OverviewThe AppLive PreviewAI Logic TrustDesign SystemResearch StrategyArchetypesJourneys IAComponentsScreens MetricsRisks
Product concept · Canadian retail banking · Internal design & research package

Anchorline

Money, explained before it moves.

Anchorline is an agentic guidance layer built inside Borealis, a full-service Canadian retail bank. Everything a customer does day to day is here — Interac e-Transfers, bills, registered investing, credit, mortgage — and running through all of it is an AI that tells them where their next dollar should go, explains why, and never moves it without approval.

The bank
Borealis — chequing, savings, TFSA/RRSP/RESP, cards, mortgage
The feature
Anchorline — agentic allocation guidance
Trust rule
The AI suggests and explains. The customer confirms and controls.
How you spot it
The bank is monochrome. Anchorline is the only colour in the app.
Platforms
iOS first · Responsive web · Tablet
01 — The Application

A whole bank, with an opinion

This is the working product, not a picture of one. Move real money between real accounts, send an Interac e-Transfer, pay a bill, contribute to a TFSA, pay down the Visa — every balance on every screen updates as you go. Anchorline sits inside all of it, and the same decision follows you when you switch instruments.

Anchorline is the only colour in the app ⤢ Full responsive preview
Device:

Try it end to end: send an e-Transfer, pay the Visa, contribute to the TFSA — or approve what Anchorline is recommending. Balances update everywhere, and switching instruments keeps your place.

9:41●●●
Always one tap away

Anchorline holds the centre tab on mobile and a pinned sidebar slot on web, plus a floating button that follows you onto every screen in the bank.

Where the decision is

Contextual violet cards appear inside Accounts, Cards, Investments and Borrow — the AI shows up next to the money it's talking about, not in a separate silo.

Top of the dashboard

The home hub is the first thing a customer sees: what's recommended this week, what it's worth, and the approve / adjust / not-now choice.

Proactive, never pushy

Badges and "needs your approval" pills mark the tiles that warrant attention — and disappear the moment the customer decides, including when they decide to hold.

What the prototype actually does

AreaFlowDepthCanadian specifics modelled
AnchorlineRecommend → explain → approve / adjust / decline → receipt → progressFull flowBuffer-first allocation, high-interest debt priority
Move moneyInterac e-Transfer — recipient, amount, account, review, confirmFull flowAutodeposit vs. security question, $0 fee, transfer limits
Move moneyPay a bill — payee, amount, date, review, confirmFull flowCanadian billers, post-dated payments
Move moneyBetween accounts · Deposit a chequeFull flow · Entry pointSame-day settlement, mobile cheque hold periods
InvestContribute to TFSA / RRSP / RESP — account, amount, review, confirmFull flowContribution room tracking, CESG grant on RESP, RRSP deadline
CardsCredit card detail → make a payment (min / statement / custom)Full flowMinimum payment, statement balance, 22.9% APR, due date
BorrowMortgage · HELOC · Credit scoreFull flowFixed-rate renewal date, HELOC room, TransUnion score
MoreRewards · Statements · Security · Alerts · Branch/ATM · InsuranceEntry pointCDIC deposit insurance notice, void cheque / direct deposit

What's invariant vs. what adapts

DimensioniOSResponsive WebTablet
Navigation model5-tab bar, Anchorline raised in the centreGrouped sidebar, Anchorline pinned in violetMaster-detail — nav list + full detail pane
Dashboard densityLow — one decision in view at a timeHigh — two-column, plus a persistent explain railMedium — nav and content together, no scroll-hunting
Approval experienceThumb-reachable card, full anatomy inlineInline card + always-open "why" railInline card + explain block in the detail pane
Multi-step flowsOne step per screen, progress bar in the headerOne step per screen, wider review tableOne step per screen, wider review table
Escalation to a humanIdentical on all three — a "Talk to someone" link inside every recommendation and every at-risk state
The trust ruleIdentical on all three — the AI prepares, the customer approves. Nothing moves on its own.
9:41●●●
03 — AI Logic Model

A bounded decision-support system

Anchorline's AI is not described as magic — it is a responsible, explainable system with clear inputs, a fixed prioritisation order, explicit approval boundaries, and a defined behaviour for when it doesn't know enough. Every rule below is what makes the trust rule enforceable in code, not just in copy.

