To know whether AI automation works, the key is recording what things look like now, before launch. Three kinds of numbers matter most: hours spent each week, how many mistakes happen, and how long it takes to reply to customers or colleagues. After launch, record them the same way and compare. To estimate payback, multiply the hours saved each month by the hourly staff cost and set that against the build cost and monthly maintenance. Without a baseline, all you can say after launch is that it feels faster, which makes it hard to decide whether to do the next workflow.
1. Record a baseline before launch
The baseline is how things work before automation. We suggest spending two to four weeks before launch with the colleagues involved keeping an honest record. It doesn't need to be precise, but use a method you can repeat afterwards.
| What to record | How | Examples |
|---|---|---|
| Hours spent each week | The colleague notes how long the task took each day | Preparing quotes, copying orders into the ERP, assembling the weekly report |
| Mistakes | Note each wrong entry, omission or misdirected send, and why | Wrong quantity typed, wrong quote version sent, a lead nobody answered |
| Reply time | How long from a customer or colleague asking to someone answering | From inquiry to quote sent, from billing statement to manager approval |
| Items stuck (optional) | At the end of each week, how many are still unfinished | Forms awaiting approval, leads not yet answered |
Record at your normal pace; don't rush or slow down because automation is coming. The more realistic the baseline, the more meaningful the comparison later. Recording can be simple: a shared Google Sheet filled in for a minute or two each day is enough.
2. Record again the same way after launch
After launch, record for a while using the same fields and the same method. When comparing, watch three things:
- Compare the same task: if you recorded quotes before launch, compare only quotes after, without mixing in other work
- Count the exceptions: how many items landed on the pending list and how long people spent on them belong in the total
- Allow for busy and quiet periods: volumes differ at month end, year end and around trade shows, so compare similar periods
The automation keeps its own records: when each item came in, when it was finished, and which ones failed and were rerun. Many of the post-launch numbers can come straight from these records, without asking colleagues to note them by hand.
3. How to estimate payback
A payback estimate can be simple. Work out one number on each side:
- Saved: hours saved each month × hourly staff cost
- Spent: the build cost, plus monthly maintenance, plus usage-based fees such as AI usage and LINE messages
For example, record the time a sales assistant spends preparing quotes each week, once before launch and once after; the difference is the hours saved per week, which you then convert to a monthly figure. If several colleagues do the same task, add up the time each of them saves.
Divide the build cost by the monthly savings minus the monthly maintenance and usage fees, and you get a rough number of months to pay back. Hourly staff cost can be estimated from monthly salary plus employer contributions such as labor and health insurance, divided by monthly working hours. This is only a rough estimate, and saved time doesn't always turn into money, but it shows whether a workflow is worth doing.
4. Look beyond hours saved
Some effects are hard to turn into hours but still matter. For example:
- Fewer mistakes: one wrong cell carries into every team downstream, and going back to fix it takes time too
- Less dependence on one person: reports and follow-ups keep running when the colleague in charge is on leave
- Managers can see progress: no need to ask around, since you can look up where each item stands
- Customer wait time: compare against the reply-time baseline to see whether it got shorter
Write these into your monthly review and read them alongside the numbers.
5. When to adjust or switch off
If after a while the pending list stays long, or colleagues still fill in a lot of data by hand, the rules or the AI's instructions need adjusting. If a workflow runs only a few times a month and the time saved doesn't cover maintenance, consider switching it off and moving the effort to a process that takes more time. After adjusting, record again the same way for a while to see whether the list gets shorter.
A good sign is when colleagues stop going back to the old way and start asking whether something else can be automated too. That's the time to pick the next workflow, and record its baseline first in the same way.
The short version
Before launch, record weekly hours, mistakes and reply times, then record them the same way after launch; for payback, multiply hours saved by staff cost and compare with build and maintenance costs.
Frequently asked questions
- Do we really need a baseline? What if we're too busy to record one?
- We suggest recording at least weekly hours and mistakes for one workflow. Two weeks and one colleague is enough; what matters is recording the same way again afterwards.
- How soon can we see results?
- It depends on how often the workflow runs. A workflow that runs every day builds up enough records to compare within a few weeks of launch; a report that runs once a month needs several rounds before a difference shows.
- How is the cost of automation worked out?
- Pricing is scoped to your needs, with a quote after the free 30-minute consultation. The quote depends on how many systems and workflows are involved, how complex the rules are, AI usage and the scope of maintenance after launch. Third-party fees such as AI usage and LINE messages are charged at actual usage with no markup.
- Will AI automation definitely save money?
- No. How much it saves depends on how often the workflow runs, how long each run took by hand, and the build and maintenance costs. That's why you record a baseline first, compare after launch and decide with your own numbers.