- AI in a company comes down to four questions: why, where, how it works and how to do it. Answer them in that order before you choose a tool.
- Start with one job that repeats every day, in one department, and write down its numbers before anything changes.
- Before it goes live, someone owns the job, a person approves the risky steps, and there is a rule for which data may go to AI.
- Owners and IT read in a different order, then choose the first job together.
AI in a company starts with four questions
Most owners start with which AI tool to buy. Most IT teams start with how it will connect to the systems they already run. Both questions matter, and both come too early. Before choosing a tool, a company needs four other answers: why use AI and how to measure it, which job to use it on, how the system works, and how to put it in place.
This guide has nine chapters. This one is the map; the other eight answer the four questions in turn.
Why: benefits you can put a number on
The first question is what AI will improve and how you will know. Most measurable gains sit in work that repeats every day: documents retyped into a system, customer questions answered the same way each time, and approvals waiting on a desk.
Chapter 1, 7 benefits of AI in a company you can put a number on, splits the gains into seven, each with what to measure and how to take the reading before you start. Read it alongside How much is manual work costing your business each month?. Those numbers become the base for the ROI your board will ask about later.
Where: jobs in each department
Once you know what to measure, the next question is where. Sales wants quotes out faster. Accounting wants to stop retyping Thai-language invoices. Purchasing wants to compare supplier prices sooner. HR wants staff to find answers about benefits on their own.
Chapter 2, AI use cases in business, by department: sales, accounting, purchasing, HR, collects examples from each department and separates the jobs that can use AI today from those that must wait until the data is ready. The scoring method for picking the first department is in If you want to cut staff costs, which department should start using AI first?.
How it works: LLMs, RAG, agents and the system around them
This is the part people skip and meet again after signing. An owner should know what parts an AI system has, because each part carries its own cost and risk.
Chapter 3, How AI apps work: LLM, RAG and AI agents, explained through a real app, explains the three main parts. An LLM is the language model that reads and writes text. RAG makes the system search your company's documents before it answers, so the answer comes from your data. An AI agent takes one job and opens the systems it needs until the job is done.
Chapter 4, The 7 layers of an enterprise AI system, and what IT decides in each, is written for IT. One AI system has seven layers, from the channel, such as LINE OA, through the model, the data and the controls, to measurement. Each layer has a question to settle before building. If customer data cannot leave the company, the option is a private LLM; see private AI options for companies whose data can't leave the building.
Chapter 5, AI agent, agentic AI and generative AI: what is the difference?, separates three terms from vendor decks. Generative AI drafts and stops. An AI agent finishes one job with the permissions you give it. Agentic AI coordinates many steps and systems inside limits people set. The chapter has a comparison table and says when an ordinary chatbot is enough.
How to do it: set rules, change the work, hire
The last question has three chapters, because putting AI in place means three pieces of work running side by side.
Chapter 6, An AI usage policy your staff will follow, and what Thailand's PDPA asks of it, belongs before your first system goes live, because in many companies staff already use AI that IT has never seen. A one-page policy says which kind of data may go into which tool, who approves a new tool, what happens when someone makes a mistake, and what Thailand's Personal Data Protection Act (PDPA) requires.
Chapter 7, What is AI transformation? Where to start and what the first 90 days should deliver, separates buying tools from changing how the company works. AI transformation changes processes, roles and the way results are measured, with AI inside daily work. The chapter says what the first 90 days should end with. The plan, phase by phase, is in A 90-day plan to make your company AI-First.
Chapter 8, App development FAQ: 20 questions before you hire, collects what buyers ask before signing, in four groups: price and time, scope and technology, working together, and after handover. Put the same set to every developer. What to prepare on your side first is in What to prepare before you hire an AI developer.
Hypothetical case
A family-owned frozen-food maker in Nakhon Pathom (names and details invented) supplies restaurants. In the first quarter, management worked through why and where. They timed how long sales spent taking orders by LINE and typing them into the system. Order taking became the first job, because it repeats every day and the sales manager agreed to own it.
In the second quarter, IT found that the old sales system had no API, so a connection had to be built first. The company issued a one-page policy at the same time, because some staff were already pasting customer lists into personal AI chat apps. In the third quarter, an agent took orders from LINE OA and drafted the sales order in the system for sales to confirm. The first mistake was units: a customer ordered by the case and the system recorded bags, so the team added a rule to ask back whenever the unit is unclear. In the fourth quarter, the company compared its numbers with the ones taken at the start of the year, then chose supplier invoices in accounting as the second job.
