When to Put Your Trust in AI Recommendations
One could describe consumer trust in AI recommendations in Australia as practical, if not a little conditional and cautious. An Australian is happy to go along with an AI suggestion that is pertinent, verifiable and a time saver. Make the system opaque about its reasoning, or have it use personal data without proper consent and put on airs of knowing what is best, and they are not so amenable.
This is of some import given AI’s presence at every turn of the online shopping experience from product search and discovery to customer service and the checkout. The issue is not if Australians will make use of AI but if they have sufficient faith in the recommendation to buy, to click and compare and return for more. This is where e-commerce site personalisation with AI matters. I have had my share of cock-ups in the course of my work in tourism and retail, and I can tell you people are more forgiving of an error when you put it to them straight. Dominic Hawthorne here, and I see no reason why AI should be any different.
How Much Trust Do Australians Place?
You will not find a national figure that tells the whole story of Australian attitudes. It is a matter of the brand, the product and how much the customer is shown; also the fallout from being wrong.
Australians and AI recommendations are something of a sliding scale. A phone case is low risk and one can test it. But a mortgage, an investment, a health product or insurance policy is a different kettle of fish altogether.
The Stakes Determine Trust
When consumers feel in command they are more disposed to accept AI. They want the option to put the brakes on, alter the filters, talk to someone face to face if it does not seem right. With an online purchase there are small ways to earn trust: a size call that is on the money, a good explanation, a return policy in plain sight and the sense the result has been made for you. One wayward recommendation is of no consequence. Put out a string of irrelevant ones and the system is rough as guts.
| Recommendation Type | Likely Customer Response | Proof of Trust |
|---|---|---|
| Low-cost accessories | Open to it | Easy to return and compare, a simple case to make |
| Groceries and household items | Of use if the preferences are real | Data controls and editable preferences |
| Big ticket items | Will want to do their own research | Have verified info, reviews and support on hand |
| Matters of finance or health | Cautious, will look for an expert | Need evidence, accountability and human oversight |
What Has Australians Wary
They are not opposed to AI per se, just systems that project confidence yet are unaccountable. A chatbot may be persuasive and still miss the mark with the customer or be driven by a commercial imperative. Then there are the broader issues of deepfakes, scams, automated decisions and privacy which colour one’s view of generative AI. Having seen an AI make up an answer, a customer is within his rights to question whether a product recommendation has been vetted.
What is Often Missed
People do not think of AI as neutral. They know retailers are in the business of selling. What is lacking is disclosure. Is the recommendation down to an algorithmic prediction, stock levels, popularity or the fact it is sponsored? Customers want to be told. “Recommended for you” is a vacuous phrase. Better to say in plain English “Suggested because you put a wide fit waterproof hiking shoe in your sights.” That way the customer has something to go on instead of having to boil the billy and trust the smoke.
AI and the Customer Journey
These days AI is part of the path to purchase in more than a carousel. It can dictate what is seen first, what is compared and if the shopper goes through with the checkout. The experiences that are trusted employ AI as a guide rather than a hard sell. There should be no automated funnel holding the customer back from changing course or looking at the evidence.
Discovery to Checkout
In the course of a shop AI can whittle down a catalogue to a shortlist, or at comparison point point out the trade-offs and compatible products. At the point of checkout it can be of service in putting forward delivery options or an accessory, assuming it is warranted. Digital shoppers need a clear run from the search to the product details and on to the delivery and return terms. There is little room for error with an AI assistant. Should it put a person on a page that goes nowhere, obscure the means of contact or make returning something a chore, any trust will be eroded in short order.
One-Day Shopping Scenario
When a customer is under the gun to decide in a day, one should put the reasoning for the recommendation front and centre. It is the one section to put first. Put the customer in the picture as to what information was at hand, which other options were put to one side and where there is still some doubt.
Mobile devices are no different; in fact they are more so when someone is making comparisons between meetings or on the train while a child wants a snack. A brief explanation will do, one that does not have the salient caveats lost in a mountain of marketing copy but puts the comparison in plain view.
What Makes Recommendations Trustworthy?
An AI recommendation that is to be trusted is relevant and can be challenged, and it is explainable. There is no pretence of an algorithm having personal wisdom; useful evidence is put forward and the customer is left to make the final call. For a brand this is a matter of being understandable, not of having an AI system come across as clever. A fair dinkum reason is called for, not a technical lecture.
Five Practical Trust Signals
- In plain language, give the chief reason for the recommendation.
- Indicate if popularity, sponsorship, availability or a person’s history has had a bearing on the result.
- Customers must be able to opt out of personalisation, clear their history and put in their own preferences.
- Rely on verified data for stock, prices and policy.
- Where a decision is of high risk or complexity, put a straight line of sight to human expertise.
Then there is the question of editorial provenance. An AI assistant may summarise buying advice but the customer ought to know the source of it and if a qualified individual has been over it. With safety, health or legal rights in play, verified information is all the more important.
Transparency Privacy And Oversight
You will not find transparency in a privacy policy tucked away in the dunny at the foot of a site. It is what the customer is shown in practical terms when an AI makes a recommendation or uses his information. Data privacy and personalisation have to be in sync. The onus is on the business to make clear what data is taken, for what purpose and for how long, and how a customer might alter his choices. If after a long day a normal person cannot make head or tail of the explanation, it is failing.
