The architecture of money decisions: a premise for personal finance

The financial-services industry has been answering a specific question for more than four decades: how do we give people better information about their money? Better visibility, clearer dashboards, categorised transactions, financial-literacy modules, educational content, retirement calculators, budget planners.

I’ve spent my career in banking and fintech. The industry I’ve been part of has been answering one version of that question for the whole of that time. I know what the answers have looked like. And I have come to think that the question itself is wrong.

The problem is not information. The research base built up over those same four decades points to a different conclusion: that information is often necessary but rarely sufficient. For many routine money behaviours, the stronger lever is the structure and timing of the decision itself. In other words it is the architecture of money decisions, not the information surrounding them, that determines how people act.

By ‘architecture of money decisions’ we mean the underlying conditions that shape how money decisions actually get made: the context the person is operating in, what happens by default, how and when attention is called to decisions, how easy it is to act, and who is doing the watching for what matters.

This essay sets out the standard solutions, what the research says, where it is contested and what we are building at Lucie Money in response.

The standard solutions

Financial literacy

We know from many studies that many (if not most) people say that they don’t feel on top of their money and don’t know what to do about it. And the most obvious response to that has been to help them understand and teach them to plan: financial literacy. Regulators, banks, governments, international bodies, NGOs: the effort has converged on a single theory of change. Give people better information, better tools and greater capability, and they will make better decisions about their money.

The scale of that effort is considerable. In Australia, ASIC’s Moneysmart records more than eleven million visits a year and each of the major Australian banks runs a literacy programme. CBA runs Start Smart in schools, ANZ has MoneyMinded (now reaching more than 850,000 adults) and the ANZ and Brotherhood of St Laurence Saver Plus matched-savings programme has over 45,000 participants. Internationally the pattern is the same. The OECD’s International Network on Financial Education brings together close to 300 public institutions across more than 130 countries, including every G20 member; in the United Kingdom alone, the Money and Pensions Service maps 110 financial-education programmes with combined annual spend of £13.5 million to reach 7.2 million children. The labels vary (literacy, education, capability, wellbeing) but the mode of response is the same.

Money tools

Alongside financial literacy, the ecosystem has produced a generation of personal financial management (PFM) and budgeting tools: Mint, YNAB, PocketGuard, Raiz, Frollo, plus the budgeting modules built into most neobank and major bank apps. These tools differ meaningfully in method and audience (YNAB’s active-allocation method is not Mint’s passive tracking; Raiz is primarily a savings product); the point here is not that they are identical, but that most still rely on the user to interpret, remember and act.

The premise is common across them. Let people see their spending; let them set budgets; let them track what happens. Awareness, visibility and self-monitoring: what the research literature calls ‘passive tracking’.

The scale has been considerable. Mint, the largest of these tools, claimed to have more than twenty million registered users at its peak in 2016¹. Each of the major Australian banks has shipped some version of this functionality inside its own app.

Architectural interventions

The ecosystem has also provided another kind of response: changing the architecture of how money moves rather than teaching people how to think about it. Auto-enrolment into superannuation is a population-scale default that lifts participation without teaching anyone anything new. Direct-debit and automatic bill-pay infrastructure is near-universal. Savings “pockets” and “vaults” in some bank apps draw on mental-accounting research. Round-up savings products apply the nudge principle directly.

The impact of standard approaches

But despite all the money and effort invested in these solutions and the genuine difference they can make in particular situations, the overall measurable behavioural impact of these standard approaches has been consistently modest. The evidence base surveyed below is mixed in quality, ranging from peer-reviewed meta-analyses to industry and commercial indicators, so the picture it supports is directional rather than conclusive.

Financial literacy

• A 2014 meta-analysis by Fernandes, Lynch and Netemeyer² examined more than two hundred studies and found that financial-literacy interventions explained about 0.1% of the variance in subsequent financial behaviour.
• A later meta-analysis by Kaiser, Lusardi, Menkhoff and Urban (2022)³, covering seventy-six randomised trials and more than 160,000 participants, argued the true positive effects of financial literacy programmes were at least three times larger than Fernandes had reported. Even so, three times a very small effect is still a small effect: the absolute impact remains small.

