Calmer is not the same as working

You turned on an automation and things feel calmer, but you cannot say whether it is working. Maybe it is the system, maybe a slow month, maybe wishful thinking. Without a measurement plan, you will cancel something that works or keep paying for something that does not.

Working means three things: it runs as built, it moves a number you care about, and it costs less than that number. Each one needs its own check, which is why this plan has five checkpoints.

The fix is a short plan written before launch. It names what you will count, where the numbers come from and what you will decide at each checkpoint. An hour of writing now saves months of guessing later.

This plan fits any workflow automation, from quote follow-up to missed-call text-back to approvals. The examples use a small service business, and you can swap in your own numbers. It also works if the automation is already live. You will rebuild the baseline from the records.

Choose one number and two guardrails

A good measure ties to money, time or capacity, and it comes from records you already keep. Pick one primary number for the workflow. Then pick two guardrails: numbers that tell you whether the automation is doing harm while it helps.

Guardrails matter because automation can improve one number while hurting another. More follow-up texts can raise bookings and opt-outs together. You want to see both.

  • Quote follow-up: jobs won from quotes older than a week. Guardrails: opt-outs and complaints.
  • Missed-call text-back: missed calls that become booked jobs. Guardrails: callers who ask for a person, and urgent calls routed late.
  • Declined-work follow-up: recovered dollars. Guardrails: complaints, and safety items that reached an advisor.
  • Approvals by text: time from estimate to decision. Guardrails: disputes about what was approved.
  • Review requests: new reviews per hundred invoices. Guardrail: private complaints that reached you.

If two numbers compete, pick the one closest to money. Jobs booked beat messages sent, and dollars collected beat invoices sent. Activity counts help at week two, and the outcome number tells you the result.

For back-office automations, the primary number may be time: the hours a week someone spends on a manual step. A week of written notes is crude, but it beats a guess.

Write the three numbers on one card, with the source of each. If you cannot say where a number comes from, it is not ready to be a measure.

Take the baseline before you build

A baseline is what the workflow produced before the automation existed. Collect four weeks of it if your tools keep history, and two weeks of hand counting if they do not. Record the volume, the outcome, the time spent and the dollars.

  • Volume: the calls, quotes, estimates or invoices handled in the period.
  • Outcome: the bookings, approvals, payments or reviews that resulted.
  • Time: the hours a person spent on the manual steps, from a week of notes.
  • Dollars: the average job or invoice value from the same period.
  • Guardrails: complaints, opt-outs, errors or reversals in the same period.

Volume matters as much as the outcome. If you only record jobs booked, a busy month will look like a win. If you record calls received, jobs booked and the rate between them, you can separate the automation's effect from a busy season.

Ask your builder for the same baseline. Our process starts the same way: a free walkthrough maps the workflow and times each step as it runs today. The quote and statement of work then record the yardstick the build will be judged by.

If the automation is already live, rebuild the baseline from the records before its start date. Phone logs, quote lists and invoices are good places to look. A rough baseline is far better than none.

If the numbers live in two systems, pick one as the source and stay with it. A call log says how many calls came in. A scheduling system says how many jobs were booked. Match them by week, and write down the match rule so it never changes.

Name one person to own the numbers. It can be you, an office manager or an advisor. The job is a short block of time at each checkpoint.

The 90-day plan at a glance

Five checkpoints cover the three months. Each one asks a single question, and each has a response if the answer is no.

The 90-day check
CheckpointWhat you checkWhat good looks likeIf it is off
Baseline, before launchVolume, outcome, time and dollars for four weeks, plus the guardrailsNumbers written down, with the source of eachDo not launch until you can count the workflow
Week 2Behavior: triggers fire, messages send, handoffs reach a person, actions are loggedEvery step runs as built, and exceptions reach a humanFix the build before judging results
Day 30The primary number against the baseline, adjusted for volumeThe number moved the right way and the guardrails heldCheck the trigger, the wording and the timing, one at a time
Day 60Stability: is the change holding, and is the season skewing it?Same as day 30 or better, after adjusting for seasonLook for drift, workarounds and skipped steps
Day 90Net dollars, hours returned and risk, against what it costs to runA clear decision: keep, change, widen or stopNarrow the scope or stop, and keep what you learned

The order matters. Behavior comes before results, and results come before decisions. Judging results at week two is a good way to condemn a working system, since a handful of events cannot show a trend.

Week 2: check behavior, not results

Two weeks in, you have too little data to judge results, but you can judge behavior. Open the log and check that the automation does what you built it to do.

