Behavioural Finance, Consumer Behaviour, and the Algorithm

– David Wanis, September 2026

In the next 20 years, the use of advanced computational methods (“the algorithm”) is likely to change the nature of excess returns earned from consumer behaviour in the same way as it changed the alpha available from deep value investing over the past 20 years.

Value investing as we were taught growing up, was the discipline to buy unpopular stocks which were very cheap, and then over time waiting for them to become less unpopular and less cheap. This appeared easy to do when reading early Warren Buffett or Benjamin Graham, but much harder to do when faced with the humanity most investors possess. Cheap stocks tend to be unappealing. A weak business model, bad management, low growth, repeated underwhelming financial results, over-levered balance sheet, underperforming stock price, lack of interest by the market.

When the human investor had to pull the trigger despite these issues, it was easy to confuse the cause and effect. Notably, the reason the stock was cheap was because of all these poor attributes and hence there wasn’t really an opportunity here. And so, the average human investor passes, the stock stays cheap and only the rare investor sees the potential turn around and buys the stock.

Enter the age of quantitative investing. Most of the attributes of a cheap stock – whether fundamentals or price based – are quantitative. Quantitative value funds would systematically buy hundreds of these names, diversifying individual stock risk with small position size, and capture the value factor over time. The algorithm has no fear of looking bad (status loss to investment peers), of having to explain to a client why they bought the stock when every other fund manager agreed it was a dog. It simply followed the rules.

One perspective on why value investing got more difficult to generate the sort of excess returns it had historically was the alpha was available due to a uniquely human behaviour (over-riding shame and status loss) and that once quantitative value investing became a large enough part of the value investor market, this was no longer true.

In the same way the drone changed the nature of warfare, allowing asymmetry in the low cost of offensive drones against the very high cost of defensive missile shields, AI is providing the consumer with a tool to fight back against the corporate behavioural return harvesting machine.

The Algorithm, in this case Large Language Models (LLMs), can help consumers fight back against businesses whose excess return (like the value investor) comes from entrenched human behaviours. Over decades, many companies, with resources far exceeding most consumers, have invested in and optimized the consumer interface to their business to capture real economic value from consumer behaviour. Think about your own experience seeing these (dark) patterns at work:

Online Hotel Booking: “Only 1 left at this price”, “32 people are looking at this hotel right now”.

Gym Subscriptions: Low friction onboarding (click this link), high friction cancellation (come in person to the gym three suburbs away between 11-11.15am on the second Thursday of the month and speak only to Barry who can action your cancellation).

Retail Banking: when interest rates change, quick to price up mortgages, slower on term deposits. Attractive term deposits for new customers not offered to existing customers.

General Insurance: Claim processing frictions to reduce successful claim rates, contractual exclusions in the fine print.

Extended Warranties: Complex claim and uneconomic return processes to reduce warranty claim rates.

Airlines: A seat upgrade payment can be processed instantly in the airline app. A cancelled flight refund sends you to a Refund Request Webform which is complicated and manually painful to complete.

Subscriptions (PayTV, Telco, Streaming): Promotional pricing which rolls over to significantly higher rates in 12 months’ time, knowing a large percentage of new customers will forget to cancel and inertia will see contracted price uplifts stick.

Here is a real-world example. Late last year, my 82 year old mother was in a car accident – not her fault, another driver merged into her lane without looking. There was clear evidence of the other driver being at-fault (photos at the scene, position of both cars at rest after the accident, a text message from the other driver admitting fault, patterns of damage to both cars). The insurer, however, decided to use a strategy of claims assessor aggression and (empty) legal threats to avoid fault and remove their liability of a few thousand dollars. To be honest, under normal circumstances this may have worked because some of their tactics were completely unethical (claiming they had evidence they did not possess that indicated their insured was not at fault) and many 82 year olds whilst they have the time, do not have the confidence or accessible information to fight well-resourced national branded insurance companies.

Using AI, our approach was simple. As the insurer would call with their highly trained claims assessor and make these fictional claims (it was clear the goal was claim denial), my mother was under simple instruction – do not verbally respond, do not react, and simply ask for everything they say in writing via email.

Using a widely available LLM, we were quickly able to ascertain her consumer rights, the timeline required, the path of complaint via the insurance ombudsman (if required) and at every interaction the LLM would help draft an email back to the insurer – very factually, very politely, very unemotionally – detailing what she required from them.

Long story short, the claim was processed as originally desired, the claims assessor was sent on additional training (turns out falsely claiming you have video evidence as a negotiating tactic is not standard operating procedure) and the car was repaired with no out-of-pocket costs.

An LLM will read the entire contract and all terms and conditions. An LLM can understand state and national consumer laws which apply. An LLM will understand the specific consumer situation and can recommend a course of action most likely to result in the outcome desired – which are within their rights, but often not realised due to tactical friction introduced by the company on the other side. In the future, an LLM will stay on hold, chat with a corporate agent or person, answer emails citing contractual terms, legal obligations and consumer rights for as long as it has the token budget to keep going. The human is removed from the pain (time) of doing this personally.

New services are now appearing whose value proposition to the consumer is to leverage the power of LLMs in an agent-based application to undertake this enforcement of contractual rights and defeat the corporate dark patterns and strategic friction. Muse by Meta is currently the highest profile. The fully autonomous agent is the vision of the future which may or may not appear as promised, but the benefits of asymmetric consumer warfare against the corporate inertia machine are available to everyone today.

There are many details to work through, and the market is currently trying to price what happens when objectively inferior products and services, earning excess returns for their corporate owners, see their customer migrated by the algorithm to a better provider, a lower cost contract, an immediate refund.
Earning excess returns through dark patterns may go the same way as the alpha from value investing in the algorithmic age.

Although interest rates have been rising for most of 2026, September was when the equity market started to seriously consider the implications for stocks, sectors, and even business models which come under pressure in a higher borrowing cost environment. During the month the Australian 10yr government bond increased from 5.1% to 5.35%, and the RBA raised interest rates another 0.25%. Lessons learned investing through the GFC are embedded in the Longwave process, meaning we pay close attention to how much debt is used to finance the companies we invest in, and which business models increase in risk under higher interest rate and credit cost environments. Names from 2007 to 2009 seared into the memory of Australian investors include Babcock & Brown, Allco, RAMS Home Loans, Opes Prime, MFS Limited, ABC Learning, Centro Property, Great Southern Plantations, Timbercorp, Storm Financial, and City Pacific. Bond yields and credit conditions are not yet as restrictive as in late 2007, early 2008 – but they appear headed that way. Now is not the time for complacency.

Fund performance during the month, and quarter, reflected both sector positioning (underweight interest rate sensitive sectors like Real Estate and Financials) and stock selection into companies with positive cash flows, strong balance sheets, and reasonable valuations.

Our portfolio remains positioned in companies with positive cash flows, strong balance sheets, and business models not reliant on cheap debt – as higher interest rates may be with us for some time. As we observed in 2022, high-quality companies with superior ROEs often also possess pricing power to pass on unexpected inflation and protect shareholder returns. Low-quality businesses have no such ability. As we observed in 2008 – when the tide goes out, unexpected nudity abounds.