In general, financial education helps a bit; effects are modest but real; the benefits tend to diminish with time.

Money tools

• The largest single study of PFM-app impact, a 9,035-person randomised trial run by a behavioural-design firm, Irrational Labs⁴, found that creating a budget did not reduce spending. Setting a budget reliably increased engagement with the app; it did not change financial behaviour.
• Industry retention data shows roughly four in five users of finance apps stop using them within three months⁵.

The commercial evidence is also clear: in March 2024, Intuit closed Mint, the largest PFM tool in the world.

In general, regular users of standard PFM tools describe finding them useful for keeping a mental tab on spending and for spotting recurring charges, but the objective-behaviour gain is not consistently demonstrated.

Architectural interventions

The architectural interventions that do exist tend to be effective in their narrow domains (auto-enrolment super has moved participation at population scale; round-ups genuinely save money; overdraft alerts reduce charges). The problem is not that they don’t work; it is that they remain single-product features inside otherwise traditional architectures, and taken together they do not cover the day-to-day experience of managing money.

Across all three approaches

The evidence is clear: standard solutions are not worthless, but they are not, on their own, going to solve the problem people have with their money.

Which leaves key questions. Why is this impact so modest? What different approach might have a greater benefit? To answer these we have to look at what the behavioural research actually says about how people make money decisions in the first place.

Why people do not act as they intend

Behavioural economics as a field is more than forty years old. Kahneman and Tversky’s prospect theory paper was published in 1979⁶. Thaler’s work on the endowment effect followed in 1980⁷. Since then, the research base has grown but has consistently surfaced three findings relevant to personal finance.

The intention-behaviour gap

Decades of work synthesised in a 2016 meta-analytic review by Sheeran and Webb⁸ show that intentions are a weak predictor of behaviour. Across many studies, a person’s stated intention accounts for only about a fifth to a third of what they actually do; the rest is explained by other factors. And there is a further finding worth noting: even when researchers succeed in deliberately changing a person’s intentions (through motivational or commitment-planning interventions), the resulting change in actual behaviour is small. Most of this research is in health behaviour, so any extension to finance needs care; but the asymmetry is robust. The conclusion is commonsense but with profound implications: intending to act is not the same as acting.

Two systems of thinking

A second finding relates to ‘modes of thinking’. In a significant work in 2011 titled Thinking, Fast and Slow⁹, Kahneman set out a dual-process¹⁰ model of thinking:

• System 1 thinking is fast, automatic and largely effortless. It runs continuously in the background, processing the world through pattern recognition and learned response. It is where most of what we do during the day comes from: reading a familiar face, reacting to a sound, making a snap judgement, grabbing the coffee without thinking about it. It cannot be switched off, and, importantly, is strongly emotional in character.
• System 2 thinking is slow, deliberate and effortful. It engages only when a task requires conscious attention: a calculation, a novel situation, a careful decision. It is where analytical reasoning happens, where self-control is exercised, where deliberate plans are formed. It is costly to run and cannot be held in operation for long; the brain defaults back to System 1 as soon as the effort is released.

Money decisions are made using both modes of thinking.

Some money activity is deliberate System 2 work: sitting down to do a budget, comparing mortgage offers, doing a year-end review. More information helps there.

But most day-to-day money behaviour is not deliberate. The subscription that quietly renews, the spending that drifts, the budget that is never updated, the impulse purchase, the bill that slides past its due date: these are happening in System 1, in the flow of the day, with less considered analysis and often driven by emotional factors. The aggregate outcome of a person’s financial life is shaped far more by this continuous System 1 stream of fluid and emotional decision-making than by any conscious intentions established in System 2 sessions.

Note: The dual-process frame is a useful lens here; it is not a complete account of how money decisions get made. Liquidity constraints, habits, institutional frictions, mental accounting and social norms all play a part too.

