Amlok
The idea in brief.
A household can meet a monthly average and still face a cash shortfall on a particular day. Understanding affordability requires a view of recurring income, essential costs, existing obligations, accessible reserves, and the timing that connects them. [1]
- Income level, income variability, and payment timing are separate dimensions.
- Keep verified cash, inferred categories, and unobserved obligations distinct.
- Use uncertainty to trigger an evidence review rather than to manufacture a precise affordability score.
The external survey is a dated US population observation. Household examples are hypothetical CAD calculations. This paper proposes a review method; it does not establish a consumer-credit model, approval rule, or financial advice for an individual.
Measure the household on a consistent basis
Gross salary, take-home pay, benefits, transfers, and borrowing proceeds should not be aggregated indiscriminately as income. Identify each source and explain how recurring income is recognized. A transfer from savings can support a payment but does not increase recurring earning capacity. A new loan creates cash and a new obligation.
Expenses need similar care. Recurring essentials, discretionary spending, annual bills, and existing debt payments have different timing. A low-spending month may omit a forthcoming insurance renewal or seasonal utility cost. State which costs are observed, which are estimated, and which remain outside the connected accounts. [1][2]
Affordability is also purpose-specific. A monthly residual describes one set of cash assumptions; a lender’s policy ratio may use another. Show the calculation and its intended interpretation. A clear calculation is more useful than a universal score whose denominator changes from one applicant to another.
Financial resilience is more than current income
In the Federal Reserve’s 2024 US household survey, 63% of adults said they would cover a hypothetical USD 400 emergency entirely with cash or its equivalent. The definition includes a credit card paid off at the next statement. This is a self-reported response to a hypothetical expense, not an observed default rate or a finding about Canadian households. [3]
The result motivates a separate question about accessible buffers. Two households with similar income can have different capacity to absorb an unexpected cost. Conversely, a high balance can be earmarked for a near-term obligation. The analytical view should preserve the difference between an account balance, available liquidity, and long-term wealth.
When using a buffer measure, specify which accounts are included and whether the funds are accessible. Avoid inferring a lack of savings from incomplete connections. A household should be able to provide context or correct the underlying evidence before it influences a consequential decision.
One survey measure of financial resilience
View figure data
| Measure | Value |
|---|---|
| Cash or equivalent response | 63% |
Compare level and timing
Consider two hypothetical households, each receiving CAD 4,800 per month and paying CAD 4,200 in recurring costs. Both have a monthly residual of CAD 600. Household A receives two equal payroll deposits spread across the month. Household B receives one irregular deposit near month end. If most bills fall early, their short-term liquidity paths can differ despite identical monthly totals.
A cash-calendar model begins with opening accessible cash, adds each expected inflow on its date, and subtracts each scheduled payment. The minimum balance over the window identifies a timing pressure that a monthly total can hide. Expectations should be labelled as such; a projected payroll date is not an observed deposit.
Now consider a simplified monthly stress for either household: income falls by 15% from CAD 4,800 to CAD 4,080 while recurring costs stay CAD 4,200. The residual changes from positive CAD 600 to negative CAD 120. If the household has CAD 1,200 of accessible reserves, dividing by that shortfall gives ten months in a deliberately simplified constant-gap calculation. It ignores other shocks and should never be presented as a survival forecast.
| Measure | Baseline | 15% lower income |
|---|---|---|
| Income | 4,800 | 4,080 |
| Recurring costs | 4,200 | 4,200 |
| Residual | 600 | −120 |
| Assumed accessible reserve | 1,200 | 1,200 |
Design a reviewable change signal
Changes in observed income can result from unemployment, reduced hours, a changed pay schedule, account switching, or missing data. Classification should preserve these alternatives until the evidence resolves them. A useful signal reports the observation and the uncertainty rather than collapsing both into a distress label.
Compare equivalent windows and display the records contributing to the change. If one payroll period has not yet completed, say so. If a source has stopped refreshing, show the last successful observation and the affected calculations. A system that treats missing data as zero can generate the most alarming signal exactly when its evidence is weakest. [4]
An appropriate review workflow can request clarification, refresh a source, or route a case to a trained team under the institution’s policies. The signal should not itself imply a collection action or an automatic reduction in access. Institutional accountability and a way to correct evidence are part of the design.
From an observed change to an informed response
- 01Observe
Identify the changed amount, date, and source.
- 02Check coverage
Inspect freshness, missing accounts, and timing.
- 03Review context
Distinguish verified facts from alternative explanations.
- 04Record the response
Assign an owner and preserve corrections.
Validate the population and the consequences
A consumer-credit study needs an explicit outcome and horizon, representative authorized data, and a point-in-time feature view. It should inspect performance under different income patterns, evidence coverage, and product structures. A result from business lending is not a consumer validation.
Assess calibration as well as ranking, and inspect the workload created at the proposed review threshold. Examine errors and access implications across relevant groups using lawful, appropriate evaluation data. If the intended use changes, the evaluation should change with it. A tool designed to summarize evidence should not quietly become an unvalidated approval policy.
Amlok’s published model card distinguishes completed structured-loan and retrieval baselines from the research programme’s broader ambitions. No consumer outcome claim follows from those baselines alone. The useful near-term standard is a process where every calculation, uncertainty, and response can be reviewed. [5][6]
Sources & methodology.
Figures identify their population, period, and units. Scenarios marked illustrative are Amlok calculations using stated assumptions, not observed customer outcomes.
- 1Consumer underwriting researchAmlok research and product documentation
Affordability, temporal evidence, and consumer-specific evaluation requirements.
- 2Financial analysis and evidenceAmlok research and product documentation
Definitions, reconciliation, and analytical methods; implementation guidance rather than measured portfolio performance.
- 3Economic Well-Being of US Households in 2024: Savings and InvestmentsFederal Reserve Board · May 2025
US adults’ self-reported response to a hypothetical USD 400 emergency. Cash equivalent includes a credit card paid off at its next statement.
- 4Monitoring and change detectionAmlok research and product documentation
Proposed monitoring design, observation windows, baselines, and evaluation requirements.
- 5Research model cardAmlok research and product documentation
Intended uses, exclusions, status of model development, and limits to generalization.
- 6Evaluation designAmlok research and product documentation
Evaluation protocols and research objectives. A specified protocol is not evidence that an experiment has been completed.