Recommendation priority order

Protect the buffer

A minimum safety cushion is defended before any other move is proposed.

High-interest debt

Proposed whenever its rate outweighs the near-term value of any savings goal.

Goal-priority saving

Allocated by the user's stated order once buffer and debt are healthy.

Investment (secondary)

Proposed only once the above are stable — always prepared, never executed autonomously.

Guidance vs. control preference

ModeWhat changes
GuidedAI prepares full recommendations by default; fewer decisions surfaced per session.
BalancedAI recommends; trade-off framing shown more prominently for the user to weigh.
Manual-leaningAI surfaces analysis and scenarios; the user builds the allocation, AI checks it.

When the AI is assertive, cautious, or abstains

PostureTrigger
AssertiveHigh-confidence data, clear priority order, no conflicting goals.
CautiousRecent life-stage change, thin data history, or a stressed-state signal.
AbstainsConfidence below threshold, or the action falls outside approval boundaries.

Confidence & explanation strategy

High (≥85%)

Full recommendation with one-line rationale; approve in one step.

Medium (60–84%)

Recommendation shown with an explicit caveat on what's uncertain.

Low (<60%)

No specific number proposed — the system explains what data would raise confidence.

Approval boundary rules

ActionAI may prepareAI may auto-executeApproval required
Transfer between eligible accountsYesNoAlways
Recurring contribution setup / editYesNoAlways
Goal contribution reallocationYesNoAlways
Savings rule changeYesNoAlways
Open a new productNoNoOut of scope, v1
Execute an investment tradeSimulation onlyNoOut of scope, v1
Act on incomplete / low-confidence dataNoNoNever

Auditability

Every recommendation, its confidence score, the data snapshot behind it, and the user's decision (approve / adjust / decline) are logged as one traceable record — the same record structure a compliance reviewer, a support agent, and the user's own approval-history screen all read from.

04 — Trust & Governance

Trust as architecture, not messaging

These patterns are built into the interface itself — not stated as values in a marketing page. Each one maps to a specific, always-present piece of UI.

AI disclosure

A persistent "Recommended by Anchorline AI" mark on every AI-authored element — never blended into system copy.

Explanation pattern

What, why, data used, benefit, trade-off — the same six fields, every time, in the same order.

Confidence communication

A visible chip and, where relevant, a percentage — never confidence implied only by tone.

Fairness & bias awareness

Recommendation logic reviewed across archetypes for disparate impact before each release; flagged in the research hub's stakeholder view.

Privacy signalling

"Data used" is shown at the point of every recommendation — not buried in a separate privacy centre.

Human escalation

"Talk to someone" inside every approval panel and low-confidence state, carrying full context to the agent.

Customer control

Guidance level, data inputs, and any standing rule can be paused or reversed by the user at any time.

Post-action transparency

Every approved action produces a plain-language receipt — what happened, why, and how to undo or adjust it.

Audit trail

Recommendation, confidence, data snapshot, and decision logged as one traceable record per action.

What the product says when it's unsure

"We don't have enough recent data to recommend confidently yet. We're missing a clear view of your regular bills. You can add them now, or hold steady — nothing needs to move today."

Failure handling

  • A failed transfer never retries silently — the user sees exactly what didn't complete and why.
  • If a recommendation was based on data later found to be wrong, the user is proactively told, not left to notice.
  • System outages fall back to the last-known-good state; no approval is ever inferred from stale data.
05 — Design System

An instrument, not an interface

The design language is near-monochrome and hairline-ruled: white surfaces, 1px borders instead of shadows, and hierarchy carried by weight and scale rather than colour. Exactly one saturated hue exists in the product, and it belongs to Anchorline — so the agentic layer is never mistaken for the bank.

