch for an eight-person team at A$7,920, published on the page rather than behind a form.
Training bought on its own, without the full launch package, runs from A$490 per person for a two-hour session up to A$990 per person for a full day with role-specific tracks. All four formats sit on the Claude training page. The eight questions worth asking before signing with anyone are in our checklist for choosing an AI consultant in Sydney. An AI consultant in Sydney or anywhere else in the country pays the same per-person rate under this model, because the price sits on team size rather than postcode. Businesses further along already using Claude but wanting it managed properly usually skip straight to a Claude consultant arrangement instead of the full launch.
Why now, and why Claude specifically?
Anthropic opened a Sydney office in March 2026, its fourth in the Asia-Pacific region, citing strong demand from Australian and New Zealand businesses as the reason4. Australia already ranks fourth globally in per-capita Claude usage, according to Anthropic's own figures4, well ahead of most economies its size. That explains why so many businesses are hiring someone to make sense of a tool their own staff have often already started using without asking permission.
Take the last report your team wrote by hand this week, paste it into Claude, and instruct it: "turn this into a first draft, and flag anything you are unsure about." Time how long that draft takes, then compare it against the version you would have typed from scratch. The whole comparison fits inside fifteen minutes, and nobody else needs to be in the room.
Sources
Footnotes
-
Anthropic (2026). How Australia uses Claude. https://www.anthropic.com/research/how-australia-uses-claude ↩
-
Model Context Protocol (2026). What is the Model Context Protocol?. https://modelcontextprotocol.io/docs/getting-started/intro ↩
-
XLev (2026). Pricing. https://xlev.ai/ ↩
-
Anthropic (2026). Sydney will become Anthropic's fourth office in Asia-Pacific. https://www.anthropic.com/news/sydney-fourth-office-asia-pacific ↩ ↩2