List the recurring tasks in a week and score each on three axes: frequency (how often), predictability (does the same input always need roughly the same output) and cost of error. Automate the high-frequency, high-predictability, low-error-cost work first. That is almost always: answering common enquiries, tidying data, and producing the weekly report.
| Step | Manual today | With an agent |
|---|---|---|
| 1. Message arrives | Staff notices it eventually | Agent receives it instantly on WhatsApp |
| 2. Classify | Staff reads and judges | Agent classifies: price question, booking, complaint, other |
| 3. Answer | Staff looks up the price list | Agent answers from the knowledge base in seconds |
| 4. Record | Staff copies details into the CRM — or forgets | Agent writes the record with the customer, intent and outcome |
| 5. Follow-up | Someone remembers, or does not | Agent schedules the follow-up message |
| 6. Escalate | Ad hoc | Agent hands complaints to a human with the transcript attached |
The gain is not only speed. Steps 4 and 5 are the ones humans skip when they are busy, and they are exactly the steps that lose revenue.
| Metric | How to capture it |
|---|---|
| Volume handled | Count of conversations / tasks per week |
| First response time | Timestamp of message vs first reply |
| Human touch rate | Percentage escalated |
| Hours saved | (Manual minutes − automated minutes) × volume |
| Error rate | Corrections and complaints per 100 tasks |
Record these for one week before switching the agent on. Without a baseline, a successful automation looks like a rumour.
The second workflow is far cheaper than the first because the platform, the knowledge base and the guard rails already exist. The mistake is scaling before the first one is stable: extend once the human touch rate and error rate are both within the targets you set, not before.
A high-frequency, highly predictable task with a low cost of error — commonly customer enquiries, data tidying or the weekly report.
Capture volume, first response time, human touch rate, hours saved and error rate for a week before launch, then compare over the following weeks.
Read-only by default, human confirmation before money or commitments, a full action log, a kill switch and one accountable owner.
Yes. Once the platform, knowledge base and guard rails exist, each additional workflow is significantly cheaper than the first.
Want to know how MOKAGI fits your business, need a quote, or book a demo? Reach us any of the ways below — we usually reply within one business day.