Intersectionally-Aware Guidance & Support

The dominant paradigm in AI tooling is de-contextualisation as a feature — one model that works for everyone. Most financial products assume a similar baseline: stable income, predictable costs, and equal access to time, mobility, housing, credit, and support networks. Many users do not experience that baseline.

The more interesting frontier is the opposite: tooling that takes context seriously and adapts to it. My MA was in social anthropology precisely because I find this question — how people make decisions inside specific cultural and institutional contexts — genuinely fascinating. Every user-research project I run hits the same pattern: the aggregate finding is interesting; the cluster-level finding is more interesting; the individual story that does not fit the cluster is most interesting of all.

This concept is a bet that a guidance system can respect the individual fit the way a good ethnographer does — not through a black-box model, but through a transparent, opt-in layer that invites users to share context that changes what "good" guidance means for them.

Tools typically treat users as isolated individuals with comparable constraints. In reality, people navigate structural barriers that shape income, costs, risk, and access to help — often simultaneously:

• Care responsibilities constrain time, earning capacity, and spending flexibility in ways a generic budget template cannot model.
• Disability and chronic illness create high, non-negotiable costs that make standard "cut spending" advice harmful rather than helpful.
• Discrimination in credit, employment, and housing limits the options that standard guidance assumes are available.
• Migration or visa complexity can exclude people from benefits, banking products, and legal protections.
• Gender-based economic risk — financial abuse, pension gaps, safety costs — is invisible to most fintech tools.

These factors rarely arrive alone. A single parent managing both insecure work and disability-related costs faces a qualitatively different financial reality from someone experiencing only one of those pressures.

The cost of ignoring this

For users: guidance feels irrelevant, judgmental, or inaccessible — eroding trust and engagement at the exact moment it's most needed.

For support teams: coordinators absorb the cost of repairing misdirected or mis-framed advice, driving up contact volume on complex cases.

For the product: retention drops in the cohorts with the highest need and the deepest loyalty potential — the people most likely to stay if the product works for them.

For the brand: claims of inclusivity ring hollow without structural follow-through. Opt-in signals intention; repeated failure signals otherwise.

The stances below are not UX polish — they're constraints on what the product is allowed to do. Each one closes off a class of failure mode.

Serving, not profiling

Disclosed context is used only for stated purposes — guidance, referrals, support routing — with clear user controls. The system is not allowed to infer identity categories from disclosed situations, build hidden scores, or share context beyond the user's stated purpose. Purpose limitation is written into the data model, not enforced by goodwill.

Situations before labels

The primary UX prefers functional descriptors ("irregular income," "caring for someone with high support needs," "high unavoidable health costs") over identity categories — while still allowing user-led identification where people actively want recognition for relevance. The distinction matters: labels can essentialise; situations describe constraints without fixing identity. Users who want to name their experience can; users who don't won't be asked to.

Rules-based first

A transparent, human-reviewable mapping from opted-in context to curated playbooks and resources. No black-box model required to ship early value. The logic is auditable by coordinators and legal. If the system can't explain why a user saw a particular recommendation, it shouldn't have made it.

The five implementation principles

1. User agency — opt-in, granular, revocable. Plain-language explanation of why we ask and what changes.
2. Minimum necessary — prefer situation descriptors over identity labels where the guidance outcome is the same.
3. No essentialising — content variants address constraints, not stereotypes. No demographic shortcuts.
4. Human safety valve — easy, dignified path to a person when automation is insufficient or stakes are high.
5. Equity by evaluation — test with diverse participants. Monitor for disparate outcomes if shipped. Publish what we learn.

An opt-in layer — not a hidden score — that invites users to share aspects of context that change what "good" financial guidance means for them, in exchange for tailored next steps:

• Budgeting patterns for volatile or constrained income — variable-income cashflow framing, not generic spend-less advice.
• Benefits and entitlements eligibility checks and application readiness — surfacing money users didn't know they could access.
• Local services, NGOs, and trusted signposting — geo-aware where data exists, partner-validated before inclusion.
• Financial education matched to real constraints, not assumed demographics — sequenced by what's actually actionable given the user's situation.
• Access-friendly tooling suggestions — approaches suited to how the user describes their own needs, when they say that is relevant to them. Not inferred. Stated.
• Escalation to human support when the situation exceeds self-serve capacity — with coordinator context already populated so the user doesn't have to repeat themselves.

This is the recommended first phase before committing to MVP build. The feature touches sensitive context, structural inequality, and operational safety. Jumping to design without evidence from the people closest to the problem risks building the wrong abstraction — or causing harm. Sequential ordering (coordinators first) protects against designing user journeys that break ops or create ethical or legal exposure.

Research ethics

Informed consent and the right to skip any question in every session. De-identified synthesis; no small-n attribution that could identify participants. Clear scope (not therapy, not legal advice); debrief and signposting resources available throughout. Recruitment screened on situations not identity probing — while still aiming for diversity across relevant dimensions. Appropriate incentives, accessibility provisions (captioning, async options), and trauma-informed moderation.

Intersectionality (Crenshaw, 1989; Hill Collins; King 1988 “multiple jeopardy”) is the framework from Black feminist scholarship that explains how interlocking systems of power and marginalisation — racism, sexism, ableism, classism, cis/heteronormativity, and others — compound on individual experience to produce burdens that single-category thinking cannot see.

Two framings sit alongside one another:

• Multiplicative, not additive. The effects of intersecting systems of power compound on individual experience in a way that is not reducible to the sum of their parts. Hill Collins describes this as the “matrix of domination”; the quantitative-intersectionality literature operationalises it (with appropriate methodological caveats) as multiplicative interaction.
• Qualitatively distinct. The compounding produces a different kind of experience, not just more of the same. Crenshaw's foundational point: legal systems failed Black women because they treated race and gender as separate categories; the experience couldn't be captured by analysing racism and sexism in isolation.

Both descriptions point at the same phenomenon. The critical guardrail is that the framework is fundamentally about systems of power, not individual identity attributes. Reducing intersectionality to “design for diverse users” — mapping characteristics to features without theorising the systems that produce marginalisation — is widely critiqued in the literature, and rightly. The systemic frame is what the design holds onto.

For fintech, this means:

• A budgeting tool works differently for someone juggling care duties + insecure housing + racial discrimination than for someone facing only one of those — qualitatively differently, not just three times worse.
• The same financial advice may empower one user and fail another — not because the advice is wrong in the abstract, but because the situation differs in ways the advice doesn't model.
• Templates break when care, health, labour-market access, credit history, and safety interact.

This is a design principle: we intervene where guidance and referrals meet real ecosystems, not where assumptions meet averages. The framework is borrowed from legal and social theory not as academic cover but as a precise description of what the product is actually trying to solve — and as a tool for surfacing the risks of harm that flat user models conceal.

A freelancer with irregular income. Describes ADHD as relevant to how they manage money. Primary carer for a parent with high support needs.

Emotional outcome we are designing for: seen, not sorted. Useful next steps, not a label dump. Time saved. Money found that they did not know they could access.

This journey is not a user story for engineering. It's a benchmark: if the opted-in context we collect and the outputs we surface don't produce this emotional outcome in the research sessions, the abstraction is wrong and we go back to Phase 1.