Blog

Which Tasks to Automate First in a Small Firm

Start with repetitive, rules-based work that eats staff hours. Here is how to score tasks, map what your team does and pick the first workflow to automate.

11 minute read · Published 22 November 2026 · Calibre Forum

Automate first the work that is high in volume, repetitive, rules-based and costly when it goes wrong: invoice reminders, lead capture and follow-up, enquiry routing, and copying data between systems. The catch is that most owners cannot see which tasks those are, because the work lives in individual staff habits. A short anonymous survey of everyone on the team, followed by a simple scoring exercise, shows where the hours go. Our AI readiness assessment is built to give you that picture across your whole organisation.

CalibreforumSee Where AI Can Help Your FirmCalibre Forum’s AI readiness assessment gathers staff insights and delivers a scored report with a clear 90-day action plan.Visit

TL;DR

  • Start with repetitive, rules-based tasks where manual handling causes delays or errors, such as invoicing, lead follow-up and enquiry routing.
  • Score each candidate on volume, repetitiveness, data availability, error tolerance and current cost, using the National AI Centre matrix.
  • Map tasks by talking to the people who do them, not from the org chart. Staff know the workarounds.
  • Pick one clearly defined workflow, measure time saved and errors avoided, then move to the next.
  • Decide who checks the output before anything goes live.

Table of Contents

Why the first choice matters more than the tool

Small firms rarely fail at automation because the technology is weak. They fail because the first project is the wrong size. A flashy, cross-department build stalls, the team loses patience, and the next idea never gets budget. A narrow, boring task that saves four hours a week for six people builds the case for everything after it.

The second trap is guessing. Owners tend to automate what annoys them personally, while the biggest time sinks sit with staff they rarely watch work: the person re-keying job details into the accounting system, the office manager chasing quotes by email. You need evidence of what your team actually does before you choose.

The kinds of tasks that usually come first

A panel of Australian business experts writing for Dynamic Business advises SMEs to begin with repetitive, rules-based work where manual handling causes delays or errors. Their examples fall into three groups:

Area Typical first candidates Why they suit early automation
Finance Recurring invoices, payment reminders Fixed rules, clear trigger (a due date), easy to measure
Sales Lead capture, follow-up emails, appointment booking High volume, same steps every time, leads go cold when ignored
Customer service Enquiry routing, standard replies Predictable categories, fast response matters
Data handling Moving information between systems Same fields each time, errors compound

What these share is a clear start, a clear end and a rule for what happens in between. If a person can explain the task in five plain sentences, it is a good candidate. If it depends on judgement, relationships or knowing the client’s history, leave it for later.

For more concrete illustrations of repetitive service-business tasks, see these workflow automation examples for service businesses.

Count the hours before you choose

The question “how many hours go on repetitive admin” has no honest industry average, because it depends on your systems. What you can do is calculate your own. ADAI News describes manual data transfer as a “copy-paste tax”: the time cost is real, and the cost of errors stacks on top. Its worked example is a 10-person firm where each person does 20 manual transfers a day at 60 seconds each. That adds up to 858 hours a year, nearly $43,000 at average rates.

The formula is simple: transfers per week × time per transfer × 52 weeks = annual hours.

Here is an illustration with made-up numbers for a firm like yours. Eight staff each re-enter job details into a second system about 10 times a day, at 2 minutes each. That is 160 minutes a day, about 13.3 hours a week, or roughly 693 hours a year. Multiply by your own loaded hourly cost and you have the price of doing nothing.

Annual hours from repeated job-detail transfers

The same source advises against trying to fix every copy-paste. Identify the five highest-cost transfer routes and automate those first. That keeps the project small and the payback visible.

Score the candidates with a simple matrix

Once you have a list, rank it. The National AI Centre, the Australian Government body that publishes practical guidance for businesses adopting AI, rates each opportunity on five criteria, each from 1 to 5:

  1. Volume: how often the task happens.
  2. Repetitiveness: how similar each instance is.
  3. Data availability: whether the information needed is already captured digitally.
  4. Error tolerance: how much a mistake matters.
  5. Current cost: the time and money the task takes now.

Totals of 5 to 12 count as low AI fit, and 13 to 25 as high fit. You then plot fit against business impact. The guidance says to prioritize high-fit, high-impact opportunities for rollout and treat high-fit, low-impact ones as quick wins. Low-fit, high-impact processes should first be improved with templates, simple rules-based automation or streamlining.

That last point saves many small firms money. If a process is messy, automating it only makes the mess faster. Fix the steps, then automate them.

A worked example

Say your bookkeeper spends a day a week chasing overdue invoices by hand. Volume scores high, since it happens every week. Repetitiveness is high because the reminder wording barely changes. Data is already in your accounting system. A wrong reminder is awkward but not catastrophic. Current cost is a full day of skilled time. That could easily total in the high teens or more, a high-fit candidate with solid impact. It becomes your first project.

Now compare a task like drafting bespoke proposals for large clients. Impact is high, but each one differs, so repetitiveness and data availability are low. It goes on the list for later, probably after you have simplified your proposal templates.

How to map staff tasks so the list is real

The National AI Centre’s guidance on mapping your processes is practical. Map a process step by step, recording the action, who is responsible, the inputs, the outputs and the time it takes. Talk to staff about workarounds, delays and what would make their job easier. Then note the trouble spots: bottlenecks, rework, waiting, manual data entry and inconsistency.

A workable routine for a firm of 10 to 100 people:

  1. List the recurring tasks by role. Ask each team lead for the ten things their people do every week.
  2. Pick the top five processes by time. Do not map everything.
  3. Walk each one with the person who does it. Write down every step, including the ones that are not in any procedure.
  4. Record the time and the tools. Note which systems are touched and which copies of data are made by hand.
  5. Mark the pain points. Waiting, rework and re-keying are the signals.