Which order to read in, by role
Owners and IT decide different things, so they read in a different order and meet when they choose the first job.
The four-question map
- Why Chapter 1. Output: a before number for the job you will measure.
- Where Chapter 2. Output: the first job in one department, with its owner's name.
- How it works Chapters 3, 4 and 5. Output: the systems to connect, which data may leave the company, and the level of the job.
- How to do it Chapters 6, 7 and 8. Output: a one-page policy, a 90-day plan and the questions for developers.
Order for owners: chapters 1, 2, 5, 7, 6, then 8.
Order for IT: chapters 3, 4, 5, 6, then 8.
Reading rule: if "why" has no number yet, hold off on choosing a tool.
| Chapter | Who should read it | Question it answers | Decision it supports |
|---|---|---|---|
| 1 Benefits you can put a number on | Owners, finance heads | What will AI improve, and how is it measured? | Which numbers to take before starting |
| 2 Use cases by department | Owners, department heads | Which jobs in our department can use AI? | The first job and its owner |
| 3 How AI apps work | IT, executives | How do LLMs, RAG and AI agents differ? | Which parts the first job needs |
| 4 The 7 layers of an enterprise AI system | IT | What does an AI system connect to in the company? | Which model, where the data sits, who approves |
| 5 AI agent, agentic AI, generative AI | Owners, IT | Which level is a vendor offering? | Which level the first job should be at |
| 6 AI usage policy | HR, legal, IT | Which data may go into AI under the PDPA? | A one-page policy ready to announce |
| 7 What AI transformation is | Owners, senior management | How does it differ from buying tools? | What the first 90 days should end with |
| 8 20 questions before you hire | Owners, purchasing | What should we ask a developer before signing? | A developer chosen on comparable answers |
When you are ready to build
Before you talk to a developer, check this list. If a line is still open, go back to the chapter that answers it.
- One first job in one department, with the name of the person who owns it
- A before number for that job, such as time per case
- You know which systems the job touches and whether they have an API
- Agreement on which kinds of customer data may leave the company and which must stay inside
- An AI usage policy that staff know about, even a one-page one
- You know which steps need a person's approval, and whose
- One set of questions to ask every developer
For the development team · What IT should read first
Start with chapter 3, so that LLM, RAG and agent mean the same thing inside your team and with vendors. Then read chapter 4, because the hardest decisions to reverse sit in the data and control layers: which system holds the master copy of the data, whether it has an API or needs RPA, whether the model runs in the cloud or inside the company, and where the audit log is kept. Before meeting a developer, prepare a list of systems with their APIs, a map of how personal data flows, and the narrowest set of permissions the agent needs. Then read chapter 6, because IT is the team that makes the data policy real in the systems.
How we work with you along this path
The path in this guide is the way DNA Maker works with clients. We start by listening to the people who do the work every day, because the rules a system has to get right live with them. We write what we hear as process steps, rules, the data each step uses and the points where a person decides. That consulting stage answers the first two questions on the map, and we say plainly when a job is not ready for AI yet.
Once the first job is agreed, scope and price are fixed before work starts. You approve a prototype of the screens, we build in short sprints with a demo every week and a senior engineer reviewing the work, then test, go live, train your staff and keep looking after the system on a monthly care plan. Everyone on the project is on the DNA Maker team, with no subcontractors. We do this kind of work for banks, hospitals and large Thai groups. See AI Transformation and AI App Development, or describe your job to Dream in the chat and get a first estimate in minutes.
SOFTWARE ENGINEERING GLOSSARY
Software engineering glossary
| Term | What it is | A simple example | What executives should ask the development team |
|---|---|---|---|
| LLM | A large language model that reads and writes text | The part that drafts replies to customers in LINE OA | Which model do we use, and can we switch? |
| RAG | Making the system search company documents before it answers | Staff ask about leave and get the answer from the latest HR rules | When a document changes, who updates the system? |
| AI agent | Software that takes one job and finishes it, within the permissions given | An agent takes an order from LINE OA and drafts the sales order | Which systems can this agent touch, and who granted that? |
| ROI | The return compared with the money invested | Hours saved each month against the cost of the system | What is our before number, and who measures it? |
| Private LLM | A language model that runs on the company's own systems | Customer data never leaves the company's servers | Which of our jobs truly need a private LLM? |