Artificial intelligence has moved rapidly from experiment to expectation in ecommerce.
Retailers are being promised smarter search, automated merchandising, individualised customer journeys, faster content production and autonomous shopping agents. Yet many ecommerce teams are still working with fragmented product information, disconnected customer data and platforms that make basic improvements difficult.
That creates a practical question: where can AI produce meaningful commercial value today, and where is it simply adding complexity?
For established retailers, the answer is not to introduce AI everywhere. It is to apply it selectively to high-volume decisions where better prediction, classification or automation can improve revenue, margin or operating efficiency.
Australian shoppers are already using AI
AI-assisted shopping is no longer a distant possibility.
Australia Post reports that 32% of Australians are already using AI for shopping advice, while six in ten are comfortable using the technology. Half believe AI will eventually become a normal part of online shopping. At the same time, Australian businesses appear more enthusiastic about autonomous commerce than their customers: 44% of businesses advocate for agent-led purchasing, compared with only 16% of shoppers.
That gap matters. It suggests retailers should prepare for AI-mediated shopping without assuming customers are ready to surrender control of the purchase.
Adoption within the retail industry also remains relatively immature. The Australian Bureau of Statistics found that only 9% of retail businesses reported using AI during 2024–25. Across all industries, larger and innovation-active businesses were considerably more likely to have adopted it.
The opportunity is real, but the market is not yet settled. Retailers still have time to build an advantage—provided they invest in useful capabilities rather than novelty.
1. Product discovery and site search
Product discovery is one of the clearest applications for AI in ecommerce.
Traditional site search depends heavily on exact keyword matching. A customer searching for “lightweight waterproof jacket for Tasmania” may receive poor results if the catalogue only describes products by brand, colour and category.
AI-supported search can interpret meaning as well as words. It can account for:
- Natural-language queries
- Synonyms and common misspellings
- Product attributes and intended uses
- Customer behaviour
- Inventory and availability
- Commercial priorities
This gives retailers an opportunity to make search behave more like a capable sales assistant.
The objective, however, should not be to install an “AI search” badge. It should be to improve measurable outcomes such as search conversion rate, revenue per search session, zero-result frequency and product discovery time.
Retailers should begin with their own search data. Repeated zero-result searches, frequent query refinements and high-exit search pages reveal where customers’ language and the product catalogue do not align. AI can help close that gap, but it cannot compensate indefinitely for missing attributes or inconsistent product information.
2. Product recommendations and personalisation
Recommendations are another established use case, but effective personalisation requires more than displaying products similar to those previously viewed.
Retailers can use AI to choose:
- Which products to recommend
- When to make the recommendation
- Which channel to use
- Whether to prioritise relevance, margin or stock position
- When not to personalise at all
The last point is important. Not every customer interaction needs to be individualised.
A returning customer may benefit from recommendations based on purchase history. A first-time visitor may be better served by bestsellers, strong category navigation and clear buying guidance. Excessive personalisation can narrow discovery, make the experience unpredictable or create uncomfortable assumptions about the shopper.
According to Adobe’s retail research, 65% of consumers value personalised recommendations, but only 41% believe brands deliver them effectively. The same research found that 87% consider responsible handling of personal data important, while just 46% believe brands meet that expectation.
The lesson is not simply to personalise more. It is to personalise when the retailer has sufficient data, a clear customer benefit and an appropriate level of consent.
3. Product data enrichment
Product data is less visible than generative shopping assistants, but it may offer a greater immediate return.
Large catalogues often contain inconsistent titles, incomplete specifications, missing attributes and supplier descriptions written in incompatible formats. These problems affect search, filters, comparison tools, recommendations, accessibility and organic visibility.
AI can assist with:
- Classifying products into the correct taxonomy
- Extracting attributes from supplier information
- Standardising units and terminology
- Identifying missing or contradictory data
- Drafting product descriptions and summaries
- Creating image alt text
- Translating and localising catalogue content
- Flagging potentially duplicated products
This is particularly useful when a retailer manages thousands of products across multiple suppliers or channels.
Human review remains necessary. Generated specifications must be checked against authoritative product data, and claims involving safety, compatibility, materials or performance require additional scrutiny.
The real benefit is not automatic copywriting. It is creating better structured product information at a scale that manual catalogue management cannot easily achieve.
That structured information will also become more important as shoppers use AI assistants to compare products. If an external system cannot reliably determine a product’s dimensions, compatibility, availability or delivery options, it is less likely to recommend it confidently.
4. Merchandising and promotion
Retail merchandising requires teams to balance customer relevance, inventory, margin, seasonality, promotions and brand priorities.
AI can help analyse these variables and recommend:
- Category-page product ordering
- Products to feature or suppress
- Cross-sell and bundle opportunities
- Promotion timing
- Markdown candidates
- Stock that should receive greater exposure
- Differences in demand across locations or channels
The strongest model is usually assisted merchandising, not fully autonomous merchandising.
A system may detect that a product is converting well, but it may not understand why the product should be held back for a campaign. It may prioritise immediate revenue while overlooking margin, supplier agreements, brand presentation or longer-term inventory needs.
AI should reduce the amount of time merchandisers spend finding patterns and performing repetitive changes. Commercial teams should retain control of objectives, constraints and exceptions.
5. Customer service
Customer service is often the first place retailers experiment with generative AI. It is also one of the easiest places to damage trust.
