- AI transformation means changing how the company works so AI handles some steps of real processes, measured against how those processes ran before.
- Licenses for everyone or a chatbot on the website are tools. If the steps of the work stay the same, the company's numbers stay the same.
- When it is real, four things change together: the work, the data, the people and the controls.
- Start with one process that has a measurable problem and one owner.
- The first 90 days should end with a measured pilot, an AI-use policy, a team that can use the system, and a decision on the next three processes.
What AI transformation is
AI transformation is changing how a company works so that AI takes over some steps inside real processes, with data the system can reach, people who know what to check and rules for who approves what, and with the result measured against how the process ran before.
"Transformation" here means the steps change. If everyone has an AI account but accounts payable still types each invoice into the system by hand, the process is the one you had last year.
Definitions vary. Some consultancies count AI transformation as part of digital transformation; others treat it as the stage after it. The definition above is the one a development team uses when designing the work, because you can check whether it has happened.
Three things often called AI transformation
The first is buying licenses for everyone. ChatGPT or Copilot accounts help each person draft emails or summarize meetings faster, yet the company's processes run as before, and without a policy, customer data can end up in tools the company does not manage.
The second is a chatbot on the website that answers general questions from documents. When a question needs someone to look up a status or create a record, the work lands back with the same employee. A chatbot counts as a change in how you work once it is connected to the back office.
The third is digital transformation, which moves paper and manual work into systems, such as an ERP or an online sales channel. AI transformation builds on that by changing who does which step inside the systems you already have. A company whose core data is still on paper has to do the digital part at the same time, because AI cannot read data that is not in a system.
| Digital transformation | AI transformation | Buying AI tools for staff | |
|---|---|---|---|
| What changes | Paper and manual work move into systems; data starts being captured digitally | The steps of a process: AI does some, people decide and check the exceptions | Each person's own tasks get faster; company processes stay the same |
| Who leads | IT with the process owners | The executive who owns the outcome, with IT and the supervisors who do the work daily | IT or purchasing |
| How you know it worked | Work runs in the system; data is complete and retrievable | The process number moves against the baseline, such as time per item or items needing correction | Number of active users, which does not yet show a business result |
| Common failure | New system, old way of working | Starting several processes at once, so none can be measured | Company data flowing into employees' personal tools |
Four things that change when it is real
The work changes first. Each process is rewritten to say which steps AI does, which steps a person decides and which need a person's approval before moving on. This is called human-in-the-loop. Start with work that repeats every day and can be corrected if it goes wrong.
The data needs one version the company treats as correct, a source of truth, reachable through a system. If product prices live in three Excel files that disagree, the AI answers from whichever file it opens. You also need to know where personal data sits and which tools it may go into under Thailand's PDPA. Checking this before you start is called data readiness.
People change roles. Staff who used to key in data become the ones who check the items the system is unsure about, and supervisors set the criteria for what passes. In Thai companies, a new system often goes unused for down-to-earth reasons: it reads Thai documents poorly, or staff worry it will replace them. Plan change management from day one. The skills your team needs are in 10 skills employees need so AI doesn't leave them behind.
Controls are written down: which kinds of data may go into which tools, which outputs a person must check before use, and a record of every run in an audit log. If staff already use personal AI apps with company data, known as shadow AI, publish a policy and provide tools the company manages first. A ban on its own pushes the use to places IT cannot see.
Hypothetical case
An auto-parts distributor in Chonburi (name and details invented) bought AI licenses for every office employee and called the project its AI transformation. A quarter later, some staff were drafting emails with it, yet nobody could say what the company had gained. Accounts payable still typed each supplier invoice into the system, and someone had pasted a customer price file into a personal AI app.
On the second attempt, management picked one process: recording supplier invoices and matching them to purchase orders. The team measured time per invoice and the number needing correction before starting, and the head of accounts took ownership of those numbers. The development team built a system that reads Thai-language invoices with OCR, matches them to purchase orders in the ERP and sends only the mismatches to a person. Meanwhile the company wrote its AI-use policy. After three months, management had before-and-after numbers and used them to choose the next three processes.
Signs your company is ready
You can check this list within a week. With five or more in place, start the first process. With fewer, the missing items are the first job.
