AI Adoption Challenges for Australian Small Businesses: What Gets in the Way

Describe the last point where the work progressed and the next point where it stopped. Use an example from this week rather than a general statement that the business needs to be more innovative. A stalled quote, unreconciled invoice or unused training exercise gives the discussion something observable. Treat the table as a diagnostic aid, not a test that produces a definitive score. Several problems can coexist. Choose the one preventing the next safe step, assign an owner and agree what evidence would show it has improved. That creates a smaller and more useful scope for external help, too.

A business can afford software and still lack the capacity to implement it. Someone has to collect examples, explain the exceptions, test draft outputs and answer questions from the rest of the team. If that work is added to an unchanged client workload, progress depends on spare time that may never appear. In a public account of an unsuccessful accounting automation attempt, Box Advisory owner Davie Mach describes staff lacking capacity to teach and check a tax-review system while doing client work. They reverted to manual processing. This is one owner’s experience, not an estimate of failure across accounting firms. Give the pilot a real allocation of time. For an illustrative trial lasting 14 days, that might mean a process owner has scheduled preparation time, a reviewer has a defined checking window and participants have permission to pause ordinary work for practice. The allocation must suit your workload; there is no universal number of training hours. Count that time in the project cost. At the review, ask whether it is declining as the team learns or staying high because the task remains difficult to check. Persistent checking effort may be a reason to narrow the task rather than buy more licences.

Many tasks depend on information that lives in somebody’s memory. The owner knows which jobs attract a surcharge. A senior accountant knows when a client needs a different request. A service manager knows which enquiry should bypass the standard queue. An AI tool cannot consistently apply a rule the business has never made available. The same accounting owner’s proposal example identifies pricing details missing from meetings and difficulty connecting systems. The lesson is to inspect the inputs before assuming that a different model will fix the output. Take a recent completed task and work backwards. Which facts were required? Where did each fact come from? Which judgement changed the result? Identify missing fields, conflicting records and decisions requiring professional review. Do this with the person who performs the work, not just the software supplier. Use our AI workflow documentation template to record those dependencies. Start with approved or fictional examples while assessing the process. Choosing a paid tool alone does not settle permission to use client information. Agree the information boundary before using live records, then test whether the task is suitable at all.

A rollout asks people to change something they already know how to do. They may worry about errors, extra checking, loss of control or what management will expect after time is saved. Calling every concern resistance makes it harder to learn which part of the proposed workflow is weak. In a National AI Centre interview, Lee Hickin relays a repair-business example in which a promising quotation experiment was abandoned after staff felt excluded. It is a second-hand example, useful for the question it raises rather than proof that participation alone guarantees adoption. Include the people doing the task before choosing the final process. Ask them to identify a difficult example, explain what they check and demonstrate the quickest reliable way of completing the work today. Invite someone with reservations to help define the acceptance criteria. Their concerns can become tests rather than a debate about enthusiasm. Agree how released capacity will be used. Better client attention, less overtime, a smaller backlog and additional sales capacity are different outcomes. Our small-business AI change management guide sets out a practical way to run that conversation and support the team after an initial workshop.

The visible AI step is often only part of the process. Drafting a client email faster helps little if somebody must retype the client details, correct unsupported statements, find the attachments and update three records afterwards. A local time saving can create work elsewhere. Map one item from arrival to completion. Include the queue before the task, the work itself, review, sending and record keeping. Ask where an item waits and what information gets copied. Distinguish a useful drafting assistant from an automation that changes records or contacts customers. The latter needs controls around those actions, not just a good prompt. In an illustrative workflow, a five-minute draft followed by fifteen minutes of checking may be worse than the previous fifteen-minute process. Those figures are an example calculation, not a measured customer result. Measure your own complete task, including failed attempts and the time taken to recover from them. Keep the first trial narrow enough to inspect. You might allow draft creation and manual approval while leaving sending and system updates unchanged. If the repeated copy-and-paste step is the real bottleneck, consider plain automation alongside AI rather than asking the model to improvise around disconnected systems.

A small business does not need to copy a large enterprise’s organisational chart. It does need answers to practical questions: who can approve the workflow, who checks difficult cases, who changes the instructions and who pauses it when something goes wrong? One person may hold several responsibilities, but they still need time and authority. An AI Collab discussion aimed at 10–30-person accounting and legal firms recommends designing workflows together with the people who understand the process. That is an adviser’s recommendation for a named audience, not evidence that every firm in that size range has the same needs. Record the ownership alongside the process. Identify the ordinary operator, the reviewer and the person responsible for updates. Define a fallback that somebody can actually perform. If the workflow depends on a particular employee’s personal account or undocumented setup, the handover is incomplete. Business size is only a rough guide to available capability. A 20-person firm can have a capable technical owner; a much larger business can lack time for this project. When considering outside support, assess the skills and capacity available for the workflow rather than assuming a dedicated AI department appears at a particular staff count.

Match the engagement to the work needed next. If people need supervised practice on an agreed task, start with team training. If the task is poorly understood, start with discovery and documentation. If it requires dependable movement between systems, scope an implementation. If it works but depends on one person, prioritise ownership and handover. Ask what your team must contribute and what the provider will deliver. A supplier cannot reconstruct every unwritten business rule from access to software. Equally, a training session does not automatically include integrations or ongoing support. Put those boundaries in the scope before comparing prices. If you are unsure which category applies, bring one recent example and the steps currently involved. Explore team workshops for practical learning or start the workflow audit when the priority task is still unclear. You should leave the next conversation with a smaller, reviewable decision.

What are the main barriers to AI adoption for Australian SMEs? Public business discussions repeatedly raise staff capacity, uncertain use cases, missing process information, checking effort, disconnected systems and unclear ownership. This guide uses those as diagnostic themes. It does not claim a representative national ranking or prevalence estimate. Should we buy more tools if staff are not using AI? First find out why the existing tools are unused. People may lack a relevant task, permission to use the necessary information, time to practise or confidence in the output. More software addresses only some of those problems. Does a small business need an internal AI team? It needs accountable people with enough capability and time for the work being attempted. Some firms can manage that internally; others need training, implementation help or ongoing support. Headcount alone does not determine the answer. Where should we start if our pilot has stalled? Choose one recent attempt and identify the point where it stopped. Check inputs, review effort, staff involvement and handoffs before changing tools. Agree the smallest change worth testing, who owns it and when you will review the result.