Automation Audit

Sample Client A · A/C service · McAllen, TX · 9 staff

How to read this

An audit is a written account of how the work actually moves through your business, what each manual step costs you per month, and which of those steps are worth automating first. It is produced from one hour with you and a look at the tools you already use.

It is not a proposal, a quote, or a list of software to buy. Nothing in it obliges you to build anything, and the sections that matter most are the ones that argue against building — §7 exists because an audit that recommends everything is a sales document wearing a lab coat.

Every figure carries its source. Where a number came from the owner's estimate rather than something counted, the table says so on that row. Read those rows as the weaker ones; they are the first thing a real engagement measures properly.

SESSION

One hour with the owner and the dispatcher

SCOPE

Inbound call through to paid invoice. Field work and pricing excluded.

RATE USED

$38 / hour loaded, agreed at the session — wage plus employer cost, not billing rate

The business does not have a technology problem. It has a handoff problem, and July makes it visible.

Work arrives by phone, almost entirely. The dispatcher answers what she can and the rest goes to voicemail, where it sits until she has a gap. There is no list of calls that were never returned — the missed-call log on the handset is the list, and it scrolls. Everyone knows this and nobody can point at how many jobs it costs, which is itself the finding.

A tech finishes a job and texts photographs and a few lines of notes to the office. Someone in the office turns that into a quote by hand, in the same quoting tool every time, typing from the text thread. The quote is usually accurate. It is rarely fast, and in July it is slow enough that customers call to ask where it is — which lands back on the dispatcher, during the hours she is least able to take it.

Invoices go out from the accounting package. Whether one gets chased depends on whether anyone opens the aged-receivables view that week. The office staff describe the chase rule three different ways, which matters later in this document.

Reviews are asked for when someone remembers. The owner believes this is the single largest missed opportunity in the business and is probably right; it is also the step most often skipped, because it happens at the end of the day when the person who would do it is finishing something else.

Two things are working and should be left alone. Scheduling inside the existing calendar is fine — the dispatcher is fast in it and no tool would beat her. And the tech-to-office text thread, though it looks like a mess, is a genuinely good capture step: the photographs are consistently there. The problem is what happens to that message afterwards, not the message.

One job, left to right, from the phone ringing to the invoice being paid. Lanes are roles. The marks say how each step runs today.

Every manual step inside the scope, costed. Hours per month are computed from the frequency and the minutes, not estimated separately.

Assumptions printed so they can be argued with. $38 / hour loaded, agreed at the session — wage plus employer cost, not the billing rate. 22 working days and 4.33 weeks per month. Rows are rounded to 0.1 hours; the total is computed from unrounded values, so it does not equal the visible column summed by hand. Two rows are owner estimates rather than counted — together they carry 28.4 of the 78.6 hours, and they are the first two things a real engagement would measure properly before anyone builds against them.

What to do first, and why that order. Hours carry forward from §4 unchanged.

Effort: S = under 8 hours of build. M = 8–24. L = more than 24, or unknown until a dependency is settled. No score column. A weighted score invented for this document would be a fake number in a report whose whole argument is that the numbers are real. Ties are broken by dependency: sequence 3 sits behind sequence 1 because it uses the same SMS path, not because it is worth less.

Three, in the order of §5. Each says what it does not fix, because that is the part you cannot get from a demo.

Three things in this business should not be automated now, and one of them is the request that started the conversation.

The invoice chase — because nobody agrees what the rule is

Asked when an invoice gets chased, three people gave three answers: at 14 days; at 30 days unless it is a commercial account; and "when I notice it in the aged list". All three are in use. None is wrong — they are three different policies running at once, and the difference has never mattered because a person was applying judgement each time.

Automating this now means picking one of the three and hardening it, silently, into the system of record. If the chosen rule is the wrong one, the business will not find out from a bug report; it will find out from a commercial customer who got chased at 14 days and did not appreciate it.