The bandwidth tax of scarcity

The third finding concerns the concept of ‘cognitive scarcity’. Mullainathan and Shafir’s 2013 book Scarcity¹¹ drew on years of experimental and field work by the authors and their collaborators. Their thesis is that being short of something (money, time, attention) imposes a particular kind of cognitive tax. The mind focuses disproportionately on what is scarce (what the authors called ‘tunneling’), and the capacity to attend to anything else (working memory, executive control, self-control) is reduced. Scarcity is not just an external constraint on what we can do; it changes how we think.

The implication for personal finance is direct: being short of money is precisely what makes managing money well cognitively hardest. Scarcity of money taxes the very mental bandwidth that careful financial management requires. This is not a moral or educational failure; it is a structural consequence of the scarcity itself. And it points to something about a useful response: one that reduces the mental load of money management rather than adding to it.

Three research findings, one picture

The research summarised here highlights a consistent and reinforcing pattern:

• Intentions do not reliably become behaviour: the distance between what a person plans to do with their money and what they actually do is large, and resistant even to deliberate efforts to close it.
• The money decisions that shape a person’s financial life are made mostly by fast, automatic thinking, in the flow of the day. Information that depends on deliberate analysis tends to reach the wrong part of the mind at the wrong moment.
• The people for whom careful money management would matter most have the least cognitive room to do it. Scarcity itself reduces the bandwidth that careful management requires.

None of this removes the importance of material constraint. Where the binding problem is insufficient income or extreme volatility, better decision architecture can help at the margin but cannot solve the underlying economics. This is a boundary condition on everything that follows, not a footnote.

Taken together, the three findings describe a problem poorly served by the financial industry’s standard responses. People’s money lives are not shaped by a lack of knowledge, a lack of intention, or an unwillingness to act. They are shaped by a continuous stream of fast, emotional, habit-driven decisions taken in whatever attentional conditions the person happens to be operating under. An approach built on giving people better information to reason with is pointing at the wrong part of the problem.

What the research says does change behaviour

For any intervention to affect behaviour driven by System 1 thinking, it has to arrive at the moments the behaviour is actually happening, in a form that doesn’t require the user to stop and switch into System 2 mode.

The research line that reliably changes financial behaviour is not about teaching people anything. It is about changing the architecture of the decision itself: the context in which a decision happens, what the defaults do, when and how attention is drawn, and how easy it is to act. The research literature has focused most on two of these levers (defaults and timely notifications); the evidence on them is worth a closer look.

Removing the need for action

The canonical case is Brigitte Madrian and Dennis Shea’s 2001 paper on automatic enrolment in 401(k) retirement plans in the United States¹²; opt-out defaults produced dramatic increases in participation without teaching anyone anything. The principle is now thoroughly absorbed into the financial industry: automatic bill-pay, direct-debit infrastructure, auto-enrolment super, round-up savings products and countless sign-up toggles are all applications of the same idea.

Notifying at the moment it matters in ways that matter

The literature studies these interventions under various names (reminders, alerts, nudges). “Notification” is the general product term; we use it here and note the specific research terms where the citations do.

Direct research on notifications in financial contexts is more recent. Karlan, Mullainathan, Zinman and colleagues’ 2016 paper in Management Science found that savings reminders increased goal attainment by 3% and total savings by 6%¹³. Grubb, Kelly, Nieboer, Osborne and Sakaguchi’s 2025 paper in the Journal of Finance found that automatic enrolment (a default practice) into just-in-time SMS overdraft alerts at major UK banks reduced overdraft charges by 17 to 19%¹⁴.

The more interesting finding in this body of research is not that notifications change behaviour but what makes them work when they do. Three conditions recur:

• Relevance to the user’s specific situation and goals. Karlan’s 2016 result that goal-specific reminders were roughly twice as effective as generic ones is the clearest example. A notification that lands as “this is about the thing I am actually trying to do” works; the content must be right.
• Timing that coincides with the behaviour. The strongest effects come from notifications that arrive at or near the moment a decision is being made. Grubb’s overdraft alerts were effective because they arrived when the overdraft was imminent, not at the end of the statement month. The just-in-time adaptive intervention literature¹⁵ frames this as a design principle: the content must not just be right, but also arrive at the right time.
• An immediate action path. Notifications that bring the resolving action close at hand (a transfer in the same app, a one-tap confirmation) produce larger behaviour change than notifications that require the user to stop, open another app, figure out what to do and execute it.

