Where a small practice's hours actually go, and which of them automate

A small accounting firm saves hours with AI automation by measuring its own admin time first, then automating the tasks that pass a four-part test. There is no reliable general figure. Published "hours saved per week" numbers describe other firms, with client mixes, software and staffing you know nothing about.

Your own number takes two weeks to establish. Doing that first changes what you build, because most firms find their largest time cost is not the task they were about to automate. This guide shows how to measure it, which tasks pass the test, and what the arithmetic looks like once you include the cost of building and keeping the automation.

What we'll cover

• Why published time-saving figures are not yours
• How to run a time audit
• The categories worth separating
• The test a task has to pass
• Building it with CodeWords
• The arithmetic, done honestly
• What the recovered hours become
• Where the savings disappear
• The ceiling on automation
• How to make the automation reliable
• Limits of time-saving estimates
• What to measure first
• Frequently asked questions

Key takeaways

• A two-week time audit gives a firm its own hours-saved number, and that number should drive the build instead of a published figure.
• A task is worth automating when it has a definite trigger, identical steps, a bounded cost of error and enough frequency.
• Recovered hours become revenue only if demand is waiting for the extra capacity.
• Automation reduces admin time but does not remove it, so count review, upkeep and exceptions in the case.
• Rank tasks by frequency, not by irritation, because the task people complain about is rarely the one using the most hours.

Why published time-saving figures are not yours

Published time-saving figures are not yours because admin load depends on local variables. Two practices with the same headcount can have very different admin loads.

Client mix. Forty limited companies on a clean ledger system is a different business from a hundred sole traders sending photographs.

Software. A practice whose ledger, practice management and document store already talk to each other has automated most of the connective work without calling it automation.

Who does what. A firm where a qualified accountant answers the phone loses expensive hours; a firm with an administrator does not.

The fee model. Fixed-fee practices feel admin as lost margin. Hourly practices sometimes bill it, which changes the calculation entirely.

A figure from another firm shows that a category of problem exists. It does not show the size of yours, and building against another firm's number produces automations that save forty minutes a month.

How to run a time audit

Run the time audit for two ordinary weeks, with everybody tagging time at the point of doing the work. Reconstruction on Friday is where audits go wrong, because people remember the annoying tasks and forget the frequent ones.

Keep it to eight categories. More than eight and people stop tagging accurately. Fewer and the result is too coarse to act on.

Record interruptions separately from tasks. A two-minute query that costs ten minutes of lost thread should show as ten. Logging only the two understates the cost of interruption.

Pick ordinary weeks. Avoid January, year-end weeks and holiday weeks.

Include the partners. Senior time is the most expensive and the least examined, and it is where the surprise often sits.

Count frequency as well as duration. A four-minute task that happens sixty times a week matters more than a two-hour task that happens monthly, and the first is usually easier to automate.

Two weeks of this produces a ranked list of what your practice spends time on, with numbers you trust because you collected them.

The categories worth separating

Most practices find their hours fall into eight categories. The distinctions matter because the categories automate very differently.

Category	What it covers	Automation candidate
Chasing clients	Requests, reminders, checking what arrived	Yes
Answering the same question	Status, deadlines, what is needed, how to pay	Yes
Moving data between systems	Rekeying, exporting, reformatting, uploading	Yes
Reviewing and correcting	Checking machine output or junior work	Yes
Scheduling and coordination	Arranging calls, confirming and rearranging meetings	Yes
Internal reporting	Building the same view of the same numbers every week	Yes
Genuine client work	The work you are paid for	No
Genuine advisory conversation	The work you are paid best for	No

The first six categories are candidates. The last two are the point of the business. If the audit shows the last two take less than half the week, the firm has a structural problem that automation will not fix on its own.

The test a task has to pass

A task is worth automating when it clears four bars. Not everything in the first six categories does.

It has a definite trigger. A date, a status change or an arrival. "When it feels like time" is not a trigger, and tasks without one are the commonest cause of half-built automations.

The steps are the same every time. If the process branches on judgement, automation can prepare the work but should not complete it.

The cost of getting it wrong is bounded. Sending a duplicate reminder is recoverable. Filing something incorrectly is not.

It happens often enough to matter. Four minutes sixty times a week is worth building. Two hours once a quarter rarely is, whatever the annual figure looks like.

Tasks that fail the second bar are often still worth half-automating. Preparing a draft, assembling the information and presenting it for a decision removes most of the time and leaves the judgement with a person.

Building it with CodeWords

CodeWords connects to 3,000+ integrations (CodeWords integrations page, 2026), and each connection is made once and reused.

1. Open CodeWords and start a new automation.
2. Describe what should happen in plain language to Cody, the automation builder: the trigger, the steps as they happen today, what reaches a person, and what stops the whole process.
3. Authorize the connections to your ledger, practice management and document systems, and whatever channel clients use.
4. Describe the exceptions. This step decides whether the automation survives a real week.
5. Run it in parallel with the manual process for a full cycle and compare the results before switching.

You describe the outcome; Cody builds it, connects it, and deploys it. The Free plan costs $0, Pro costs $39 per month and Business costs $100 per month (CodeWords pricing page, 2026). Details are on the pricing page.

The arithmetic, done honestly

Honest arithmetic subtracts review, exceptions and upkeep from the time saved. Take one task from your audit. Suppose it takes twelve minutes and happens forty times a week: eight hours (an illustrative case).

Automation does not take that to zero. It takes the task to the cost of review and exception handling. Assume a well-built process leaves a quarter to a third of the original time, which makes the recovered time about five hours (an assumption to replace with your own audit data).