Design principles

Trust before automation

Nothing moves until the user says so — the interface never implies otherwise.

Clarity before density

One recommendation at a time. Depth is available, never defaulted to.

Recommendation before execution

The AI's voice is always proposing, never announcing something done.

Explainability before persuasion

Reasoning is shown, not sold. No urgency copy, no artificial scarcity.

Progress before gamification

Momentum is shown as real trend lines, not points, streaks, or badges.

Premium restraint

Hairlines and white space over shadow and glow. The one gradient in the system marks Anchorline.

Brand & product adjectives

ComposedPreciseCandid GroundedFluentUnhurried

Colour strategy

The bank is monochrome — paper, ink, and a hairline. Violet is reserved exclusively for Anchorline, which makes AI-authored content self-evident without a single badge. Status colour is a separate semantic layer, never the accent.

PaperGround — near-white, neutral
InkText and bank primary actions
VioletAnchorline only — never bank chrome
Violet tintAnchorline surfaces & inline cards
Signal — goodOn-track, ahead
Signal — watchAt-risk, needs review

Typography strategy

One family, four weights. Instrument Sans runs everything, with hierarchy coming from weight, size and tracking rather than a second typeface. Figures are tabular everywhere so columns of money align; a mono face is held back for reference codes only.

Display / 600We recommend moving $260
UI / 600Approve this transfer
Body / 400Paying the Visa now saves about $47 in interest this month.
Data / tabular$1,940.00 · 92% confidence

Spacing, layout & iconography

  • Base unit — 4px grid, 8px rhythm for component spacing, 24px for section rhythm.
  • One idea per screen — a recommendation, a check-in, a decision. Depth sits behind "why" and "details," never inline by default.
  • Iconography — single-weight line icons only (1.5–1.6px stroke); no filled glyphs, no coins, rockets, or lightning bolts.
  • Corner language — 8–12px radii on cards and controls, 24–34px on device-level surfaces; nothing fully pill-shaped except real actions and toggles.

Motion & component philosophy

  • Motion communicates state, not delight — a number settling, a status changing, a panel opening. 150–220ms, standard easing.
  • No celebratory animation on money movement; confirmation is calm, not confetti.
  • Components are composable "instruments" — cards, panels, and tiles share one elevation language: hairline border first, shadow only at the topmost layer (modals, approval sheets).
  • prefers-reduced-motion is honoured everywhere; nothing depends on animation to be understood.

Trust-signalling & accessibility rules

  • Every AI-authored element carries a small Recommended mark — never presented as a neutral system statement.
  • An explanation ("why") is always adjacent to a recommendation, never hidden behind extra taps.
  • Confidence is always visible where an approval is requested — never only in a details screen.
  • Status is never colour-only: every pill pairs colour with a word and a dot glyph.
  • Text contrast holds WCAG AA minimum (4.5:1 body, 3:1 large text) in both themes.
  • Tap targets ≥44×44pt on iOS and tablet; ≥40×40px on web with visible focus rings.
  • Approval flows are fully operable and announced via screen reader — amount, action, and consequence read in one pass.
  • No information is conveyed by animation or hover alone; every state has a static equivalent.

Content tone & microcopy rules

SituationInstead ofAnchorline says
Debt recommendation"You're wasting money on interest!""Paying the Visa now saves about $47 in interest this month."
Low confidence"We couldn't calculate this.""We don't have enough recent spending data to recommend confidently yet."
Approval CTA"Optimize now""Approve this transfer"
Nothing to do(silence, or a forced tip)"Holding steady is the right move this week — nothing needs your approval."
Missed goal"You failed to reach your goal.""You're a little behind pace. Here's one adjustment that gets you back on track."