AI can be useful for routine, well-defined interactions such as:
- Finding an order
- Explaining delivery options
- Checking return eligibility
- Answering product questions
- Summarising previous conversations
- Suggesting responses for service agents
- Directing customers to the right team
It becomes riskier when the customer has a complex complaint, the source information is unreliable or the system can take actions without appropriate controls.
A useful service assistant should be grounded in current policies, product information and order data. It should disclose its automated nature, distinguish verified information from generated guidance and make escalation easy.
Success should be measured through resolution rate, customer satisfaction, repeat contact and escalation quality—not simply by the number of conversations deflected from human agents.
Automating a frustrating experience does not make it better. It only makes the frustration cheaper to deliver.
6. Content operations
Generative AI can accelerate the production of product summaries, category introductions, campaign variations, email subject lines and internal briefs.
Used carefully, this can remove repetitive work and allow content teams to spend more time on propositions, creative direction and customer insight.
The temptation is to turn that efficiency into a dramatic increase in publishing volume. That generally produces generic content, duplicated ideas and a growing editorial maintenance burden.
Retailers should instead use AI to improve a controlled content operation:
- Begin with verified product and brand data.
- Generate a structured draft for a defined purpose.
- Apply brand, legal and merchandising rules.
- Review important claims.
- Measure whether the content changes customer behaviour.
AI can make content production faster. It does not decide whether the content deserves to be ranked.
7. Forecasting and operational decisions
AI also has applications behind the storefront.
Demand forecasting, inventory allocation, anomaly detection and customer segmentation can improve operational decisions, especially when retailers manage multiple channels or locations.
These models can help teams identify:
- Unexpected changes in demand
- Products at risk of overstock or stockout
- Regional variations
- Customers likely to lapse
- Unusual return patterns
- Promotion and pricing effects
The limitation is data quality. Forecasts trained on incomplete history or unusual trading periods may produce confident but unreliable conclusions.
Retailers should therefore treat AI-generated forecasts as decision support. Teams need to understand the inputs, test performance against a baseline and establish when human review is required.
Where retailers should be cautious
Not every AI initiative deserves investment.
A generic chatbot without reliable data
A chatbot connected to weak product information and outdated policies gives customers faster access to incorrect answers. Fix the information layer first.
Personalisation without a defined objective
“Creating a more personalised experience” is not a sufficient business case. The retailer should identify the behaviour it expects to change and the metric it will use to prove the change.
Mass-produced SEO content
Producing hundreds of lightly differentiated pages may increase the quantity of indexed content without improving its value. It can also create duplication, factual errors and substantial future maintenance.
Fully autonomous commercial decisions
Pricing, refunds, promotions and inventory movements can carry significant financial and reputational consequences. Autonomous action requires explicit limits, monitoring and a reliable path for intervention.
Technology selected before the use case
AI should not begin with a platform demonstration. It should begin with an expensive, repetitive or poorly performing customer or operational decision.
The data foundation comes first
Retailers often describe AI as a technology project. In practice, it is usually a data and operating-model project.
Search, recommendations, content generation and shopping agents all depend on accessible, accurate information. That includes:
- Structured product attributes
- Current prices and promotions
- Location-level inventory
- Delivery and returns information
- Customer consent and preference data
- Consistent identifiers across systems
- Clear ownership of data quality
Salesforce’s research found that 88% of retailers expect unified commerce to have a significant effect on their goals. That reflects a broader reality: AI is most useful when customer, product, order and inventory information can work together.
Without that foundation, adding another intelligent interface may simply obscure the same underlying fragmentation.
How to prioritise an AI initiative
A retailer does not need an enterprise-wide AI transformation to begin.
A more effective approach is to select one use case with:
- A meaningful volume of decisions or interactions
- A measurable commercial or operational problem
- Sufficient reliable data
- A clear business owner
- A low-cost path to human review
- A credible control group or baseline
For example, an ecommerce team could use AI to improve product attributes within one category, then measure the effect on filter use, zero-result searches and conversion. A service team could test response suggestions for one enquiry type while tracking accuracy, handling time and customer satisfaction.
The technology should prove its value within a limited scope before it is extended.
What should retailers invest in first?
For many established retailers, the priority order should be:
- Product, inventory and customer-data quality
- Product discovery and search
- Catalogue enrichment
- Assisted merchandising
- Service-agent support
- Controlled personalisation
- Customer-facing autonomous agents
This sequence is deliberately unglamorous. The earlier capabilities create the information and controls required by the later ones.
Autonomous shopping may ultimately change how products are discovered and purchased. Australia Post reports that 85% of businesses are already taking steps to prepare for that future. But visibility to shopping agents will depend on familiar fundamentals: structured product information, accurate availability, transparent pricing, dependable fulfilment and a trustworthy brand.
The competitive advantage is not AI alone
Most retailers will eventually have access to similar models and software.
The durable advantage will come from what those systems can learn from: better product data, stronger customer understanding, connected operations and clearly defined commercial rules.
AI can help customers find the right product, help teams make faster decisions and remove large amounts of repetitive work. It can also scale poor information, weak processes and bad judgement.
The retailers that benefit most will not be those that adopt the greatest number of AI tools. They will be the ones that choose valuable problems, build trustworthy foundations and measure the results.