Then count the costs against it:

Building it. Days, not hours, for anything that touches more than one system, including the time spent working out what the process is.

Keeping it working. Software changes, a supplier changes a format, a rule changes. Budget for this up front.

The failures. Every automated process fails occasionally, and somebody has to notice and fix it.

The tail. Clients who will not use it, cases the rules do not cover, and exceptions that stay manual.

Five recovered hours against a two-day build and an hour a month of upkeep is a clear yes. The same five hours against a three-week integration project is not, and only your own numbers can show which case you are in.

What the recovered hours become

Recovered hours do not turn into billable work by themselves. They become billable work only if demand is waiting.

The other outcomes are capacity for more clients, if you are acquiring them, or earlier finishes, if that is the goal. Several are legitimate and only one shows up in revenue. Decide which you are aiming for before you build, because the choice changes what you build.

For many small firms, the first recovered hours go into work already being done badly for lack of time: the rushed review, the planning conversation that never happened, the client who should have been called in November. That value is real and does not appear as a line in the accounts for a year or two.

Where the savings disappear

Four patterns consume the savings.

Reviewing the automation for longer than the task took. If the output needs checking line by line, the automation is not finished.

Maintaining something nobody owns. A process with no owner degrades, and staff work around it manually while it still appears to run.

Handling exceptions badly. A rule that covers the common cases while the rest arrives unstructured can cost more than the original process.

Building for the client mix you wish you had. Automations designed for tidy clients leave the messy clients manual, and the messy clients were the expensive part.

The ceiling on automation

Automation has an upper bound, and knowing it before you start prevents disappointment. The first six audit categories are where automation lives, and automation removes a large share of them, not all.

In a typical small practice, the work that remains is client work and advisory conversation, which is what you sell. A practice that has automated well still has admin. The admin that remains is the part that required a person, and nobody spends Thursday evening rekeying a spreadsheet.

How to make the automation reliable

A reliable automation has an owner, a measured baseline and a documented manual fallback.

Re-run the audit annually. The ranked list changes as the client base and software change.

Give every automation an owner and a review date.

Measure the same thing afterwards. Without the before figure you cannot tell whether the automation worked, and everyone will hold a different opinion.

Document the manual route. You will need it during an outage and nobody will remember it.

Count the exceptions. A rising exception rate is the first sign that a process no longer matches reality.

Limits of time-saving estimates

Time-saving estimates have four limits.

Published figures describe other firms. Use them to identify categories, not as targets.

Automation does not create demand. Recovered capacity becomes revenue only if work is waiting for it.

Judgement does not automate, though the preparation around it usually does.

A messy client stays messy. Automating your side of a broken process does not tidy theirs.

Some of the worst hours are structural. Wrong software, wrong staffing mix and the wrong fee model are not repaired by automation, and automation can disguise them for a while.

What to measure first

Measure two weeks of tagged time first: eight categories, everybody including the partners, interruptions counted at their real cost. Build nothing during those two weeks.

Two habits make the difference. Rank by frequency, not by irritation, because the task everyone complains about is rarely the one consuming the hours. And record the before figure for whatever you automate first, because without it you cannot tell whether the next automation is worth doing.

Frequently asked questions

How much time does automation save a small practice?

No general figure is reliable. The result depends on client mix, existing software, who does which work and how you charge. A two-week time audit gives your own number, and that number should drive any build instead of a figure from another firm.

How do I run a time audit?

Run the audit for two ordinary weeks with every team member, including partners. Tag time at the point of doing the work instead of reconstructing it later, use eight categories, and log interruptions at their real cost, including the lost thread.

Which tasks are worth automating?

Tasks with a definite trigger, identical steps each time, a bounded cost of error, and enough frequency to matter. Short, high-frequency tasks beat rare, long tasks almost every time, because the saved minutes repeat many times a week. Start with the task that costs the most hours and has the least judgement in it.

What if a task needs judgement?

Automate the preparation and stop there. Assembling the information and presenting it for a decision removes most of the time, and the decision stays with a person. This keeps the speed benefit without moving professional responsibility to software. Automate the preparation around the judgement, so the person decides faster with everything in front of them.

What costs should go into the case?

Include the build, ongoing upkeep, the failures somebody has to fix, and the tail of clients and cases that stay manual. Measuring recovered time against build time alone makes every project look better than it is. Count the review time that remains after automation, because it is part of the true cost of the task.

Does automation turn into billable hours?

Only if demand is waiting. Otherwise the recovered time becomes capacity, earlier finishes, or work previously skipped. All of these can be worth having, so decide which outcome you are aiming for before you build the automation. Decide in advance where the freed hours will go, and measure that destination.

Why do time savings disappear?

Savings disappear when review takes longer than the task, when a process has no owner, when exceptions arrive unstructured, or when the automation fits tidy clients while messy clients stay manual. Each of these is preventable at design time. Name a destination for the saved time before you automate, and track it each month afterwards.

How much of a practice can be automated?

Automation covers chasing, repeated answers, data movement, coordination and internal reporting, and it removes a large share of that work but not all of it. Client work and advisory conversation remain, and they are the point of the business. Measure the share task by task in your own firm, not from a general percentage.

How often should the audit be repeated?

Repeat the audit every year. The ranked list shifts as clients, software and staffing change, and an automation built against a three-year-old picture often solves a problem you no longer have. Repeat it every six months, and also after any change in clients, staff or software that shifts where time goes.

Related reading

• Tax season as a queueing problem
• Closing the loops a practice leaves open
• Receipts, records and what HMRC needs