Recommendation card — anatomy

  1. Recommended tagMarks this as AI-proposed, not a system fact.
  2. Headline actionOne sentence, plain language, the amount and destination.
  3. One-line rationaleThe "why," always present, never a tap away.
  4. Confidence chipHigh / Medium / Low, shown at a glance.
  5. Primary actionsApprove · Adjust · Not now — three ways out, never one.

Approval panel — anatomy

  1. WhatThe exact action and amount, stated plainly.
  2. WhyThe reasoning, in the user's terms.
  3. Data usedWhat inputs influenced this — balances, goals, spend pattern.
  4. Expected benefit / trade-offWhat improves, and what the honest cost is.
  5. What happens next & what's editableConfirms this isn't a one-way door.
  6. Confidence & escalationConfidence level, plus a visible path to a human.

Progress & status visualisation language

On track

Real trend line vs. target line — never a badge or streak count.

At risk

Shown with the one action that would restore pace — never alarm-coloured alone.

Ahead of pace

Framed as an option to redirect surplus, not pressure to "optimize more."

Cross-platform behaviour rules

Stays constantAdapts by platform
Colour tokens, type roles, iconography, motion timingNavigation model (tab bar / sidebar / split view)
Recommendation & approval panel anatomy (all fields, every field order)Panel presentation (sheet / modal / inline)
The trust rule — recommend, explain, approveControl density and how much planning depth is shown by default
Microcopy voice and terminologyLayout structure (single column / multi-column / master-detail)
06 — UX Research Hub

A command centre, not a findings deck

The hub is internal-first — built for product, design, research, and compliance stakeholders to trace every product decision back to evidence, and to see, at a glance, what is known, what is assumed, and what still needs validating.

Research objectives

  • Establish what "trustworthy financial guidance" means to a low-confidence retail customer.
  • Determine the guidance-vs-control preference distribution across life stages.
  • Identify the exact moments explanation is required for approval, vs. optional.
  • Validate that trust-led engagement predicts recommendation follow-through better than urgency-based nudging.

Critical unknowns

  • Does explanation depth increase approval speed, or slow it past a usable threshold?
  • How much does trust decay after one incorrect or poorly-timed recommendation?
  • What's the right re-engagement pattern for a user like Marcus who disengages under stress?
  • Where is the line between "helpful nudge" and "pressure," in this specific product's voice?

Problem framing & jobs to be done

"Help me know where my next dollar should go, without needing to become a finance expert to trust the answer."

Main JTBD

When money moves (payday, a bill, a windfall), tell me the smartest next step and let me approve it in seconds.

Related JTBD

When my situation changes, help me re-plan without starting over.

Related JTBD

Show me I'm making progress in a way I can believe, not just a number that moved.

Behavioural tensions & trust barriers

TensionWhat's happeningDesign response
Wants help / fears judgmentUsers avoid tools that feel like a scorecard on their choices.Non-evaluative language; momentum framing over comparison.
Wants automation / fears losing controlAuto-pilot budgeting apps have burned users with silent changes.Recommend-then-approve as an absolute rule, always visible.
Wants simplicity / needs to trust the mathOversimplified advice reads as untrustworthy to confident users (Elaine, Devon)."Why" always available, depth on demand, never forced.
Engages when stable / avoids when stressedDebt-stressed users (Marcus) disengage exactly when guidance matters most.Low-pressure re-entry points; no guilt-based nudges.

Evidence types & confidence weighting

Evidence typeExampleConfidence weight
Behavioural / transactional dataActual transfer & approval logs from pilot cohortHigh
Moderated usability testingTask-based sessions on approval & explanation comprehensionHigh
Diary studiesWeekly check-in reactions over 6–8 weeks per archetypeMedium
Survey / attitudinalTrust and satisfaction self-reportMedium
Expert / heuristic reviewCompliance & accessibility auditMedium
Internal stakeholder assumptionUnvalidated product hypothesisLow — flagged