What to do instead: settle the rule first — a fifteen-minute decision, not a project. It is sequence 4 in §5 precisely because the dependency is a decision rather than an API. Once the rule is written down, the build is small.

Parts ordering — because the arithmetic does not support it yet

The owner asked about automating parts ordering, and it is the most technically interesting thing here. It is also the weakest case in the document:

3 orders / week × 7 minutes = 1.5 hours / month = $57.59 / month [owner estimate]
build effort M = 16–24 hours payback 11 to 16 months
sequences 1 and 2, for comparison, return 44.0 hours / month [§4]

These three figures are the only ones in this document that are not carried forward from §4, because parts ordering was raised in the session and never inventoried. They are the owner's estimate, marked as such here for the same reason the inventory marks its two — they are doing the work of ranking something out of the plan, so they should be the first thing challenged.

The payback is not absurd, which is what makes this the honest example rather than a straw one. It is simply ranked out: the same build hours spent on sequences 1 and 2 return twenty-nine times the monthly hours. Come back to it when the first phase is running and the volume has grown — the case improves on its own if the business does.

Rebuilding the day after a cancellation — because a person is better at it

It is the third-largest line in §4 at 8.7 hours a month, and it is deliberately in no sequence in §5. Scheduling is one of the two things §2 found already working: the dispatcher is fast in the calendar and knows which customer will take a same-day slot, which is judgement rather than a rule. An automation would have to encode that judgement, and every version of it we discussed was a worse dispatcher than the dispatcher.

It is costed anyway, because leaving it out of the inventory would flatter the total. Not everything expensive is worth removing.

Plain English, on purpose. If you would rather hand this page to your bookkeeper than read it yourself, it should still make sense.

The tools

The automation itself runs in n8n, an open workflow tool. It sits between the systems you already use and moves information between them on a trigger. It does not replace your phone system, your quoting tool, your calendar or your accounting package — all four stay exactly as they are, and n8n is the thing that stops you being the transport between them.

The order

Sequence 1 first, on its own, for two weeks. It is the smallest of the three, it needs nothing that does not already exist, and running it alone proves the SMS path works before anything else depends on it. Sequence 3 goes second because it reuses that same path — by then the risky part is already known to work. Sequence 2 last of the first phase: it is the largest, and it touches the quoting tool, which is the system where a mistake is most visible to a customer.

Where a person stays in the loop

Nothing that a customer sees goes out unreviewed except the two messages you have explicitly approved the wording of: the missed-call acknowledgement and the review request. The quote always stops as a draft. That is a design decision, not a limitation — an automation that can send a wrong price to a customer is one you will end up switching off.

When a step fails

Every workflow has an error path, and it is the same one: the run stops, nothing half-finished is left behind, and a message goes to the office saying which job it was and what did not happen. The message says what failed, not what the customer wrote. Failures are loud on purpose — a silent automation that quietly stops working is worse than no automation, because you stop checking.

Who gets told

The office, by default, because they are the people who can act on it. The one exception is a failure in the accounting path, which goes to the owner as well, because that is the one whose consequences show up a month later.

Everything: the workflows, the documentation, and the credentials they run on.

They sit on a server you control, or one I run for you — your choice, and it moves either way. Stop the managed support and nothing switches off; it is documented plainly enough for you, or anyone you hire, to maintain. If you want to make small changes yourself, I will show you at handover, at no extra cost.

What arrives at handover

• Workflows exported as files, delivered to you, and confirmed received.
• Every credential in an account you own. Where an account was created for you, you hold the recovery method.
• One page per workflow: what it does, what triggers it, what it touches, and what to check when it looks wrong.
• The list of every connected service that receives data, named.
• What alerts, to whom, and what the alert does and does not contain.
• Known limits, and anything deliberately left manual with the reason.

The session itself

• We walk one live run end to end, together.
• You make one small change yourself, with me watching.
• You log in unaided, on your own device, before we finish.

If you leave

• A written statement of what keeps running and what stops.
• Nothing is time-bombed, licence-locked, or dependent on a Kerrovia account staying open.