In September's job ads, every AI programme we could classify involved the employer's own operations. The eight ads that named a measure mostly chose everyday ones: time saved, errors, cost.

1. Where is the AI work in these ads? In the employer's own operations, for the fourth month running

Twenty-seven September roles described an AI programme belonging to the employer doing the hiring. Twenty-one involve internal work only. Five involve internal work and also tie the work to customers or the community. One didn't say enough to call.

The internal work includes:

• processing staff joiners, movers and leavers;
• OCR order processing;
• triaging production incidents;
• CRM workflows;
• the patient-support team's workflows and pharmacy orders;
• the admin load on frontline health workers.

The five that also tie the work to customers or the community cover contact-centre call quality, customer-acquisition campaigns, service outcomes at a council, participation in a sport, and customer problems at a national organisation.

That's 115 of 118 across four months.

2. What does the AI job look like in a small business? Usually hands-on, under a range of titles

In this report, a small business is one with fewer than 200 staff. Six September employers that size advertised: five small businesses and a small not-for-profit. One of the six, a part-time admin role, asks only for comfort with tools like ChatGPT, so the counts below use the other five, each tied to an AI programme: four small businesses and the not-for-profit.

Four of the five ask the person to build it themselves:

• “design and implement workflow automations”;
• “design and build scripts, integrations, and automation workflows”;
• “get it into production”;
• “design, build and deploy agents in code”.

Three of the five advertised under titles without “AI”: Business Improvement Lead, Digital Transformation Coordinator and Technical Specialist. In two of the five, AI is one duty among several: the role also runs business improvement, or administers the CRM.

Titles swing month to month, from none of them using “AI” to all of them.

The mid-sized employers. All three (200 to 500 staff) advertised under titles without “AI”, with AI as one duty among several: a head of business systems who owns the ERP, a design-applications specialist, and a business analyst.

In four of these eight ads, the same person looks after business systems and also looks for the work worth automating. Two are small businesses (a CRM; finance and sales systems) and two are mid-sized (an ERP; design software).

The large organisations. Of the 12 roles at large organisations with an AI programme (two through one recruiter, possibly for the same client), 10 put AI in the job title.

Our reading: that's a difference in structure, not sophistication. A large organisation can make AI its own job. In four of these five ads, the small business gave it to someone who can build it, sometimes alongside the systems they already look after. That is how a lean team covers more ground with each hire.

Applicants were asked for an example. Three of the five small employers asked to see the applicant's work:

• one wanted examples of AI or automation work “including the tools used and results achieved”;
• another said applications “without a build example won't be considered”;
• a third asked for one workflow or agent: “the problem, what you built, and what changed”.

3. How will they know it's paying off? The specific ones use everyday numbers

Eight of the 27 programmes name something specific to measure. Ten more use phrases like “measurable business value” without saying what that is.

Six of the eight name more than adoption: time saved (one sets a target of five hours a week), error reduction, resolution times, cost per use case (licences, compute and staff time), and service quality. The other two track adoption, one of them alongside licence and credit consumption.

Sound familiar? First, are people using it? Then, what is it saving? Where July's ads named a measure, it was mostly the first question. Seventeen of its 52 ads named any measure, and most were about adoption: active users, usage, feedback. September's specific measures lean towards the second question: time, errors, cost. The most detailed ads track both.

Our reading: when AI is new, the first check is whether people are using it. Once the invoices arrive, the question becomes what it's saving.

Hand-run searches can't tell a trend from a difference in the ads we happened to find, and August's ads already named both kinds of measure. July counted any measure across all 52 ads; September counts specific measures in 27 programmes.

Three set the number before anything is built:

• A group transformation role defines each use case with “a baseline measure, a target measure, a data source, and a named business owner”. It sets the pass mark “before build starts so the result is judged against a number agreed in advance”.
• A small health-services business tells its automation lead to “establish a baseline before you start, then track what changed”. It says success “means what's still running six months later, not how many things were built”.
• A small software business sets a twelve-month goal of staff “saving at least five hours a week on average”, and wants results reported “including where the forecast was wrong”.

4. Who looks after it once it's built, and what keeps it safe? In several ads, the builder hands it over, and the checks are written down

Several September ads make the handover part of the build role:

• one moves each workflow to “its business owner with a run book, a support path, and reporting”;
• a small business hands automations over “once they're stable, documented and the team knows how to run it”;
• a large national organisation requires its business, product, technology, data and service owners to be named;
• a large logistics employer wants automations “used and trusted, not shelved”.

The roles that build and run these systems name specific controls:

• a human review point, an escalation path, and a fallback for when the system is unavailable;
• go/no-go, stop and rollback criteria, and stopping work that misses its threshold at pilot;
• guardrails, permissions, logging and monitoring for agents (a small business);
• approved tools and approved data (one of the three ads is a small business);
• the judgement to know where AI should be applied “and where it should not”, and to recommend what to “build, buy or leave alone” (the second from a small business);
• the rules the business already works under: privacy law, security certifications, sector regulation (between them, two small businesses name all three).

