Taking a process baseline before automation means writing down, over a few days, how often the process happens, how long it takes, how much manual work it needs and which errors it has. Without that measurement there is no honest way to know afterwards whether automating it with AI has worked.
This article is part of AI-First hybridisation consultancy.
In short
- The baseline is a picture of a process in numbers, taken before anything changes.
- Four figures and a note are enough: volume, working time, waiting time, errors and exceptions.
- It takes a week, a sheet of paper and the owner of the process.
- You measure the process, not the people who do it.
- The same measurement is repeated later, with the same definition, to compare.
What the baseline of a process is
The baseline is a picture of how a process works today, taken in numbers before anything changes. It answers four questions: how often the process happens, how long it takes, how much manual work it needs and which errors it has.
It is the step most often skipped in company hybridisation with AI. It gets skipped because it looks like a delay: the team wants to see the agent working, and measuring what is already done is not exciting. Three months later someone asks whether it was worth it, and all anyone has is an impression.
A week, a sheet of paper and a person who answers for the process are enough to take one.
What to measure: four figures and a note
Four figures are enough for most processes in a services company. The fifth row is a note, and it is often the one that teaches the most:
| Measure | The question it answers | How to get it without tools |
|---|---|---|
| Volume | How many times it happens in a week | Count what comes in: emails, orders, invoices or calls |
| Working time | How many minutes of a person each case needs | Time a sample of cases, from the first step to the last |
| Waiting time | How long a case takes from arriving to leaving | Note the date it came in and the date it went out |
| Errors and rework | How many cases have to be redone, and why | Write down each correction and its cause |
| Exceptions | Which cases fall outside the rule | A note on each odd case and how it was resolved |
Working time and waiting time are different things. A case can need ten minutes of a person and take three days to leave, because it sits in a tray. An agent usually shortens the wait first.
Exceptions get their own row because they are what break an automation. Every odd case you write down today is a rule the agent will need tomorrow.
How to measure in a week without stopping the work
Follow the order and do not chase precision. A simple measurement that gets finished is worth more than a perfect one that never does:
- Choose one process and name its owner. If nobody answers for it, that is your first finding.
- Write down where it starts and where it ends. «From the email arriving to the invoice being in the accounts» is a definition. «Handling invoices» is too loose.
- Count the volume over five working days in a row.
- Time a sample. Twenty cases picked without looking give a reliable idea of the working time.
- Write down each error and each exception as it happens, in one line.
- Put it all on one page and ask the owner of the process to review it and sign it.
The signature has a purpose: when the owner approves today's measurement, they also accept that the later one will be compared with it.
An example of a baseline on one page
This is what the page would look like for an example made up for this article: an accountancy practice that sorts the documents its clients send by email.
- Process: from the client's email arriving to the document being sorted and recorded.
- Owner: the head of administration.
- Volume: 180 documents in five working days.
- Working time: 4 minutes per document in the sample of twenty.
- Waiting time: a day and a half on average between arrival and recording.
- Errors: 11 documents redone, nearly all because they arrived without a tax number or twice.
- Exceptions: crooked photos from three senders, which have to be requested again.
With this page, the practice knows what to ask of the agent and what to compare a month later. Your company's figures will be different. What you copy is the sheet.
The most common mistakes when measuring
Nearly every baseline that turns out to be useless fails for one of these five reasons:
- Measuring afterwards. Once the agent is running, nobody remembers how long things used to take, and memory always favours whoever is asking.
- Measuring only time. A process can get faster and make more mistakes.
- Changing the definition along the way. If «before» ends when the message is sent and «after» ends when the client replies, the two figures cannot be compared.
- Using an unusual week. Quarter end or the middle of August do not stand for the rest.
- Trusting the estimate of the person who does the work. Almost all of us misjudge what we do every day.
Measure the process, not the people
A baseline counts cases, minutes and errors in a process. It does not need to know who handled each case, and it is better if it does not: a team helps more readily with a measurement that does not single anyone out.
It is also a data protection matter. The GDPR requires personal data to be adequate, relevant and limited to what is necessary for the purpose, which it calls data minimisation (Article 5(1)(c)). To find out how long a process takes, it is enough to note the cases, without the name of whoever handles them. Spain's data protection authority, the AEPD, devotes a section of its guide on data protection at work to this principle. Other jurisdictions have their own guidance.
If the measurement is going to rely on records from the company's systems, tell the team first: what is being measured, for how long and what for.
What to do with the baseline afterwards
The baseline is used three times. First, to choose: with the figures in front of you, it is clear which process deserves to go first, as we explain in which processes to automate with AI first. Second, to set the agent's target using the same measure. Third, to compare: once the agent has been part of the daily work for a while, the measurement is repeated with the same definition.
In the Legathum AI-First Hybridisation Method, the baseline is the product of the first deliverable of the implementation, the CAIO Audit. The first result in production arrives within 2 days and is measured against it. During the follow-up, the monthly review goes back to the same numbers.
Without that comparison, a company can have many agents and not know which one helps. With it, every process that comes in leaves a figure, and the AI-first company is built on results that can be shown. The full work is that of the AI-First hybridisation consultancy.
Frequently asked questions
What is the baseline of a process?
The measurement of how it works today, before anything changes: how often it happens, how long it takes, how much manual work it needs and which errors it has.
How long does it take to measure a process?
A week is usually enough: five working days of counting and a sample of twenty timed cases.
Do I need a tool to measure?
No. A spreadsheet or a sheet of paper by the desk is enough. What matters is defining where the process starts and where it ends, and not changing that later.
Can I measure without monitoring my staff?
Yes, and it is the better way. You count cases, minutes and errors in the process, without noting who handled each one.
What if the process is already automated?
Measure it now and use that figure as the starting point for the next change. You will not have the original «before», but you will stop working blind.
Who takes the measurement in a project with Legathum?
The baseline comes out of the CAIO Audit, the first deliverable of the implementation, and is reviewed with the owner of each process. The first step is a 15-minute meeting with Deibson Silva.