If you want a worked method for structuring this kind of mapping across a team, this guide to a business accountability workflow is a useful companion to the steps above.

Where manual mapping goes wrong

Workshops and interviews have a known weakness: people describe the process as it should run, not as it runs. Staff may also be wary of telling a manager they use a personal AI account to draft emails or summarise meetings. The result is a map that misses exactly the unofficial tool use you most need to see.

An anonymous survey of every staff member fixes both problems. That is how our assessment works: each person answers privately about the tools they use and how their work flows, with nothing installed and no individual data collected. Our guide to running an AI staff survey honestly and anonymously explains how to get candid answers, and the shadow AI page covers the accounts, extensions and recorders that tend to surface.

Document just enough before you build

You do not need a binder of procedures. The National AI Centre’s guide to redesigning your workflow recommends a short workflow specification covering:

  • a process flow diagram showing where AI is used;
  • roles and responsibilities;
  • decision rules and escalation paths;
  • quality standards and acceptance criteria;
  • training requirements;
  • technical requirements, including systems and data sources.

One page per workflow is enough. The same guidance suggests testing before launch with paper prototypes, stakeholder review, exception-case testing and load testing. In plain terms: walk through the new process on paper with the people who will use it, try the odd cases (the client with two entities, the invoice with no purchase order), and check it copes with a busy week.

Decide who checks the output

Automation that nobody reviews is a liability. The same guide says every AI workflow needs checkpoints and sets out four oversight patterns:

Pattern When it fits
Check before acting (human in the loop) High-impact decisions
Watch while acting (human on the loop) Medium impact, high volume
Spot-checks Low impact, very high volume
Exception handling Predictable edge cases

So an automated payment reminder might need only spot-checks, while an AI-drafted message to a client about a dispute should be read by a person before it goes out. Match the checking effort to the impact, and write down who owns it. Designing that review step is covered well in this overview of tools for online work.

Before staff paste client information into any tool, set the rules. Our pages on what staff can paste into AI tools under the Privacy Act, AI data leakage and AI policy for small business give you a starting point. If a vendor will handle your data, run it through the one-page AI vendor risk checklist first.

A 90-day sequence for the first workflow

  • Weeks 1 to 2: survey staff, list recurring tasks and tools, and calculate hours using the formula above.
  • Weeks 3 to 4: score the top candidates with the five-criteria matrix and choose one workflow.
  • Weeks 5 to 6: write the one-page specification, set the review pattern and agree the success measure (hours saved, errors avoided).
  • Weeks 7 to 10: build or configure, test with exception cases, train the people involved.
  • Weeks 11 to 13: run it live, measure against the baseline, decide whether to extend or pick the next workflow.

The Dynamic Business experts make the same point about scope: start with one clearly defined workflow and measure whether it saves time or reduces mistakes. A measured win makes the second project easier to approve.

What our assessment adds

The Calibre Forum report is a one-off A$1,990. It combines anonymous answers from your whole team into a scored report that maps current tool usage and capabilities, flags risks, and sets out a tailored 90-day action plan. For firms that want to track progress, the membership is A$2,388 per year and covers recurring surveys and reports each quarter. You can see exactly what is in the readiness report and how the process runs from setup to report before you decide.

Perspective

We built Calibre Forum around one observation: management usually cannot see how work is really being done with AI tools, and the people who can are reluctant to say it in a meeting. So we ask every staff member, anonymously, and we collect no individual data and install no software.

That matters for the question of what to automate first. A shortlist built from what a few managers remember is a shortlist built on guesses. A map built from the whole team shows where tools are already used, where work is repeated by hand and where the risks sit, and it does so in the language of the people doing the work.

Our scored report turns that map into a 90-day plan with concrete recommendations for operations and security. We think the order matters: see first, then choose, then automate one thing well. Firms that follow that order spend their money on the tasks that actually cost them time.

Anthony

Get the picture for your own firm

If you would rather start from your team’s real answers than from a guess, the Calibre Forum assessment report gives you a scored map and a 90-day plan for A$1,990 one-off.

Calibreforum

Firms that want to repeat the exercise every quarter as tools and habits change can take the annual membership at A$2,388 per year. If you are an IT provider, accountant or broker advising clients on this, see our page for firms.

FAQ

What is the single best first task to automate?

There is no universal answer, but recurring invoices and payment reminders are a common starting point because the rules are fixed and the benefit is easy to measure. Lead capture and follow-up is the usual sales equivalent. Score your own candidates on volume, repetitiveness, data availability, error tolerance and current cost, then take the top scorer.

How do I know if a task is too complex to automate yet?

If it depends on judgement, varies from case to case or relies on information that is not stored digitally, it scores low on fit. The National AI Centre advises improving such processes first with templates, simple rules or streamlining. Come back to them once the steps are consistent.

How much should I document before automating?

A one-page workflow specification is enough for a small firm: a flow diagram, who is responsible, decision rules and escalation, quality standards, training needs and the systems and data involved. Test it on paper with the people who will use it before anything goes live.

How many hours does repetitive admin take in a small firm?

It depends on your systems, so measure it. Multiply transfers per week by time per transfer by 52. In the ADAI News example, a 10-person firm doing 20 one-minute transfers per person per day loses 858 hours a year.

Do I need to tell staff before surveying them about tasks and tools?

Yes. Explain why you are asking, that answers are anonymous and that the aim is to improve how work gets done. Candid answers are what make the map accurate.

Sources

Recommended

General information only. Articles on this blog are written with the help of AI writing tools and published by Calibre Forum. They are not legal, insurance, tax or financial advice, so check anything important with a qualified adviser.

See how your business really works.

Ten minutes of setup. Your team answers on their phones. A report arrives when the survey closes.