Keep A Human In The Loop
The value of human oversight is in the serious consequences, in a disputed outcome or when the system is not sure. Someone should be on hand to put right bad data, review a problematic recommendation and see patterns an automated system would overlook. This is not to say a staff member is to be manually checking every low-value product suggestion; it is about the organisation having accountability and the processes to escalate. With agentic commerce and autonomous agents one has to be even more circumspect, for a system can go from making a recommendation to acting on it by putting an item in a basket or ordering it.
Building Trust Through Personalisation
In e-commerce the best AI personalisation is like a good shop assistant, not some stranger going through your drawers. The idea is to have enough reliable information to be of use, not to amass every detail. Those looking to personalise their e-commerce site with AI would do well to begin with a narrow use case such as product discovery or compatibility advice, which are easier to put to the test than full autonomy. See if the help is appreciated by the customer, not just if clicks are up.
Expectation Versus Reality
It is expected that sales will follow in the wake of greater personalisation. The reality is that the wrong kind of it can leave a customer feeling manipulated or as though he is being watched. One might think a sophisticated model is the answer to poor product data. But an AI working from inaccurate descriptions or old stock will put out nonsense with confidence. More often than not better source material and some regular testing will be of more service than another bit of technical sparkle.
Choose A Safer Starting Point
A brand would be wise to define the customer problem before concluding AI is the tool for it. A pilot could be run to recommend compatible products on the basis of what the customer has supplied, and the business can then put the accuracy, return rates and complaints to the test.
The Growth Distillery is one of the entities of note in this area but one should not assume performance figures without the evidence to back it up. That is the way of things; even a trusted brand has to put its AI to the proof.
AI In High-Stakes Decisions
One does not place the same trust in an AI shopping assistant as one would in a financial or health recommendation put out by an automated system. Where there is greater scope for harm, there must be correspondingly more accountability and verified information from human hands.
This calls for a degree of care on the part of health providers, insurers and financial institutions in any system that bears on treatment options, affordability, eligibility or service access. A well turned-out interface will not of itself render a high-stakes decision safe.
When to Steer Clear of Automation
There are times when a consumer should be wary of an AI system: if it is proffering an urgent investment, telling them to forgo professional counsel, asking for sensitive data or putting forward a decision with no avenue for appeal. One can not put blind faith in an automated suggestion in such circumstances.
AI stocks present a case in point. A recommendation has a way of turning into financial advice in a hurry and there is no definitive list of “best” AI stocks. An investor would do well to look at risk, fees, diversification and his own situation before making a move. Markets are fluid and what was a shrewd choice yesterday may be a financial nuisance today.
Consumer Data Rights and Regulation
While regulatory frameworks set the parameters for responsible AI, they are not in themselves a source of consumer confidence. Organisations have to have people on hand to answer for a system when it fails, as well as accurate records and unambiguous processes.
The Consumer Data Right is of some import in that it confers rights on eligible consumers as to the use and sharing of their data. It is incumbent on businesses to put their data practices in terms a customer can follow and not see consent as something to put a tick against and put out of mind. The Australian Competition and Consumer Commission’s guidance on the matter is where to go for up to date information on data sharing and participation. Before rolling out a data-driven service, businesses would be wise to review the latest from the official sources, requirements being what they are.
A Checklist for Trust
- What should the AI be recommending and what is off limits? That needs to be defined.
- Make a note of the assumptions, permissions and data sources at work in the system.
- Put the recommendations to the test in a variety of situations and with different customers and devices.
- Provide the customer with a means of control and explanation, as well as a person to speak to.
- Keep an eye on bias, privacy issues, errors and any other unanticipated results.
The private sector could learn from the Artificial Intelligence in Government and Public Sector Strategy of the Australian Government which makes plain the expectation of responsible governance for the public good. Responsibility is not something to be ceded to a model.
Frequently Asked Questions
Given that AI recommendations are somewhere between an advertisement, personal decision-making and convenience, these are questions we get a lot. The short answer will serve but the customer should make sure he is familiar with the product and data practices in question.
Do Consumers Have Faith in AI?
In the context of a low-risk shopping decision that is transparent and easily put aside, some do. But if the system is opaque, uses sensitive data or has a bearing on a major decision, trust wanes. It is a matter of context.
What of the 30% Rule?
You will not find a “30% rule” in Australia that is universally accepted as a measure of an AI’s trustworthiness. To use it as a benchmark a business ought to lay out the method, sample and base for the decision it is to underpin. Otherwise it is like a road sign with no direction.
Best AI Stocks in Australia?
No list is without risk or can be called the best. Depending on the company, its debt, revenue and the sector, AI investments are subject to variation. Do not mistake an online list for your own financial advice; check the disclosures.
Is Bunnings the most trusted brand?
Ask the surveyor and you will get an answer that hinges on the category, year and sample. One study may put a brand at the top and another will not. The lesson for AI is that marketing copy does not create trust, consistent experience and fair dealing do.
Can an AI be free of bias?
Assume nothing of the sort. There will be gaps in the data or in the design and business objectives will show through. A sensible organisation will offer human review and a way for the customer to put a recommendation right.
Trust Is What Matters
An Australian will continue to use AI recommendations so long as the advantage is evident and he is in the driving seat. The better systems have the sense to know when human input is required and they will back their reasoning with verified information.
For the shopper the rule of thumb is to let AI do the comparing and then take things slowly where privacy or money is concerned. As for businesses, trust is no ornament. Without it personalisation is just a sales pitch.
Dominic Hawthorne
