An AI Strategy Doesn’t Start With AI

For a lot of businesses, AI arrived before the strategy did. Someone started using a chatbot to draft product descriptions, a marketing platform switched on an AI feature, a staff member found a tool that summarised supplier emails. Each made sense on its own. Put together, they rarely add up to a plan, and it’s often hard to say what any of them is actually achieving.

That’s a common position to be in. The National AI Centre’s tracking of Australian SMEs found that 43% were using AI in some form over the summer quarter to February 2026. It also found encouraging signs of maturity: AI being embedded across several parts of a business was at its highest level in seven months, while one-off use was declining.

NAB’s research, surveying around 670 Australian SMEs late last year, put overall use at a similar 42%, but found retail well behind at just 22%, with nearly two-thirds of retailers not using AI at all. It suggests that for many eCommerce businesses, the larger opportunities, in operations, customer service, inventory and finance, are still mostly untouched. That’s where a strategy earns its keep.

For an eCommerce business, the list tends to look familiar: answering order status enquiries, keeping product content consistent across channels, reconciling orders between the storefront and the ERP, preparing quotes for B2B customers, spotting stock that’s about to run out or has stopped moving. Some of those are good candidates for AI. Some are better solved by a straightforward integration or automation that has nothing to do with AI at all. A good strategy is honest about the difference.

It also helps to rank each problem by two things: how much it would be worth to fix, and how much it would hurt if the AI got it wrong. Drafting internal summaries is low risk. Quoting a price to a customer or promising a delivery date is not. Starting with useful, low-risk work builds confidence and experience before the higher-stakes uses come along.

The next step is to look at the data each problem depends on, and this is where most eCommerce AI projects succeed or stall. An AI tool can only work with the information it can reach, and in many businesses that information is split across a storefront, an ERP, an accounting system, a marketing platform and a few spreadsheets that someone updates when they have time.

For each problem on the list, it’s worth asking where the data lives, whether it’s accurate, how current it is and whether a tool could reach it without someone exporting it by hand. An AI assistant answering stock enquiries from yesterday’s export will give confident, wrong answers. An AI drafting B2B quotes without access to customer-specific pricing will produce something nobody can send. In both cases the fix isn’t a better AI model. It’s connecting the systems properly first, which also happens to pay off whether or not AI is involved.

The National AI Centre’s tracking found that trust is the biggest reason SMEs hold back from AI, cited by around two-thirds of businesses with no plans to adopt it. It also found that only about half of businesses already using AI check its output before it reaches customers. Those two findings point in the same direction. A small amount of structure makes AI both safer and easier to trust.

The federal government’s Guidance for AI Adoption, released in October 2025, is a sensible place to start. It’s voluntary, and it boils responsible AI down to six practices: deciding who is accountable, understanding the impact on the people affected, managing risks, being open about how AI is used, testing and monitoring, and keeping meaningful human control. The guidance comes with a short policy template and an AI register template, a simple list of which AI tools the business uses and for what. For most SMEs, filling in those two documents would be a significant step forward. It also tends to surface AI use nobody knew about, along with any customer or commercial data that has been pasted into tools without much thought.

AI tools are changing quickly, and the best option for a task today may not be the best in six months. A strategy built around one vendor’s product is more fragile than one built around the business’s own systems and data. Where AI connects to the business through well-designed integrations, rather than being wired into every system separately, swapping one tool for another becomes a manageable change rather than a rebuild.

It’s also worth planning for AI arriving whether or not you choose it. Ecommerce platforms, ERPs and marketing tools are adding AI features at a steady pace, and customers increasingly use AI assistants to find and compare products. A strategy that covers what you buy, what’s switched on inside the tools you already use and how your products appear to AI tools outside your business will age better than one that only covers new purchases.

Finally, every AI project should have a simple measure attached before it starts: hours saved, enquiries resolved without escalation, errors avoided, faster quote turnaround. It doesn’t need to be precise. It just needs to exist, so that after three months the business can decide whether to expand the project, change it or stop.

An AI strategy for an eCommerce business doesn’t need to be long or ambitious. It needs a short list of real problems, an honest look at the data behind them, a few clear rules about who checks the output, and a way to tell whether it’s working. The businesses that get the most from AI over the next few years probably won’t be the ones that adopted the most tools. They’ll be the ones that knew what each tool was for.