Of the 31 classifiable AI programmes in July's job ads, 28 do work inside the business. Three are purely customer-facing.

1. What the AI is for: mostly the work inside the business

The employer-adoption sample contains 32 programmes. One is too vague to classify by audience. The other 31 sit across 30 employers; one advertised two roles. What the ads say the AI is for:

The three purely customer-facing programmes: a global remarketing business governing customer-facing digital products, a car dealer group running AI that talks to inbound leads, and a youth mental health institute running two clinical trials.

The internal work sits on jobs that were already there: finance and people teams, investment research, workforce planning and forecasting, legal practice, test coverage, quoting and invoicing.

In June, none of the 30 classifiable programmes was customer-facing only; six of the 33 employer-adoption programmes combined internal and customer-facing work, and three were unclear. These are programmes described in job ads, not proof that the systems are running: 18 of July's 32 were at planning or piloting stage.

2. Where they are in the cycle: all four stages at once

We placed every described programme at the stage the ad's own words support.

There's no single answer to “how far along is everyone” — all four stages are happening in the same month, in the same city. A family-owned retailer is writing “lead our push to use AI” while a dealer group is already coaching an AI that talks to its customers and checking its accuracy against its best salespeople.

Both rows are restated after the September re-read. Enough programmes moved between adjacent stages to change the shape of the table. The useful reading is that all four stages appear in both samples, not that the market moved between them.

3. What they're measuring: mostly whether people are using it

Thirty-five of the 52 ads name no measure at all. Of the 17 that do, most measure adoption: monthly active users, usage, impact, feedback loops. The few measuring whether it pays use plain operational numbers:

• overhead per student, and cost to serve per campus and per program
• cost to serve across a fulfilment network
• faster onboarding and fewer support tickets
• lead drop-off and lead-to-appointment conversion
• “hours transformed”, feeding decisions on resourcing, pricing and workforce design

The most specific measurement instruction in the sample isn't a metric. It's an inventory. A superannuation administrator is hiring someone to maintain a register of automation, analytics and AI tools — ownership, usage, risk classification and compliance status — because AI is arriving in the business whether it's centrally delivered or locally built.

That register is the first step of any AI governance worth the name, and the same artefact tells you what to measure. You can't report an outcome for a system nobody has listed, and the owner field puts a name against the number.

A lighter version is here too: a 300-person education business gives its operations manager authority to require an owner, a baseline and a measure before any new process or tool goes live — and to stop it if it doesn't have them. The same discipline at a weight a lean business can carry.

4. What guardrails show up in roles that do the work

Several ads put practical controls in the hands of the person who builds or runs the system, rather than reserving them for a governance role. The examples include:

• A human check. Safe, secure and on-policy AI use, with human review checkpoints.
• An approved list. Approved AI solutions, approved enterprise platforms and services.
• A gate with teeth. The pre-go-live check above, held by someone with the authority to say no.
• The security obligations already on the business. A family-owned retailer's technology manager holds PCI-DSS and cyber posture in the same seat as AI adoption; a facilities group's BI specialist holds row-level security and access controls alongside its Copilot rollout.
• Accuracy, where AI already faces the customer. The dealer group's ad asks for AI that responds quickly and accurately, reviewed conversation by conversation.

From an 11–50 staff distributor, the constraint most SME operators will recognise: build operating rhythms, reporting and controls “without creating unnecessary bureaucracy”.

5. The rest of the sample

• The small businesses were again filed under operations. The four SME ads surfaced under “technology manager”, “technology officer ai”, “operations manager ai” and “ai automation”. Our two highest-yield AI searches, “AI enablement” (14 ads) and “AI lead” (9 ads), returned none. One of the four never uses the word “AI” at all.
• AI fluency appeared as a personal capability. Four ads asked for it: a business development manager, a QA engineer who'd be tested at interview, a growth operator and a product manager. In two, AI appears nowhere in the duties.
• Employers hired for the judgement not to use AI. Four unrelated ads asked candidates to know when training, workflow redesign or ordinary software was the better answer.
• The same tools turn up across very different sizes. n8n appears in an 11–50 staff distributor, a scaling logistics business and the AI engineer role at a multi-billion-dollar retail group. Copilot and Claude are named together in four ads as one working environment.
• Salary stayed mostly hidden. Eleven of 52 published a figure: $70–80K for an SME operations generalist, $80–100K for an AI engineer building agents, $140–200K for AI leads and governance managers.

The composite SME ad of the month

Composited from the four SME-flagged ads and recurring patterns. No single employer is represented.

That's six functions in one seat with a direct line to the CEO. Covering that much ground is what a lean business runs on — a rational response to scale, and that reporting line is an advantage a large organisation can't easily buy. The decision that matters is how to resource it: seniority with hands-on operational know-how over the cheapest available hire.

What this means if you run a small business

• Starting inside the business puts you in good company. Purely customer-facing AI is the exception here, and in all three cases someone is paid to look after it daily. If you're weighing a customer-facing chatbot against fixing your quoting process, the early movers in this sample have mostly made that call one way.
• Start with the list, not the framework. Before a policy or a committee, 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 gives you a starting point for risk and measurement. The AI readiness assessment can help you take stock.
• Measure the number you already have. The employers measuring payback use cost to serve, support tickets, onboarding time and conversion, not AI metrics. Write down the operational number you already watch before you change anything, and check it after. A costed automation plan makes that comparison concrete.
• Right-size the guardrails on purpose. A human check, an approved tool list, an owner with a baseline. Which controls earn their keep right now is a senior judgement call, and part of fractional technology leadership.

Cite this report

GraftPoint SME AI Hiring Index, July 2026 edition, restated September 2026. Sample of 52 Melbourne-area AI and automation job-ad captures from 17–27 July 2026. Of 32 programmes in the employer-adoption sample, 20 were internal only, 8 combined internal and customer-facing work, 3 were customer-facing only and 1 was unclear. Methodology: graftpoint.com.au/sme-ai-hiring-index/methodology/