Can AI replace this task?

23 Sept

A practical guide for busy Australian small-to-medium businesses and not-for-profit organisations

It's 4:45pm on a Friday, and the office manager at a 22-person engineering firm in outer Sydney is still re-keying numbers from supplier invoices into a spreadsheet, because the accounting software doesn't talk to the project management tool. Two desks over, someone is manually copying names and email addresses from a sign-up form into a mailing list, one row at a time. This is not their favourite part of the job for either of them. Both of them have, at some point this year, muttered some version of ‘surely a computer can do this by now’.

Artificial intelligence tools have moved from novelty to mainstream in a remarkably short space of time, and the tools themselves have become far more capable of handling everyday business tasks. But ‘AI can do a lot of things’ is very different from ‘AI should do this specific task in this specific business'. This article works through the questions businesses and organisations can ask as a practical, task-by-task way of determining what 'can' versus what 'should' be managed with AI.

Why the automation question is suddenly everywhere

A few years ago, ‘automation’ in a small business usually meant a macro in Excel or a basic rule in an email inbox. Today, the tools available to a five-person retailer or a fifty-person not-for-profit look a lot more like what a large enterprise IT department might have had access to a decade ago, at a fraction of the cost and with no coding required. Generative AI tools can now draft correspondence, summarise long documents, answer routine customer questions, transcribe meetings and generate first-pass reports, quite often using the same software suite a business is already paying for – they just don’t know it, or simply don’t know how to use it.

This is an important shift for Australian small-to-medium enterprises (SMEs) in particular, because so much operational load in smaller organisations sits with only a few generalist staff rather than having inhouse dedicated specialists who are experts in this stuff. The ATO's Small Business Benchmarks have shown that administrative overhead is one of the more persistent drains on profitability for businesses under 20 staff, and it's not because people are inefficient. It's because they're doing ten jobs at once. When a genuinely capable tool becomes available to take even one of those ten jobs off someone's plate, it's worth taking a good hard look into it.

It's an area where the goal posts keep shifting. What a tool couldn't reliably do twelve months ago, may now handle those same tasks reasonably well, and the reverse can also be true if a business tries to use a tool beyond what it's actually built for. Treating ‘can AI replace this task’? as a one-off question, rather than something that should be asked every 6-12 months, is something many organisations fail to do.

What ‘AI’ actually means at the task level

Part of what makes this question hard to answer is that AI gets used as a single label for a bunch of different things. It's worth breaking a task down before deciding whether AI has any role in it at all.

Some tasks are largely about pattern recognition and repetition: sorting invoices into categories; flagging unusual transactions; matching a resume against a job description. These are often well suited to automation.

Automation is the use of technology to perform repetitive or rule-based tasks automatically, without necessarily using AI.

Other tasks involve producing a first draft of something: a job ad; a social media post; a meeting summary; a grant acquittal report. These sorts of things are where a generative AI tool can be a very useful thing to have working for you…provided a person still reviews and finalises the output.

Generative AI is technology that can create new content such as text, images, audio, video or code which has been based on patterns it has learned from existing data.

A third category involves judgement calls that depend on context an AI tool simply doesn't have access to, for example, reading the room in a difficult staff conversation, deciding how to handle a long-standing client's complaint, or weighing up a values-based decision on behalf of a community the organisation serves.

Judgement calls are decisions that rely on human context, experience and information that an AI tool does not have the access to or the ability to fully understand.

Separating a task into ‘which part is repetitive,’ ‘which part is drafting,’ and ‘which part is judgement’ tends to produce a much clearer answer than asking ‘should AI do my bookkeeping’ as one big, vague question.

Tasks where AI tools can add incredible value to SMEs and NFPs

Across many Australian SMEs and not-for-profit organisations (NFPs), a fairly consistent shortlist of tasks has emerged as genuinely well suited to AI-assisted tools right now.

First-draft writing is one of the clearest examples: newsletters, job descriptions, social media captions, meeting minutes and internal policy documents can often go from blank page to a workable first draft in minutes rather than hours, with a human editing for accuracy and tone before anything goes out the door.

Data summarisation is another strong candidate. Long customer feedback surveys, lengthy PDF reports, or a year's worth of meeting notes can be condensed into a readable summary far faster than a person reading through them line by line - useful for a committee preparing for an annual general meeting, or a manager trying to get across a topic before a client call.

Routine customer and stakeholder queries are increasingly handled, at least in the first instance, by AI-assisted chat tools that can answer common questions from a knowledge base, freeing up staff time for the enquiries that actually need a person. Scheduling, basic data entry from structured documents, and simple translation or transcription work also tend to sit comfortably in this category. In each of these cases, the pattern is similar: the tool handles volume and speed, and a person still checks the output before it's relied upon.