Hypothesis tracking

IDHypothesisStatusEvidenceLinked decision
H-01Explaining "why" at the point of approval increases approval rate vs. a bare CTA.SupportedModerated testing, n=24Approval panel anatomy (§03)
H-02A visible guidance-level toggle increases trust for high-confidence users (Builder, Stabilizer).Partially supportedDiary study, 2 archetypesGuidance-vs-control model (§07)
H-03Debt-stressed users disengage after a single "you're behind" framing.SupportedDiary study + support transcriptsReset archetype copy rules (§02)
H-04Investment layer visibility before debt payoff reduces trust in the recommendation engine.UnvalidatedNone yet — planned studyv1 scope: save/debt first (§01)
H-05A confidence chip alone (without a number) is sufficient for approval comprehension.UnvalidatedNone yet — planned studyConfidence & explanation strategy (§07)

Research → decision → solution traceability

InsightProduct decisionSolution areaDesign system rule
Silent auto-changes destroy trustAll meaningful actions require explicit approvalApproval & confirmation flowApproval panel anatomy
Shame language causes disengagementNon-evaluative, coach-toned microcopyWeekly progress flowContent tone rules
Confident users distrust oversimplification"Why" always present, depth on demandRecommendation flowRecommendation card anatomy
Guidance preference varies widely by archetypeGuided / Balanced / Manual modesGuided setup + AI logicCross-platform control density

Prioritisation framework

Research and design backlog scored on: Trust Impact × Reach × Evidence Confidence, divided by Effort. Anything touching approval, explanation, or money-movement trust is weighted 2× regardless of score — it is treated as a compliance-adjacent surface, not a growth lever.

Internal stakeholder views

  • Product — plan acceptance & follow-through by archetype, prioritisation backlog.
  • Design — component-level usability findings, accessibility audit trail.
  • Compliance & risk — approval-boundary adherence, audit log completeness, fairness review status.
  • Research — hypothesis status, evidence confidence, open validation queue.

Open questions for future validation

  • What confidence threshold should suppress a recommendation entirely rather than show it as "low confidence"?
  • Does tablet's side-by-side planning view change approval speed or comprehension vs. mobile's sequential flow?
  • How should the product reintroduce itself to a user who disengaged for 60+ days during a financial setback?
  • What's the right disclosure moment for investment-layer eligibility without it reading as a sales trigger?
07 — Product-Owner Strategy

Trust is the growth strategy

Anchorline is framed as a trust-led engagement product, not a feature. Every downstream decision — what the AI is allowed to do, what it must explain, what ships in v1 — traces back to this framing.

Strategic frame

  • Primary user — mass retail banking customer, low-to-medium financial confidence, broad age range.
  • Core job to be done — "Help me make better money decisions without needing expert knowledge."
  • Core user value — clear recommendations, visible progress, low effort, confidence, control.
  • Core bank value — stronger digital engagement, retention, trust, selective product penetration.
  • Experience tone — a smart financial coach. Never a salesperson. Never an overbearing autopilot.

v1 scope discipline

  • In scope — savings behaviour, debt payoff efficiency, cash buffer health, goal-based reallocation.
  • Secondary layer — investment intelligence, introduced only once saving & debt posture is stable.
  • Always required — explicit approval before any meaningful money movement.
  • Always available — a manual, non-AI route for every guided action.
  • Out of v1 — autonomous execution, product opening, trading, action on low-confidence data.

KPI framework

TierMetricWhat it tells the businessWhat it tells the user
North StarTrust-led engagementCustomers return because they believe the guidance, not because they're nudged."I trust what this tells me."
SecondaryPlan acceptance rateRecommendation quality & explanation clarity are working."This actually fits my situation."
SecondaryRepeat usage (weekly check-in return rate)The habit loop is forming without being coercive."Checking in feels worth it."
SecondaryRecommendation follow-throughAdvice converts to real financial behaviour change."I did the thing, and it helped."
SecondarySatisfaction & reduced effortThe product is lowering cognitive load, not adding to it."I didn't have to think hard about this."
SecondarySelective product adoptionTrust converts into deeper banking relationships, earned not pushed."This felt like the right next step, not a pitch."
08 — User Archetypes

Five people, one trust model

Anchorline's guidance logic flexes by life stage and confidence level, but the trust rule never does. These five archetypes anchor every design and AI-logic decision in the rest of this package.

