What to automate in your first ninety days, in the order that actually works

The rule the whole plan rests on

Do one thing at a time and finish it. That sounds obvious and almost nobody does it, because once you start looking at your operation properly you find eleven things worth fixing and they all feel urgent.

The order below is not arbitrary. Each phase produces the raw material the next one needs. Capture creates a record of every enquiry, which is what follow-up runs on. Follow-up creates outcomes, which is what a dashboard measures. Build them out of order and phase three shows you a chart of nothing.

Phase one, days 1 to 30: capture every missed call and message

The goal of the first month is narrow and absolute: nothing that arrives goes unanswered and nothing goes unrecorded. Not "fewer things", nothing.

This is first because it is the only phase whose value does not depend on anything else working, and because it is the one you can feel within days. It also produces the dataset every later phase needs, which is a complete list of who contacted you, through what, about what, and what happened next.

1. List every door into the business. Phone, text, the website form, email, Instagram, Facebook, WhatsApp, the number on the van, the second number nobody uses any more. Write them all down. Most owners find two or three they had forgotten, and forgotten doors are where enquiries die.
2. Pick the one that leaks most. For nearly every local business it is the phone, because a missed call leaves no trace. If you are not sure, the honest test is whether an enquiry through that channel at 7pm on a Friday would still be answered on Saturday.
3. Make it answer automatically, with something substantive. Not "we got your message". An answer that resolves the routine question, or asks the one thing you need, and offers the next step. Speed only helps if what arrives is useful.
4. Write everything into one place. Every enquiry becomes a record with a source, a timestamp and what was asked. This is boring and it is the single most valuable thing you build all quarter.
5. Set the escalation rule and be conservative. Decide what the system must never handle alone, and make it hand over early rather than late. A system that escalates too often is a tuning problem in week six. One that improvised an answer to an angry customer is a different kind of problem.
6. Add the second channel only once the first is boring. Boring is the acceptance criterion: a week where nobody mentions it because nothing went wrong.

Phase two, days 31 to 60: automate the follow-up

Phase one stops you losing people at the front door. Phase two stops you losing them afterwards, which is where the larger and quieter loss lives.

Almost every business has a version of the same hole: something gets quoted, offered, trialled or discussed, and then nothing happens because following up is nobody's actual job. It gets done well by one person on their good weeks, which means your conversion depends on the rota.

1. Name the thing that goes quiet. Quotes in the trades. Trials in a gym. Viewings in real estate. Consultations in a clinic. Carts in retail. There is usually exactly one that matters most, and everyone in the business already knows what it is.
2. Write the sequence as a person would send it. Three or four contacts over two weeks, each one saying something new. If a message exists only to bump the thread, delete it: that is the message that gets you muted.
3. Make it stop properly. It ends the moment they reply, book, buy or opt out. A sequence that keeps going after someone has answered is worse than no sequence, because it tells the customer nobody is reading.
4. Decide where a human takes over, and say so. The handoff should be announced, in plain words. Customers do not mind automation. They mind being unable to tell when it stopped.
5. Add reminders and confirmations to whatever gets booked. No-shows are a follow-up problem wearing a scheduling costume, and this is the cheapest part of the whole quarter.
6. Turn on the review request last. Timed to the moment a job completes, one tap. It only works once phase one is recording completion reliably, which is why it belongs here rather than in month one.

Phase three, days 61 to 90: a dashboard someone actually opens

Phase three is the one that gets built first by people who like software and skipped entirely by people who do not. Both are mistakes, and it belongs here because now there is real data in it.

The test of a dashboard is not what it contains. It is whether anyone looks at it in week twelve without being asked. Almost all of them fail that test, and they fail for the same two reasons: they show everything instead of the few things that drive a decision, and they have to be visited rather than arriving.

• Five numbers, not fifty. What came in, what got answered and how fast, what got booked, what is still open, and what completed. If a number would not change what you do next week, it is decoration.
• Push it, do not host it. The version people read is the one that arrives on Monday morning in a message they already open. A dashboard nobody navigates to is a dashboard nobody sees.
• Compare against your own baseline, not an industry average. The only comparison that means anything is you before, versus you now.
• Show the failures too. What the system escalated, what it could not answer, what nobody followed up. A dashboard that only shows wins gets trusted for a month and ignored forever after.
• One owner. Somebody's name is on it. Numbers that belong to everyone get read by no one.

What to do before day one

Two things, and they take an afternoon between them.

Measure the thing you are about to change. Your median response time this week. How many enquiries you got. How many you can actually account for. Without a number from before, you will never know whether it worked, and more practically, you will not be able to justify the next phase to yourself in month two.

Write down what you will not automate. This is more useful than it sounds. A short list of moments that must always involve a person, complaints, anything clinical or legal, anything where a wrong answer costs you a customer, turns into your escalation rules without any further work.

• Median time from enquiry arriving to a real answer, measured this week, not remembered.
• Enquiry count by channel over four weeks, so you know which door matters.
• The share you can account for at all, which is usually the number that stings.
• The list of moments that must always reach a human.
• Who owns this internally. Not the vendor, someone whose job it is.

The failure modes, in the order we see them

None of these are hypothetical. They are the ones that show up.

• Building all three phases in parallel. Everything is 70 percent done at day 90 and nothing is trusted. This is the most common failure by a distance.
• Automating a broken process. If nobody can say what happens to an enquiry today, automation makes the confusion faster rather than fixing it. Write the current process down first, badly, on one page.
• No baseline. Three months in, the honest answer to "did it work" is a shrug, and the project quietly loses its funding.
• Escalation set too loose at launch. Start conservative. Loosening a system that hands over too eagerly is a good week's work; recovering from one that confidently told a customer something untrue is not.
• Nobody owns it. The build finishes, the agency leaves, and six months later three things are broken because no one was watching. What that actually costs is in what a broken automation costs.

Where to start reading next

If you do not know which job is yours, do not guess. The automation scanner asks nine questions about how your week runs and ranks the seven categories against your own answers.

If phase one is obviously your problem, the speed-to-lead calculator sizes it, and the speed-to-lead guide covers what the published research does and does not establish.

If you would rather see the whole operating system before choosing a piece, the back office blueprint lays out the seven layers and which of them run themselves once built. And before committing to a build at all, build versus buy is worth ten minutes, because for a meaningful share of businesses the right first move is configuring something you already pay for.