  • Triggers fired on the events they should, and stayed quiet on the ones they should not.
  • Messages went out on time, in your voice, to the right people.
  • Exceptions reached a person with context attached, and the person acted on them.
  • Stop rules worked: replies, bookings and opt-outs ended the sequence.
  • Every action appears in the log, with a time and an outcome.

Read ten real messages end to end, and test one odd case on purpose. Send yourself a quote that never gets a reply, or call after hours, and watch what happens. Fix anything that surprises you now, before it has run for a month.

Every automated action is logged, so the log is also your first source of counts. If a step does not appear in the log, treat it as not done. If the log is hard to read, ask your builder for a weekly summary of actions taken.

Look at a quiet day as well. If nothing happens on a day with no events, that is correct. If the automation sends something with no trigger, find out why before it sends again.

Check the human steps too. Are the people who approve things actually approving them, and how fast? An automation that waits on a person who never looks at the queue will look broken when the cause is the queue.

Ask the builder to walk you through one real record from start to finish: the trigger, each message, each handoff and the log entry. If it takes a few minutes, the system is legible. If it does not, ask for documentation before you go on.

Day 30: read the result against the baseline

At day 30 you have a first real reading. Compare the primary number with the baseline, and correct for volume before you celebrate or worry. A rate tells you more than a raw count.

Say a service company turns on missed-call text-back. In the baseline month it had 60 missed calls and 9 booked jobs, which is 15 per 100. At day 30 it has 80 missed calls and 20 booked jobs, which is 25 per 100.

The raw gain looks like 11 jobs. But volume rose, and at the old rate of 15 per 100, 80 missed calls would have booked 12 jobs. The automation's share is 20 minus 12, which is 8 jobs.

At $350 a job, that is $2,800 for the month. Say the build cost $2,400 and runs for $100 a month. The month nets $2,700, so the build is paid back in about a month. Halve the gain to be careful, and the month still nets $1,300, which pays the build back in under two months.

Time-based measures work the same way. Say an office manager spent 6 hours a week retyping job details between two systems, and at day 30 it is 1.5 hours. That returns 4.5 hours a week. Across 52 weeks, that is 234 hours, and at $30 an hour it comes to $7,020.

Hours returned only count if they go somewhere: more quotes sent, invoices out sooner, an evening back. Ask where the 4.5 hours went, and write the answer next to the number.

Small samples wobble. If only a handful of events happened, treat day 30 as a first look and let day 60 settle it. Write down what else changed that month, such as a price change, a new hire or a storm.

Day 60 and day 90: separate luck from results

Day 60

At day 60, ask whether the change is holding. A result that fades can mean a step is being skipped, a message has gone stale or the first rush was a one-time backlog. Compare with the same weeks last year if you have them.

Four things can make the numbers mislead you. Check each one before you trust a result.

  • Season: a busy or slow stretch moves the number without any help from the system.
  • Volume: more inbound activity raises counts, so compare rates as well as totals.
  • Counting drift: if you count differently at day 60 than you did for the baseline, the comparison breaks. Write the counting rule down once.
  • Over-crediting: counting every booking that follows an automated message as a win, even when the customer had already decided to book.

Day 90

At day 90, decide with arithmetic. Add the dollars the automation produced and the hours it returned. Subtract what it costs to run, and look at the guardrails. Then choose one of four moves.

  • Keep it as it is, when the number moved and the guardrails held.
  • Change it, when the number moved a little and one setting looks wrong, such as timing, wording or the trigger.
  • Widen it, when it works and the same pattern exists in another location, crew or service line.
  • Stop it, when the number did not move after you fixed the obvious causes.

Take the day-90 numbers back to your builder. The yardstick in the statement of work is the first thing to check. Any change to the build gets a written quote like any other work, and you keep your data and documentation whichever way you go.

Ask for a short written summary: what was built, what the numbers show and what you would change. That summary also becomes the baseline for the next workflow you choose.

What to do this week

You can do all of this with a notebook and the reports you already have. Start before the build, or rebuild the baseline if the system is already running.

  1. Write the name of the workflow and the one number it should move.
  2. Add two guardrails, and write the source of each of the three numbers.
  3. Name one person to own the numbers, with a short block of time at each checkpoint.
  4. Pull four weeks of history for volume, outcome, time and dollars, or start hand counting today.
  5. Put the checkpoints in your calendar: week two, day 30, day 60 and day 90, counted from go-live.
  6. Ask your builder for the log and for a weekly summary of actions taken.

That is an hour or two, and it turns an automation from a feeling into a number. If you want help with the first workflow, the free walkthrough ranks your options with an impact estimate. The workflow automation page lists what gets built. A written quote and statement of work come first.