The Starter

Maya · 24 · Junior analyst
Trust sensitivity: High — fears judgment
Context
First full-time income, $6.2k student debt, $310 average balance, irregular saving.
Pain
Doesn't know what "good" looks like; associates budgeting apps with being told off.
Goals
Build a starter buffer, stop overdraft anxiety, understand her own spending.
Guidance
Guided — wants the app to decide the mechanics, not the values.
Behaviour
Checks balance often, avoids anything that feels like a lecture, responds to small wins.
AI should emphasize: tiny, low-risk first moves and visible momentum.
Must avoid: shaming language, debt comparisons, any tone of judgment.

The Stabilizer

Devon · 31 · Product manager
Trust sensitivity: Medium — wants efficiency, not hand-holding
Context
Stable income, growing savings across 3 uncoordinated goals (trip, house, buffer).
Pain
Money is scattered across goals with no logic; unsure if allocation is optimal.
Goals
Coordinate goal priority, house down payment on a clear timeline.
Guidance
Balanced — wants recommendations, but expects to adjust the logic.
Behaviour
Uses payday reallocation heavily; compares scenarios before approving.
AI should emphasize: trade-off clarity between competing goals, timelines.
Must avoid: oversimplifying — Devon will distrust a "just trust us" answer.

The Anchor

Priya & Sam · 38–42 · Household
Trust sensitivity: High — low margin for error
Context
Two kids, mortgage, childcare, RESP contributions; tight monthly slack.
Pain
No time or energy for financial admin; one mistake feels costly.
Goals
Protect the buffer, keep RESP on track, avoid new debt.
Guidance
Guided — values reassurance and simplicity over control.
Behaviour
Engages weekly at most; wants confirmation nothing has gone wrong.
AI should emphasize: stability signals — "buffer is safe," "nothing needs you."
Must avoid: frequent asks, complex trade-off framing, unnecessary urgency.

The Reset

Marcus · 45 · Recently reduced hours
Trust sensitivity: Very high — shame & anxiety present
Context
Income drop, $14k high-interest debt across two cards, depleted buffer.
Pain
Feels behind and judged by every finance product he's used.
Goals
Stop the bleeding, rebuild a minimal buffer, a believable debt path out.
Guidance
Guided, cautious — needs stabilization before optimization.
Behaviour
May avoid the app when anxious; needs re-engagement without guilt.
AI should emphasize: non-judgmental stabilization, one clear next step at a time.
Must avoid: aggressive debt payoff pressure, comparisons, silence when things worsen.

The Builder

Elaine · 52 · Operations director
Trust sensitivity: Medium — wants rigor, not reassurance
Context
Debt-free, healthy buffer, saving basics covered; underusing investment capacity.
Pain
Unsure if she's "doing enough"; existing tools feel either too basic or too salesy.
Goals
Confirm she's on track for retirement horizon; grow beyond savings.
Guidance
Manual-leaning — wants scenario tools and explainability more than automation.
Behaviour
Uses web/tablet for deeper planning; low tolerance for oversimplified advice.
AI should emphasize: rigorous explanation, scenario comparison, the "why" behind confidence levels.
Must avoid: pushing products, treating her like a novice, hiding the reasoning behind a recommendation.
09 — Key Journeys

Three journeys, optimised deliberately

These are the only three journeys the v1 product is designed to be excellent at — everything else in the IA supports them.

1 · Guided setup into a smart allocation plan

Context

Income, balances, debts, and existing goals — pulled where possible, confirmed by the user.

Priorities

User ranks goals and sets time horizon and risk comfort.

Guidance level

Guided, Balanced, or Manual-leaning — changeable anytime.

