Southeast Asia’s SMEs do not have an AI problem. They have a record problem

Southeast Asia's small and medium-sized enterprises (SMEs) do not face an artificial intelligence (AI) problem; instead, they struggle with record-keeping issues. A founder of a distribution business in Colombo demonstrated a sophisticated demand forecasting dashboard to the reporter, but the data it used was incomplete and inaccurate. This is a recurring issue across the region, not limited to a single poorly-managed company.

The regional conversation surrounding AI adoption focuses on tools, budget, and speed, but these variables are not the primary concerns. AI multiplies the quality of a business's existing operational record. When the record is accurate and comprehensive, AI can greatly enhance its effectiveness. However, when the record is incomplete or unreliable, AI produces confident yet misleading results that appear more authoritative due to its presentation as a dashboard.

The issue escalates as businesses grow in size. SMEs with around twenty to two hundred employees have enough transaction volume to necessitate a system but lack the process discipline to build one. Multiple conflicting versions of the truth exist within these businesses – the accounting package, the sales team's spreadsheet, and WhatsApp conversations where actual decisions are made. Adding an AI layer on top of this situation does not resolve the inconsistencies; it simply chooses one version to commit to.

At the regional operator scale, record fragmentation occurs due to differences in currencies, invoicing conventions, tax treatments, and languages in free-text fields. Reconciliation instead of intelligence becomes the key constraint. Only at national and enterprise scales does the problem reverse. These larger firms possess the record and face governance challenges rather than capture issues. The tools sold to SMEs often address the wrong problem.

Another critical aspect often overlooked in AI proposals is the cost. The expensive part of AI for mid-sized Asian businesses is not the license; rather, it is the reconciliation work required to standardise product naming, close gaps in the ledger, and ensure multiple systems agree on quarterly performance. This labour-intensive task is not glamorous, cannot be outsourced, and appears on the payroll line rather than the software line, which is why it usually remains unaddressed in business cases.

Furthermore, most off-the-shelf forecasting models assume that seasonality occurs on fixed calendar dates. However, in Southeast Asia, many important festivals shift over time, causing these models to mispredict crucial sales periods consistently. One example given is a distributor who overstocked for a festival that had already passed, demonstrating that the model's mathematics were correct, but its regional assumptions were flawed.

Southeast Asia's small and medium-sized enterprises (SMEs) constitute the overwhelming majority of enterprises in the region, accounting for between 97 and 99 percent of businesses and approximately 85 percent of employment. Yet, they contribute only around 40 percent of the GDP. This productivity gap is not due to a lack of AI capability but rather a lack of robust systems.

The capital intended to address this gap has significantly decreased. A recent report from Google, Temasek, and Bain indicates that private funding in the ASEAN region reached a decade low in 2024.

For businesses heavily reliant on WhatsApp orders, cash margins, and relationship pricing, the recommended AI implementation sequence is reversed. First, they must digitise the transaction record before acquiring any AI model. Creating a single source of truth for orders, standardising product naming, and maintaining a closed ledger each month are essential steps that make subsequent AI integration possible. This process takes two quarters and is considered boring but necessary.

If a business already has a clean record due to selling through platforms or fintech rails, the approach should be different. Instead of acquiring more tools, investing in capability enhancement is preferable. Acquiring a competitor with existing customers and poor systems, and implementing a working data substrate beneath it, can be a more advantageous strategy in a thin funding market.