Over 70% of Thai SMEs Now Use AI — So Why Doesn't It Show in the Numbers?

What the UOB Business Outlook Study 2026 actually says

Before analysing anything, it helps to put the numbers on the table. The UOB Business Outlook Study 2026 was published on 30 June 2026 and draws on responses from 265 business owners and senior decision-makers across Thailand. It places Thai businesses among the fastest AI adopters in the region, alongside several other signals about overseas expansion and supply-chain management.

One detail deserves attention: every figure here is self-reported by respondents, not audited financial data. That does not make the numbers wrong. It does mean they measure executive perception, which can form long before the effect ever lands in a profit-and-loss statement.

"Using AI" means at least two very different things

When two executives both say "we use AI", both are telling the truth, yet they frequently describe entirely different activities. In the first case an employee opens a chat assistant to help draft a message — personal, immediate, useful. In the second, AI reads the company's actual records, such as 24 months of sales by customer or current stock by item, and proposes what to do next. That second kind is slower to arrive and can be measured in money.

This distinction explains the survey result directly: the first kind genuinely saves people time but does not change the cost structure; the second kind changes the cost structure but requires clean data waiting for it first. Most organisations are firmly in the first category, and there is nothing wrong with that — it is simply an early rung on a ladder.

Three reasons the results have not reached the accounts yet

One: saved time does not convert into money by itself. If an accounts clerk saves 30 minutes a day by having AI draft correspondence, that time is usually absorbed by the backlog rather than removed from the expense line. Turning it into a financial result requires a deliberate decision about what the recovered hours are used for — revenue work, or closing a risk that was previously ignored.

Two: there is no baseline to compare against. Many organisations start using AI without first recording how many days the monthly close took, what unit cost was, or how much stock had been sitting for more than 90 days. Without a baseline, genuine improvement cannot be demonstrated. It is the same underlying condition that produces the gap between the reports management wants and the reports the system actually produces.

Three: the data AI needs is not where AI can reach it. This is the heavy one. In a typical mid-sized company, material cost lives in a production spreadsheet, stock balances live in a warehouse file, purchase prices live in the buyer's purchase orders, and the accounting entries are a fourth version again. With no central set of records, AI can only answer questions about whatever a human pasted into it — which is precisely why the answers feel shallow.

A five-rung data readiness ladder

The practical approach is to work out which rung the organisation is standing on, and only then decide what to ask AI to do. Attempts to skip rungs usually end with a project that demos beautifully and is quietly abandoned, because the answers it produces do not match the numbers finance is holding.

Start with the supply chain, not with paperwork

The most interesting figure in the UOB Business Outlook Study 2026 may not concern AI at all: over 90% of respondents rank supply-chain management as a priority, and 75% plan to diversify their suppliers. Those two answers point at exactly where AI can produce a financial result fastest — and simultaneously at where accurate data matters most.

Diversifying suppliers means comparing price, lead time, defect rate and payment terms across several vendors at once. If purchase history is spread across each buyer's personal files, AI can do little beyond formatting a nicer table. If purchase orders, goods receipts and supplier invoices are linked in one system, a question such as "which supplier missed the agreed lead time most often this year?" becomes a few minutes of work. It is the same root cause that makes stock balances in the system disagreeing with physical stock a bigger obstacle than most organisations estimate.

For the SMEs planning overseas expansion within two to three years, complexity rises another level immediately: multiple currencies, multiple warehouses, and different tax treatments. That is why many organisations choose to settle the back office before expanding rather than expanding first and patching afterwards — the argument we set out in five reasons SMEs need ERP.

Where ERP fits — and where it does not

To be direct: ERP is not AI, and installing ERP does not make an organisation smarter on its own. What ERP does is move the organisation from rung two or three of that ladder up to rung four or five, which is the precondition for AI answering business questions reliably rather than plausibly.

Saeree ERP consolidates cost, inventory and purchasing data into a single PostgreSQL database. Administrators inside the organisation can set per-menu user permissions, mark users inactive, and define a valid from–to period for each account themselves. Every change carries a record of who edited which entry and when, and the system exposes an API so the data can be taken into downstream analysis tools. That is preparing the raw material — not a claim that the system reasons on your behalf.

Saeree ERP's own AI Assistant is still in development and in training; it is not available for production use, and we do not recommend that any organisation build a plan around it today. What we do recommend is making the data correct and singular first, because whichever AI assistant you eventually choose, that prerequisite does not change. Organisations still deciding what an ERP is meant to cover may want to start with what an ERP system actually is.

What to do this quarter

The useful next step is not launching a new AI initiative. It is measuring where you currently stand. These five items take a few weeks and require no capital budget.

• Inventory what AI is already being used for — separate surface-level tasks from anything that touches real company records.
• Write down the baseline today — days to close the books, unit cost, value of stock aged over 90 days, average days sales outstanding. Without today's numbers, next year proves nothing.
• Identify contradicting sources — find how many places hold the same figure, and decide which one the organisation treats as authoritative when they disagree.
• Set a data egress rule — state plainly which categories of data must never be pasted into external tools, and provide a sanctioned alternative.
• Pick one business question — for example, "which product had the lowest gross margin over the last twelve months?" — and time how long the answer takes to produce. That elapsed time is the most honest readiness indicator available.

Organisations that complete these five will know immediately whether their constraint is tooling or data. In our experience it is almost always the latter — the same condition that makes the monthly accounting close take longer than it should in so many companies.

Conclusion

The 70%-plus figure from the UOB Business Outlook Study 2026 is genuinely good news and should not be talked down. It says Thai businesses are not behind anyone in their willingness to try new technology. But the 58% reporting cost reductions and 44% reporting productivity gains (UOB Business Outlook Study 2026) remain executive perceptions, and they become entries in a financial statement only once AI is wired into real operational data.

This gap is nobody's failure. It is the natural sequence of technology adoption: experiment first, embed into processes second. What separates the organisations that cross it quickly is not a better choice of model — it is having cost, inventory and receivables data that is accurate, held in one place, and traceable. Whoever finishes that work first will benefit from the next generation of AI the day it arrives, without starting from zero.

"AI does not turn wrong data into right data — it only delivers the wrong answer faster. The work that comes first is not picking a model; it is making sure the company has one set of numbers."

- The Saeree ERP team

References

• ThaiPR.NET / UOB Thailand — Thai SMEs lead the region in AI adoption: UOB Business Outlook Study 2026 (30 June 2026)
• UOB Thailand — official website (publisher of the Business Outlook Study)
• UOB Group — research and Business Outlook Study reports

Last verified 20 July 2026. All statistics in this article come from the UOB Business Outlook Study 2026, which surveyed 265 business owners and senior decision-makers across Thailand.