First plan

A full allocation plan shown with the standard explanation anatomy.

Approve or adjust

Nothing activates until approved; manual edit is always available.

2 · Weekly progress check-in and adjustment

Status summary

On-track / at-risk / ahead, for each goal, debt, and the buffer.

What changed

Spending behaviour shift explained in plain terms, not just a number.

One next-best action

A single recommendation, not a list — with full explanation anatomy.

Approve, adjust, or hold

"Holding steady" is a valid, clearly stated outcome.

3 · Reallocation during a moment of change

Trigger detected

Payday, a new expense, or a goal edited by the user.

Re-run priority order

Buffer → debt → goals → investment, recalculated against new context.

Explain what changed

Why this differs from last week's plan, in one line.

Approve the adjustment

Same anatomy, same standing rule — no shortcuts for "small" changes.

10 — Information Architecture

One system, mapped

How the bank's own IA is organised, and how the three parts of this package feed each other — not as a metaphor, but as the literal flow of information from evidence to language to live product.

Research Hubarchetypes · hypotheses · evidence
justifies →
Design Systemtokens · patterns · voice
informs →
Borealis Appflows · screens · platforms

…and the app's real usage data feeds back into the Research Hub as new evidence — closing the loop.

  • Homeoverview & next action
    • Anchorline hub always first
    • Quick actions
    • Banking · Credit · Investments · Borrowing
  • Move money
    • Interac e-Transfer
    • Pay a bill
    • Between my accounts
    • Deposit a cheque
  • Anchorlinethe agentic layer
    • Active recommendation
    • Allocation plan
    • Weekly progress
    • Decision history
  • Investments
    • TFSA · RRSP · RESP
    • Contribute
    • GICs · Portfolios & funds
  • Cards
    • Card detail & statement
    • Make a payment
    • Cash back · Lock card
  • Mortgage & loans
    • Mortgage detail · lump sum · renewal
    • Home Equity Line
    • Credit score
  • More
    • Rewards · Statements · Alerts
    • Security centre
    • Branch/ATM · Insurance · Contact
    • Settings — incl. guidance level

Where Anchorline attaches to the bank's IA

Bank surfaceAnchorline's presenceWhy here
HomeFull recommendation hub, pinned above everythingThe decision is the point of the session, so it leads.
Global navigationCentre tab (mobile) · pinned sidebar slot (web) · floating button everywhereNever more than one tap from any screen in the bank.
Accounts & savingsInline card on buffer statusThe buffer is the first thing the priority order defends.
Credit cardsInline card quantifying interest savedHighest-APR debt is the highest-return move available.
InvestmentsInline card explaining why investing is gatedExplaining a withheld suggestion builds more trust than making one.
Mortgage & loansInline card explaining why it's untouchedLow-rate debt should not be prepaid ahead of a 22.9% balance.
Credit scoreInline card on utilisation impactConnects an abstract score to the concrete action on the table.
Pay-your-Visa flowPreset amount + rationale inside the flowGuidance where the number is actually chosen, not after.
11 — Component System

The families that build every screen

FamilyPurposeKey statesGoverned by
Recommendation cardPresent one AI-proposed actionDefault, low-confidence, approved, declinedRecommendation anatomy (§03)
Approval panelFull explanation before money movesReviewing, confirming, confirmed, escalatedApproval anatomy (§03)
Status pillEncode on-track / at-risk / ahead / confidenceGood, watch, risk, accentProgress language (§03)
Stat tileShow one figure with its label, tabular-numeral setDefault, trending up/downTypography strategy (§03)
Explanation panelPersistent "why" — data, benefit, trade-off, confidenceInline (mobile/tablet), persistent (web)Approval anatomy (§03)
Waterfall allocatorShow buffer → debt → goals → investment splitRecommended, user-adjustedAI priority order (§07)
Progress trendReal line vs. target, not a badgeOn-track, at-risk, aheadProgress language (§03)
Guidance-level selectorGuided / Balanced / Manual-leaningSelected, changed-this-sessionGuidance model (§07)
Escalation entry point"Talk to someone," context-carryingDefault, in-progress handoffTrust & governance (§08)
12 — Screen Inventory

What ships in v1

Every screen below is built and interactive in the prototype unless marked as an entry point. Entry points render a real, labelled destination but stop short of a completing flow.

