Most conversations about automating a business start in the wrong place. They start with what the technology can do, which is a question with an impressive and largely irrelevant answer. The question that decides whether a project is possible, cheap or doomed is much duller: can the data get out of the system it currently lives in, and can the answer get back in?

This is a way of auditing your own systems before you talk to anyone, so you arrive knowing which parts of your business are easy, which are expensive, and which are not worth attempting this year.

The five tiers

Every system you own falls into one of these. The tier, more than anything else, sets the cost.

Tier 1 — It has a documented API

Modern cloud software: your email, your cloud storage, most CRMs, payment providers, newer accounting platforms. Somebody has already written down how to read and write data, there are libraries for it, and connecting to it is a day or two of work rather than a project.

If the process you want to automate lives entirely in tier-1 systems, you are in the easy case and should expect to be quoted accordingly. Be suspicious of large numbers here.

Tier 2 — It has an API, but you have to pay or ask for it

Common with enterprise software and with Vietnamese platforms sold on a licence model. The capability exists; access to it is a commercial conversation with your existing supplier, sometimes involving a higher tier of your current plan.

The work is still straightforward. The delay is procurement, and it is worth starting that conversation before you commission anything, because "we are waiting on the vendor" is a very common reason for a project to sit idle for six weeks.

Tier 3 — It exports files

No live connection, but it can produce a CSV or an Excel file on a schedule, or somebody can. This is where a great deal of real business software sits, including a lot of what runs manufacturing and logistics in Vietnam.

Automation here works and is not expensive, but it is batch rather than live. Your system will be acting on this morning's data, not this minute's. For invoice processing, reporting, reconciliation and most back-office work, that is completely adequate. For anything a customer is waiting on, it is not.

Tier 4 — It has a screen and nothing else

An old on-premise system, a desktop application, something a supplier built a decade ago and no longer supports. The data is visible to a person and unavailable to anything else.

There are ways in — reading the database directly if you can get access to it, or driving the user interface automatically, which is brittle and breaks whenever the screen changes. Both are real options and both cost considerably more than tiers 1 to 3. Before paying for either, ask whether the answer is to replace the system rather than to build a bridge to it.

Tier 5 — It is paper, or a PDF of paper

Scanned documents, photographs of delivery notes, forms filled in by hand, a PDF that is really a picture.

This is the tier where modern document extraction genuinely changed what is possible, and it is the single most common place where automation pays for itself in a Vietnamese SME. It is also the tier that most needs a human review step, because a model reading a creased scan will sometimes be wrong and will not tell you it is unsure.

One place Vietnam is unusually easy

Electronic invoicing has been mandatory here since 1 July 2022 for enterprises and organisations established under Vietnamese law, under a clearance model where the invoice is validated by the General Department of Taxation and carries an official code before it reaches the buyer.

The practical consequence is worth understanding. Your supplier invoices are not scans of paper. They are structured, machine-readable documents that have already been validated by a third party, available through your e-invoice provider's system. That puts a large and usually painful part of your accounts payable process in tier 1 or 2 rather than tier 5.

Businesses here routinely automate invoice handling as if it were a document-scanning problem, because that is how the international vendors sell it. If your invoices are arriving electronically, the extraction problem you are being quoted for may not exist. That is a question worth asking before you buy anything.

The honest list for a typical company

For a Vietnamese business of twenty to two hundred people — manufacturer, exporter, trading company, professional services — this is roughly where things tend to fall.

Usually straightforward:

  • Supplier invoices into your accounting system, if they arrive electronically.
  • Routing incoming email to the right person or department.
  • Producing the same report every week from data that already exists in a system.
  • Moving a record from a web form into a CRM or a spreadsheet.
  • First-draft translation of product specifications and documentation, for a person to edit.
  • Flagging unusual transactions for review against rules you can write down.

Possible, but a real project:

  • Anything involving a system with no API, where the integration has to be built sideways.
  • Extracting data from scanned or photographed documents at volume.
  • Reconciling two systems that disagree about what a customer is called.
  • Anything that has to respond within seconds rather than within a day.

Rarely worth it yet:

  • A process you run a few times a month. The build cost does not amortise.
  • A process no two people perform the same way. Automating it does not resolve the inconsistency, it freezes one version of it into software and makes it harder to change.
  • Anything where a confident wrong answer is expensive and a human check would be needed anyway — being confidently wrong is the failure mode this technology does not warn you about — in which case you have added a system without removing the work.
  • Anything depending on knowledge that lives in your staff's heads and has never been written down.

Auditing your own stack in an afternoon

Take the process you most want to automate and write down, in order, every step a person currently performs. Not the idealised version — the real one, including the bit where somebody checks a spreadsheet that is not supposed to exist.

Then, against each step, write two things: which system it touches, and which tier that system is in. You will usually find one of three patterns.

All one tier, mostly 1 to 3. Good. This is a contained piece of work and you should get a sensible quote.

A chain that drops into tier 4 or 5 in the middle. Very common. That one step is where most of the budget will go, and it is worth asking whether the step can be removed, moved to a different system, or handled by a person while everything around it is automated. Partial automation of six steps out of eight is frequently the right answer and almost never the one that gets proposed.

No system at all for several steps. The process lives in email and judgement. This is not an automation project yet. It is a process-definition project, and doing it will deliver value on its own — some companies find at this point that the automation is no longer the priority.

Two questions that save money

How often does this run, and how long does it take a person? Multiply them. A task taking fifteen minutes, forty times a week, is twenty-six hours a month and worth attention. The same task four times a month is an hour, and no automation project recovers its cost against an hour.

What happens when it is wrong? This determines the design more than anything else. If a mistake is caught immediately and costs nothing, the system can run with light review. If a mistake reaches a customer, a tax filing or a shipment, you need a review step — and a review step means the saving is "typing replaced by checking", which is still a real saving but a smaller one than the proposal will claim.

Being realistic about this at the start is what separates projects that deliver from projects that produce an impressive demonstration and then quietly stop being used.

Where to start

Pick the process with the highest number from the first question and the lowest stakes from the second. High volume, low cost of error. That is almost always document handling or routing, and it is almost always worth doing before anything more ambitious.

Build that one. Measure it for a month against the number you agreed beforehand. If you are at the point of choosing who builds it, the five questions worth asking a vendor are here. Then decide whether to do the next one, with the advantage of knowing what this actually costs in your business rather than what a proposal estimated.

If you want help putting the tier list together for your own systems, send us the process and the names of the software involved. We will tell you which tier each one is in, including when the answer is that the project is not worth doing.