How to use this checklist
Pick one process, not the whole company. Good candidates are tasks your team repeats every day: answering the same customer questions in Zalo, typing data from invoices into accounting, chasing leads that went quiet, preparing the same weekly report.
Answer each question below with yes, partly or no. Write the answer down with one example. A yes without an example usually means partly.
When you finish, count the no answers in each group. The groups are ordered: if the process group is mostly no, the later groups do not matter yet.
Group 1: the process
AI works on a task that happens often and follows a pattern. A task that happens a few times a month rarely pays back the setup work.
- Can you describe the task in steps, the way you would explain it to a new employee?
- Does it happen every day or every week, not just a few times a year?
- Do people spend real time on it today, or is it a few minutes here and there?
- Is there a clear end result: a booking created, an invoice entered, a lead passed to sales?
- Is it clear who does the task today and how they decide what to do in unusual cases?
If you cannot describe the steps, start by writing them down with the person who does the work. This alone often shows waste that needs no AI.
Group 2: the data
An assistant can only answer from information it can read. An agent can only act in systems it can reach. Data that lives in one person's head or personal phone blocks both.
- Are your prices, services, opening hours and policies written down in one place that is kept up to date?
- Are past customer conversations in a business channel (Zalo OA, Messenger page, email) rather than in staff's personal Zalo?
- Are the documents the task uses in a readable format: text PDFs, spreadsheets, exports, rather than photos of paper?
- Can you get an export of the data without asking the original vendor for a favor?
- Is there one version of the truth, or do the sales sheet and the accounting system disagree?
Group 3: your systems
Connecting AI to the tools you already use often takes the largest share of the effort. The questions below decide whether the connection is simple, needs a workaround or is not possible yet.
- Do you have admin access to your Zalo OA, website, CRM and accounting tool, or does a former agency hold it?
- Does your CRM, booking or accounting software have an API, or at least a regular export? Ask the vendor or check their documentation.
- Which plan are you on? Some vendors open the API only on higher plans.
- Is there a shared place, such as Google Sheets, where a first version could write results before a full integration?
What different answers about systems usually mean| Situation | What it means for an AI project |
|---|
| API available and you have admin access | The assistant or agent can read and write directly; the project is mostly about the logic |
|---|
| Only exports or a shared spreadsheet | A first version works with a sheet; full integration comes later or not at all |
|---|
| No access, the old vendor holds the accounts | Recover access first; until then any integration is blocked |
|---|
| Paper and personal phones | Move the process into a business channel first; AI comes after |
|---|
Group 4: people and ownership
A project can stall on ownership even when the technology works. Someone has to decide what the assistant may say, check its work in the first weeks and handle the cases it passes back.
- Is there one person who will own the result and has time for it each week?
- Will the staff who do the task today help describe it, or do they see AI as a threat to their job?
- Who will answer when the assistant hands a conversation to a human, and how fast?
- Is management willing to change the process a little, not only add AI on top of it?
Group 5: risk, data protection and limits
Some tasks should keep a human in the loop no matter how good the model is. Decide this before the project, not after a mistake.
- Does the task involve medical, financial or legal advice? Then AI can collect and route the request, and a qualified person answers.
- Will the assistant see personal data: names, phone numbers, ID documents, health details? Then consent and storage rules need a check with your lawyer.
- What happens if the assistant gives a wrong answer? A wrong opening hour is cheap; a wrong price promise or a wrong delivery date may not be.
- Can you accept that the assistant says “I will pass this to a colleague” instead of guessing?
We do not give legal advice. Rules on personal data in Vietnam have changed in recent years; we confirm the current requirements with your lawyer before a launch.
How to read your answers
Mostly yes in all five groups: you have a good candidate. The next step is to estimate cost and payback on real data, which is what a paid audit or a scoped pilot does.
Yes on process and data, no on systems: start with a version that works through a spreadsheet or exports, and plan the integration once access is sorted out.
No on process or data: AI is too early. Spend a few weeks writing the process down and moving conversations into business channels. It is cheaper than any software, and it is the base for AI later.
No on people: wait. A project without an owner usually becomes an unused tool.