ScreeniOSWebTabletStates covered
Home / overviewPending · approved · held
Anchorline hubRecommend · adjust · approved · held · escalated
Allocation planLocked / unlocked investing layer
Weekly progressOn-track · at-risk · ahead
Accounts list & detailBank · registered · credit · loan
Interac e-Transfer4 steps · autodeposit vs. security question
Pay a bill4 steps · today vs. due date
Transfer between accounts3 steps
Contribute (TFSA/RRSP/RESP)4 steps · room tracking · CESG
Credit card & payment3 steps · min / recommended / full
Mortgage & loansDefault
Credit scoreDefault · utilisation change
ActivityLive — reflects every completed flow
Deposit a chequeEntry pointEntry pointEntry point
Rewards · Statements · Alerts · SecurityEntry pointEntry pointEntry point
Branch/ATM · Insurance · Contact · SettingsEntry pointEntry pointEntry point

States that must exist, not just the happy path

StateWhat the customer seesWhat Anchorline does
Empty"Let's set up your plan — it takes about 4 minutes."Nothing yet; offers a guided or manual setup entry.
LoadingA calm, static skeleton — no spinner tricks or fake progress.Computing against current data; no partial numbers shown.
ErrorPlain explanation of what failed and what still works.Falls back to last-known-good; never silently retries an approval.
Low confidence"We don't have enough recent data to recommend confidently. Here's what's missing."Abstains from a number; offers the smallest safe action or a data prompt.
Decision held"Nothing needs your approval" — with the cost of holding stated plainly.Stops asking until the next payday trigger. No nagging.
13 — Metrics

How success is actually measured

MetricDefinitionTarget signalSource
Trust-led engagementReturn usage correlated with approval rate, not session count aloneRising together, not divergingProduct analytics + approval log
Plan acceptance rate% of recommendations approved as-is or with minor adjustment>60% by month 3Approval log
Repeat weekly check-in rate% of users returning for the weekly progress flow>50% week-over-weekProduct analytics
Recommendation follow-through% of approved actions still in effect 30 days later>80%Transaction data
Satisfaction / effortPost-approval micro-survey, effort & trust framingCSAT ≥4.3/5, low-effort ≥70%In-product survey
Selective product adoptionAdoption of a recommended product only after guidance history existsDirectional, not pushedCRM + approval log
Escalation rate & resolution% of sessions escalating to a human, and time to resolveLow volume, high resolution qualitySupport system
14 — Risks, Gaps & Validation Priorities

What this concept doesn't yet know

Risk / gapWhy it mattersMitigationPriority
Confidence threshold is not yet empirically calibratedWrong threshold either over-abstains (erodes usefulness) or under-abstains (erodes trust)Controlled pilot with graduated thresholds, H-05High
Re-engagement pattern for disengaged, stressed users unvalidatedThe Reset archetype is highest-need and highest-churn-riskDiary study + support transcript analysisHigh
Investment-layer disclosure timing untestedRisk of reading as a sales trigger, damaging trust built on save/debt guidanceH-04 validation study before layer shipsMedium
Fairness review across archetypes not yet run on live logicRecommendation logic could disadvantage a specific income or life-stage patternFormal fairness audit before GA, owned by complianceHigh
Tablet split-view approval speed unmeasuredSide-by-side comparison could speed or slow comprehension vs. sequential mobile flowComparative usability test, both layoutsMedium
Escalation handoff context-completeness unverified end-to-endIf context doesn't reach the human agent, the trust promise breaks at the exact moment it matters mostEnd-to-end handoff test with support opsHigh