AI-Washing in APAC: How to Tell Real AI from Marketing Copy

Every software vendor in APAC now claims AI. Here is how to tell who is actually using it.

In the past 18 months, the phrase "AI-powered" has appeared in the marketing materials of virtually every enterprise software vendor selling into the APAC market. ERP vendors. CRM vendors. Document management vendors. Payroll software vendors. Everybody has AI now.

Most of them do not.

This is not a new phenomenon — "AI-washing" (claiming AI capability that does not exist or is not material) has been a pattern since the first machine learning hype cycle in 2017–2019. What is new is the scale: the GPT moment in 2023 gave every marketing team a reason to rewrite their copy, and a substantial proportion of APAC enterprise software has been retroactively described as AI-powered without any underlying change to the product.

This matters because procurement decisions are being made on the basis of AI capability claims that are, in many cases, false.

What vendors actually mean when they say "AI"

There are five categories of what vendors mean when they say AI, listed from least to most technically substantive:

Keyword and rules-based logic — The system matches inputs to predefined rules. Nothing learns. No model exists. This is not AI by any definition. Many legacy document routing and customer service chatbot systems fall here.

Robotic process automation (RPA) — Software bots that follow deterministic scripts to automate repetitive tasks. Sophisticated and genuinely useful, but not AI. UiPath and Automation Anywhere products are in this category. Good vendors do not call this AI; less scrupulous ones sometimes do.

Classical machine learning — Statistical models (decision trees, random forests, gradient boosting) trained on structured data. Real machine learning, but not generative AI and not large language models. Predictive analytics tools often fall here. Legitimate AI, but often overstated.

LLM API wrapper — A thin integration over a third-party LLM API (OpenAI, Anthropic, Google) with a vendor-specific prompt. Technically AI, but the vendor has done minimal differentiation. The capability is almost entirely in the underlying model, not the product. Common in recently "AI-powered" document tools and email assistants.

Genuine ML/LLM differentiation — The vendor has fine-tuned models on domain-specific data, built retrieval systems with proprietary corpora, or developed model infrastructure that creates real differentiation from the underlying API. This is what AI-powered should mean.

Five questions to ask any vendor in a procurement conversation

These questions are designed to distinguish real AI from marketing:

1. What model powers this feature, and where is it hosted? A vendor with genuine AI capability knows the answer immediately. A vendor with an LLM wrapper may not know which model version they are on. A vendor using rules-based logic will give a non-answer.

2. What data was this trained on, and how often is it retrained? Real ML systems have training datasets and retraining pipelines. Rules-based systems do not train. If the answer is "it learns from your usage," press for specifics — what is actually changing, and on what schedule?

3. Can you show me a model card or documentation of model performance on a benchmark dataset relevant to my use case? Vendors with substantive AI work have evaluation frameworks. Vendors with marketing copy do not. The absence of this documentation is informative.

4. What happens when the model is wrong? Real AI systems have error rates and human-in-the-loop mechanisms. If the vendor cannot describe their error handling and correction mechanisms, the system likely has no mechanism — because there is no model that makes errors.

5. What is the data residency for the model inference requests? This is especially important in APAC markets with data localisation requirements (China PIPL, South Korea PIPA, Japan APPI). A vendor that cannot answer this question clearly either does not know their infrastructure or is using third-party APIs without understanding the data flow.

The APAC-specific dimension

AI-washing is more consequential in APAC than in Western markets for one reason: regulatory requirements around AI systems are tighter and more varied. Deploying a system that is actually a rules-based router as an "AI-powered decision system" may have governance implications if that system is audited under MAS, HKMA, or APPI frameworks that require documented model governance. The marketing claim becomes a liability.

For enterprises in regulated industries — financial services, healthcare, public sector — vendor due diligence on AI claims is not just about getting value for money. It is about understanding what governance obligations attach to the system being deployed.

For a structured framework for evaluating AI vendors, see the Enterprise AI Evaluation Framework.