Every AI project reaches the same fork early on. Do you build something made for your business, buy a tool someone else has already made, or leave it for now and look again later? Suppliers who build will tend to say build. Suppliers who sell software will tend to say buy. Nobody earns anything by telling you to wait, so that option rarely gets a fair hearing.
This guide sets out when each choice makes sense, and the questions that usually settle it.
When buying is the right answer
Buy when your problem is common and your way of handling it is not special. Plenty of businesses need to sort email, transcribe meetings, draft social posts or answer basic customer questions. Tools exist for all of these, and they improve every few months without you paying for the improvement.
Signs that buying fits:
- You can describe the job in one sentence and a supplier recognises it immediately.
- Your data can sit in the supplier’s system without breaking any promise you have made to customers.
- The tool works with the software you already use, or you are willing to change.
- You are comfortable with the tool’s way of doing things, even if it is not exactly yours.
The trade-off is control. You accept the supplier’s price changes, feature decisions and limits. Before you buy, check how you get your data out if you leave, and what happens to it after you close the account.
When building is the right answer
Build when the process is specific to you and important to how you make money. A microfinance company’s collections workflow, an exporter’s order confirmation process across several buyers, or a shop that sells through WhatsApp, a website and a counter at the same time, may all have details no off-the-shelf tool handles well.
Signs that building fits:
- You have tried existing tools and each one forces awkward workarounds.
- The process touches several of your own systems that need to talk to each other.
- You need tight control over what the AI can do, who approves what, and what gets logged.
- You want to be able to switch AI models later without starting again.
The trade-off is cost and responsibility. A custom system has a setup cost, a monthly running cost and someone has to maintain it. Ask any builder, before signing, who owns the code, where it runs, and what it would cost to hand it to a different developer.
The middle path
There is often a third route: buy a general toolkit, then configure or extend it for your process. This is cheaper than a full build and more flexible than a closed product. It also lets you learn what you really need before committing to custom work. Many businesses find the toolkit is enough, and some find it shows them exactly what a later build should do.
When waiting is the right answer
Waiting is a real decision, not a failure of nerve. Wait when:
- The process is not written down. If staff cannot agree how the work is done today, automate nothing yet. Write the process first.
- The data is not there. If records are on paper, in personal phones or spread across five spreadsheets that disagree, the first project is cleaning the data, not adding AI.
- The saving is small. If the task takes two hours a month, even a cheap tool may not pay back once you count setup and checking time.
- The risk is high and the tools are young. For decisions such as credit approval or anything regulated, today’s tools can support a person but should not decide. If the only version worth having is the one that decides, wait.
- Nobody has time to own it. An AI project with no owner drifts, then fails quietly.
Waiting well means setting a date to look again, usually six months out, and writing down what would need to change for the answer to become yes.
Four questions that usually decide it
If you are still unsure, answer these honestly:
- How much does the current process cost each month? Count hours and mistakes, in rupees. If you cannot work this out, you are not ready to decide.
- How different is your process from everyone else’s? Similar means buy. Different in ways that matter to customers means build or extend.
- What happens if it goes wrong? If the answer involves lost money, angry customers or a regulator, keep a person approving every important step whichever route you choose.
- Who will look after it in a year? If nobody, buy something simple or wait.
A worked example: say a small distributor spends about 40 staff hours a month re-typing orders from WhatsApp into its system. At LKR 800 an hour, that is LKR 32,000 a month. A bought tool at LKR 10,000 a month that handles most messages is a sensible first move. A custom build costing several lakhs only makes sense if the bought tool cannot cope with the way orders actually arrive.
The honest limits of this guide
No guide can replace trying things. A short, cheap trial of a bought tool often answers the build question better than any analysis. And prices, tools and models change quickly, so a “wait” decision made today may flip in six months.
If you want to talk through the decision with someone who will also say “wait” when it fits, you can get in touch. Either way, start by writing down what the current process costs you each month.