Notes
Most writing about AI is either a sales pitch or a science lecture. These are the answers we give business owners in the first conversation: what to automate first, what it costs, and how to tell whether it worked.
A chatbot answers. An agent acts. That one distinction decides what you can safely buy, what it should cost, and how carefully it needs supervising.
Where does your data go, who can see it, and is it training somebody's model. The questions that decide whether an AI build is safe, and the answers you should refuse to accept.
A firm that bills for its judgement has a specific problem with automation: the billable hour and the saved hour point in opposite directions. Where AI helps a practice, and where it quietly cannot.
Most people have already met bad support automation and they remember it. Here is the difference between a machine that helps your customers and one that stands between them and you.
Nobody enjoys chasing money, so it gets done late, inconsistently, and by the person least suited to it. That is exactly the shape of work a machine handles well, provided a person still holds the awkward conversations.
Running several sites means the same problem happening in several places at different speeds. AI is unusually good at that shape of problem, if you build it centrally and roll it out slowly.
Quotes that go out the same hour win more work. AI can read a supplier quote line by line, benchmark it against real pricing, and flag the padding before you sign it.
Screening is the most automatable task in recruitment and the most legally exposed. What AI does well in a consultancy, what it must never decide, and how to keep it defensible.
The whiteboard works until it does not. What a machine can do about the daily run, where it should stop, and why the person who currently does the scheduling is the most important part of the build.
A language model will produce a confident answer whether or not it has one. That is a property of the tool, not a bug to be fixed, and a well built machine is designed around it rather than hoping for the best.
A machine you cannot move, cannot inspect and cannot run without your supplier is not an asset, it is a subscription with your process inside it. Six questions to ask before you sign.
An AI receptionist answers the calls you are currently missing, quotes from your real prices, and books straight into your calendar. Here is what it does well, what it should never do, and what it costs.
A subscription and a built machine solve different problems. One helps a person work faster. The other does the work whether anyone is at the desk or not. How to tell which one you are actually asking for.
Real numbers, not ranges with four zeros of daylight in them. The true AI automation cost in Australia for a small to mid-sized business, what drives the price up, and what you should refuse to pay for.
Nine questions that separate a builder from a reseller, plus the four answers that should end a conversation. A checklist you can use on anyone, including us.
Most AI investments are never assessed, because the numbers needed to assess them were never collected. Three measurements taken before you start, and a page you can read in a minute afterwards.
Most AI pilots end with everyone agreeing it was interesting and nobody able to say whether it worked. Here is how to run one that returns a verdict instead of a feeling.
Every useful machine has a point where it should stop and wait for a person. Deciding where that point sits is the most important design decision in the build, and the one most often skipped.
Most AI builds stall on the same thing: the information the machine needs is scattered, inconsistent, or living in someone's head. Here is what preparing data actually involves, and why it pays off even if you never build anything.
Buying the tool is the easy part. Whether it gets used well depends on a handful of hours of proper training and one honest conversation about job security, held before the machine arrives rather than after.
The first AI project decides whether there is a second one. A practical way to pick the one process worth automating, and the four kinds of work that are wrong to start with.
The failures are boring and repetitive. Undefined scope, no baseline, nothing written down, and a machine nobody owns. Five causes, and the habits that prevent each one.
The question every owner asks quietly and few consultants answer straight. What AI actually takes off a team, what it cannot take, and how to introduce it without wrecking the place.
Your staff are already using AI. A policy is not about permission, it is about drawing the line between what goes in and what must never. One page, six clauses, written before something goes wrong.