Conclusion

What the research describes, cumulatively, is the need for a different kind of response. Not more information, not better education, but interventions that change the architecture of money decisions themselves: absorbing the ongoing cognitive load of watching, building context from the user’s situation, removing the need for a decision where possible, and (where a decision is genuinely needed) surfacing it at the moment it matters with the action close to hand.

The industry has provided some of this, but narrowly. The question is what a more thorough implementation might look like.

The literature does not point uniquely to one product form. What it does suggest is a design direction: continuous attention, context-sensitive prompts and reduced action-costs, brought together so they work as one thing rather than as isolated features.

What we are building, and why it is different

Lucie’s position is not ‘another better money management app’. It is a specific combination of four elements, designed end-to-end as one thing.

Context-declared conversational relationship. Context is built through the user’s ongoing conversation with Lucie, not only through the data Lucie has access to. Goals, situation, priorities and the shape of what matters: much of this is established directly in conversation and refined over time. It is a relationship in which the user is not just observed, but heard.

Architecture of attention. Lucie is being designed to do the watching so that the user doesn’t have to. The ongoing attention required to keep track of your money (paid bills, renewing subscriptions, drift in a budget, whether you are still on track this month) is a cognitive tax in its own right, of the same kind the scarcity research describes. Lucie is intended to absorb it. When Lucie spots something that warrants attention, it reaches out through whichever channel fits the moment, only when it matters, in the form that requires least work to act on. You don’t have to carry any of this in your head. Taking the load off is the point.

Cross-institutional view. Lucie’s attention to a user’s money has to span accounts and institutions rather than stop at one bank’s walls. A view that covered only part of someone’s financial life would miss most of what mattered. Australia’s Consumer Data Right makes this broader view possible, and Lucie is being designed around it from the start. What matters here is less the view itself than how it is used: as one of the foundations the product is built on, not something added later.

Path to permissioned agency. Notifications are not the endpoint. Over time, users who want to will be able to extend the relationship in ways that reduce what they have to attend to themselves. That is part of how Lucie is being shaped, not a later add-on.

None of these elements is unique in isolation. What is distinctive, in our view, is building all four as one system, from first principles, around the shared premise that the binding constraint is attention.

Put the four together and what you have is a personal money agent: one that knows what matters to the user, watches over their financial life, reaches out when something warrants attention, helps them fix things and is designed to do more over time on the user’s terms. That is what Lucie is being built to be.

Closing

A few things to address in conclusion.

The research base we have drawn on is not neatly applicable to personal finance without qualification. Much of the intention-behaviour literature sits outside finance; some scarcity findings are contested; financial literacy education has real benefit; and the impact of stand-alone architectural interventions is often not substantial.

There is also a risk in execution. Done badly, the approach outlined could become just another demand on the attentional bandwidth for which it is meant to be a substitute. The risk of notification fatigue is real. The research answer is that users do not reject notifications as such; they reject notifications they did not ask for, that are contextually-deficient, that arrive at the wrong time, or that they cannot control. Getting the discipline right (relevance, context, timing, user control) is critical.

It should also be remarked that Lucie is not being designed to nudge people towards saving more, spending less or optimising to a standard pattern. It is being designed to serve each person’s ambitions, whatever those are. Lucie is also not the answer for people whose binding constraint is income rather than attention, and at MVP it is not in the purchase flow so cannot intervene in impulse decisions in the moment.

What we think is defensible is the direction proposed. Intentions do not reliably become behaviour. Information often does not reach the part of the mind that makes the decision. For many routine money behaviours, the architecture of the decision appears to be a more powerful lever than information alone.

But at the time of writing, Lucie is pre-launch; we have not yet proven that a solution designed this way will deliver on that premise, and some of what we are currently confident of will turn out to be wrong. What we can say now is that the premise is worth building, the research supports the direction, and the problems Lucie is being designed to address are real and worth solving.

We’re building the MVP now. If it sounds like something you’d want, there’s a waitlist at lucie.money.