- A task repeated every day takes several staff hours a week, and you can count how many items are handled per day
- The data that task uses sits in a system you can extract from, or in documents you can collect into a set
- One supervisor is willing to own that task's numbers
- Management sets aside time for the people who do the work to help design and test
- You know which data counts as personal data under PDPA
- Staff have started using AI on their own, which shows demand and says a policy is due
Start with one process you can measure
The starting point that works is one process with a problem you can put a number on today, such as quotations that take days to issue or customer questions the team answers over and over. Starting several at once spreads the team thin, and no process ends up with numbers clear enough to decide on. Before choosing, ask these four questions about the process you have in mind.
Four questions before you call it a transformation
- Question 1: which process? Name one process and say where it starts and ends, such as from receiving an invoice to scheduling payment.
- Question 2: which number? One number you can count today, the baseline, such as time per item or items needing correction.
- Question 3: who owns it? One person who answers for that number and has the authority to change the steps, the process owner.
- Question 4: where is the data? The system that holds it, how to extract it, and which parts must stay inside the company under PDPA.
Rule: if you cannot answer all four, the process is still a tool trial. Pick another process or fill in the missing answer first.
If you still cannot choose a department, the scoring method is in which department should start using AI first.
What the first 90 days should deliver
Ninety days is enough for one pilot if you measure from day one. At the end, the company should have four things in hand.
- One pilot system working on real cases, with numbers against the baseline for what improved and what did not - An AI-use policy that says which kinds of data may go into which tools and which work a person must check - A team that can use the system, knows what to check and knows whom to tell when the system gets something wrong - A decision on the next three processes, ranked by return, with the reasons the others can wait
If the first process does not work, stopping is a valid result: the company knows where it got stuck before spending on expansion. The phase-by-phase plan, with decision gates, the steering team and the tracking dashboard, is in A 90-day plan to make your company AI-First, for business owners.
For the development team · What IT needs in place before the pilot
Data access: a read-only service account for every system the pilot uses, extraction through an API or a scheduled export, and a test dataset with personal data masked. Identity: sign-in through the company's SSO, such as Microsoft 365 or Google Workspace, role-based permissions, no shared accounts, and an AI system that sees only the documents the signed-in user may see. Logging: input, retrieved sources, output, approver, and model and prompt versions, written to an audit log users cannot edit. Classification: label data with data classification and decide which levels may go to a cloud model and which need a private LLM. Measurement: capture the baseline the same way you will measure after launch, and test the system with real examples from your own work before it goes live, agreeing what counts as passing.
How we take a company through its first 90 days
DNA Maker starts by sitting with the people who do the work every day, assessing your data, systems and team, and ranking the work AI can help with by return. Before anything is built, you receive a readiness report, a ranked list of use cases, a 90-day roadmap with team, budget and KPIs, and a safe, lawful AI-use policy. Where a process should not use AI yet, we tell you plainly.
Then DNA Maker's own engineers build the first system until it runs on real work, with no subcontracting, and you see working software every week. It connects to Microsoft 365, Google Workspace or the ERP you already run, and data that cannot leave the company stays with a private LLM inside your own environment. Toward the end we train your team to use and maintain it. Scope and price are fixed before work starts, and source code is delivered as scoped. Our 500+ projects include banks, hospitals and large Thai groups. See the AI Transformation service page, or describe your process to Dream in the chat.
SOFTWARE ENGINEERING GLOSSARY
Software engineering glossary
| Term | What it is | A simple example | What executives should ask the development team |
|---|---|---|---|
| AI transformation | Changing how work is done so AI handles some steps in real processes, with measured results | AI reads invoices; people check only the ones that do not match | Which of our steps change, and which number measures it? |
| Pilot | Running the system on real work at small scale before expanding | Starting with invoices from one group of suppliers | If the pilot fails, where can we stop? |
| Baseline | The process's numbers before the system, kept for comparison | Average time to record one invoice | Do we measure the baseline the same way we will after launch? |
| Process owner | The person who answers for a process and can change its steps | The head of accounts, who answers for month-end close time | Who signs off that the new steps work? |
| Shadow AI | Staff using personal AI tools with company data without the company knowing | Pasting a customer price file into an AI app on a phone | Which of our data sits in tools the company does not manage right now? |