Two employers put the balance in their own words. One applies its architecture, data, cyber, privacy and Trusted AI requirements “proportionate to risk and operational consequence”. Another asks the role to bring project and change-management discipline with “sufficient structure to support governance and track progress, without adding process or slowing delivery”.

5. Where are other businesses up to? All four stages, again

Planning, piloting, rolling out and running all appear again, as they have in every edition. On the same capture day we read ads from employers scoping where AI could help, and from employers hiring someone to run and support AI platforms already in use.

6. The rest of the September sample

• Who was advertising, across 39 roles:
  • six employers under 200 staff: five small businesses and one small not-for-profit;
  • three mid-sized businesses (200 to 500 staff), two close to the 200 line and one that may be above 500;
  • five consultancies;
  • two AI product companies;
  • 14 roles at large organisations (two through one recruiter, possibly for the same client);
  • nine we couldn't size.
• AI was central to 25 roles and one duty among several in 12. In the other two, AI appeared only in what the employer wanted from the candidate. Twenty of the 39 titles include “AI”.
• Eight roles were set aside because the employer sells AI delivery or products, or builds AI into what it delivers to customers: five consultancies, two AI product companies, and a defence contractor building AI into the systems it delivers to government.
• Salary stayed mostly hidden. Nine of 40 ads published a figure:
  • $27–$35 an hour for part-time admin support;
  • $60K–$64K for a sales and account coordinator;
  • $70K–$90K for an IT support and AI lead;
  • $100K–$120K for an AI and automation developer;
  • $115K–$129K for a senior AI and automation analyst;
  • $140K–$150K for a 12-month AI and automation product manager role;
  • about $160K for a practice manager;
  • $180K–$200K plus equity for a strategy and operations lead;
  • up to $785 a day for a six-month AI platform contract.

  The one small-business figure is the part-time admin role.

• Named tools: Microsoft Copilot and Azure in six ads each, ChatGPT or OpenAI and Power Platform or Power Automate in five, Claude in four.

The composite ad of the month

This composite combines duties from several September ads; it is not one employer's ad. It leans on the few ads that spelled out their measures and controls. Most ads said less about measures and controls.

How the advice builds, edition by edition

The measurement advice has repeated since July. September's ads add the target and the handover, and show what the hiring advice looks like in practice.

1. Cost the work before you write the ad (June). Map the workflows and what they're worth in hours saved and risk reduced.
2. Start with the list (July). Write down what AI is already in use, what goes in and out of it, who owns it, and what happens if it's wrong. That list still comes first. September's ads describe the steps after it.
3. Set the baseline before rollout (August). Use the measure you already run the business on: hours, cost, quality, rework.
4. Agree the number before the build, and plan the handover (September). Baseline, target and owner before anything is built. A run book and a named owner at handover. Stop and rollback points agreed before go-live.

Put together: list what you use, cost the work, pick one workflow, write its number down, build or buy, hand it over, then check the number.

What the hiring advice means now. Earlier editions advised favouring experienced, hands-on people over junior hires. September's small-business ads show what that experience looks like: four of the five asked for experience building or delivering this kind of work, and three asked to see an example.

What this means if you run a small business

• Start with a system you already own. The CRM, the ERP, the order inbox, the support queue. Each already has an owner and a number you watch.
• Write the number down before anything is built. Baseline, target, owner. Check it again once people are using the change, and keep a note of where the forecast was wrong. A costed automation plan makes that comparison concrete.
• Decide who runs it after the handover. A run book, a support path and a named owner. A builder or a vendor can hand it over; the business keeps it running.
• Keep the guardrails proportionate. A human check where the output matters, a short list of approved tools, a way to stop or roll back, and the privacy and security rules you already follow. Add more when the risk justifies it. The AI readiness assessment can help you take stock.
• Look first at who already looks after your systems. In four of September's eight small and mid-sized ads, the employer is hiring one person to look after the CRM, the ERP or the design software and to find the work worth automating. If someone in your business already does that job, start with them. Look for curiosity too: 30 of 183 ads across four editions asked for someone curious or willing to learn, at every level of role. It's a trait employers hire for, and one you can spot in your own team.
• If you hire, hire someone who has built this before, and ask them to show it. Four of September's five small-business AI roles ask the person to build it, and the same four ask for experience building or delivering this kind of work. Three asked to see an example. Ask for one thing they've built and what it changed. Fractional technology leadership is another way to bring that experience in.

Cite this report

GraftPoint SME AI Hiring Index, September 2026 edition. Manual sample of 40 AI and automation job-ad captures with a Melbourne location filter, 3–28 September 2026 (39 distinct roles). Of 27 programmes belonging to the hiring employer, all 26 classifiable programmes involved internal work: 21 internal only, and 5 that also tie the work to customers or the community. One was unclear. Eight programmes named a specific measure, and six of those named measures beyond adoption. The sample included planning, piloting, rollout and running stages. Among roles with an AI programme, three of five at employers under 200 staff advertised under titles without “AI”, and 10 of 12 roles at large organisations used AI titles. Methodology: graftpoint.com.au/sme-ai-hiring-index/methodology/