Tasks where a human still needs to be in the loop

On the other side of the coin, plenty of tasks are still firmly a human responsibility. Anything involving a final financial, legal or compliance decision: signing off on a set of accounts; agreeing to a contract's terms; deciding how to respond to a Fair Work query - these still needs a qualified person taking responsibility for the outcome, even where an AI tool helped prepare the background material.

Similarly, anything that depends on a relationship or reputation should not rely on AI as the final step in the process. Any organisation that truly values its customers would not remove human-to-human connection or use AI as its first point of written or verbal contact. A donor thank-you call, a difficult performance conversation, or a media response during a sensitive period are all situations where the ‘human touch’ isn't just a nicety, it's the very point of human-to-human interaction. Not-for-profit organisations in particular might find that donor and volunteer relationships will suffer when communication starts to feel automated or generic, even if the underlying content is accurate.

There's also a category of tasks where the risk of getting it wrong is simply too high to hand over without very close inspection. Anything touching client privacy, sensitive personal information, or a decision that could disadvantage a vulnerable person are prime examples. In these cases, many organisations are finding that AI tools can still help with drafting or research, but the final decision and final wording must remain with the person who understands the full context.

A simple framework: the four-question test

Rather than treating every task as a whole new conversation, it can help to run them through the following set of quick questions to help answer ‘should AI be doing this?’:

Is the task repetitive and rules-based, or does it require judgement about a unique situation?
What happens if the output is wrong - is it a minor inconvenience, or could it cause financial, legal or reputational harm?
Does someone have the time and expertise to properly check the AI's output before it's used, or would the checking process end up taking just as long as doing the task manually?
Does the task involve information that's private, sensitive or subject to specific regulatory obligations (client records, health information, a child's details, financial account numbers), which may limit which tools can be used and how.

Running a task through these four questions tends to sort things out fairly quickly into ‘yes, worth trying,’ ‘maybe, with tight review,’ or ‘not yet, or not without more safe-guards’. It's a simple framework business owners may find useful because it doesn't require any technical AI knowledge. It's really just a structured way of thinking about risk and time, applied to any new kind of tool.

Getting started without disrupting your team

For SMEs and NFPs still finding their feet with AI, a staged approach tends to work better than an all-or-nothing rollout. Many businesses find it useful to start with a single, low-risk, high-volume task - for example: summarising customer feedback surveys; drafting first-pass job ads; tidying up a mailing list; or transcribing team meetings - and then let staff get comfortable with reviewing and correcting AI-generated output before expanding further.

This also gives a business or organisation a chance to work out what training staff might need, and to identify any team members who are uneasy about the change, which then needs some ‘change management’ inputs and is worth taking the time to set this up right. This change management might be as simple as a short staff briefing on why the tool is being introduced and what it isn't replacing, just to help them understand.

It's also worth exploring what data and privacy settings any AI tool offers before it touches business or client information, since not all tools handle data the same way, and a not-for-profit bound by ACNC governance obligations or handling sensitive client information may have additional considerations around consent and data storage that a purely commercial business wouldn't need to think through in the same way.

Not-for-profit organisations weighing up AI-assisted tasks often have an extra layer to think through: donor and beneficiary trust. Many funders and members expect a higher degree of personal engagement from an NFP than a customer might expect from a commercial business, so it's worth taking a little extra care about whether AI-assisted drafting holds a place in the organisation.

It's also worth checking any AI tool's data handling against your organisation's privacy policy and any regulatory authority's governance standards, particularly where client, beneficiary or volunteer information is involved, and keeping the board or management committee informed of which tasks are AI-assisted as part of normal governance reporting.

Keeping a simple, shared register of which tasks are using AI assistance, and who is responsible for checking that output, can prevent the problem where AI ends up doing more than it should.

Finally, setting up an annual audit for the organisation’s AI usage is highly recommended. A task that wasn't suited to AI assistance a year ago may be worth another look today, and the reverse is also true - a tool that seemed to work well initially may need tighter review as it is now being used for higher-stakes material.

About the author

Wendy Mayer is the founder of BusinessRx, providing hands-on operational support to SMEs, NFPs and purpose-driven organisations across Australia. Over 25 years spanning business ownership as well as in business and operations management roles, she has worked alongside SME and not-for-profit leaders to strengthen their organisations. She has also facilitated governance functions in highly regulated corporate environments - experience that brings a depth and rigour smaller organisations rarely have access to. Wendy partners with owners, leaders and boards to create systems and processes that work. BusinessRx is about 'doing', not just advising - stepping into the gaps to get things moving and making businesses easier to run.

To learn more about us and what we do, you are invited to explore our website further: businessrx.com.au

Wendy Mayer https://BusinessRx.com.au