A finance workspace that matches insurer payments hitting the bank to the claims they pay for — and explains every cent that doesn’t line up.
This is a self-initiated concept. The product, organisations and data are fictional; the design problems are real.
Role
End-to-end product design
Type
Self-initiated concept
Platform
Desktop web
Focus
Trust in automation, auditability
Context
The problem and the bet
The problem
Insurers pay in bulk: one deposit can cover 40 claims, arrive before the file that explains it and differ by a few dollars. Finance teams reconcile it in spreadsheets, and month-end close slips while they chase cents.
The bet
If the system shows its reasoning — confidence, evidence and history — accountants will let it handle the routine work and focus on the exceptions.
How I’d validate it
Sit with a cash-posting team during month-end close, then test whether they accept or redo proposed matches in a prototype with anonymised data.
JK
Jonas, cash-posting accountant
Proto-persona · assumption-based, to validate
Goals
Close the month on time with zero unexplained cash
Defend every posting when auditors ask
Frustrations
Deposits arriving days before the remittance that explains them
Automation he can’t check, so he redoes it by hand
Journey
Where the product steps in
Bank feed02→Remittance→Auto-match / rule04→Exceptions01→Resolve03→Close05→Audit06
Numbers refer to the chapters below. Most deposits never need a person. The design effort goes into the few that do — and into proving what happened afterwards.
01 · Orient
Overview
Answer the question finance asks first: how much cash is still unexplained?
1
Lead with unreconciled cash
At month end the only number that blocks closing is what’s still unexplained. It gets the strongest weight and the only red on the page.
Considered insteadLeading with total received — reassuring, but not actionable.
2
Expected vs received over time
A dashed “expected” line built from remittances makes the gap visible week by week; hovering shows the shortfall.
3
Exceptions ranked, with confidence
Each exception shows the issue, the suggested match and how sure the system is.
02 · Triage
Deposits
Show every bank credit and how far along it is, without opening it.
1
Feed freshness on every account
A delayed bank feed makes totals wrong without looking wrong, so staleness is shown where people pick the account.
2
Status and confidence on every row
A label plus a small bar separate “done”, “check” and “investigate” at a glance.
3
Rule matches kept apart from auto-matches
Rules follow a decision a person already made; auto-matches are the system’s guess. Mixing them would blur what to trust.
Considered insteadOne combined “Matched” status.
03 · Resolve
Match workbench
Turn a discrepancy into an accounting decision in one screen.
1
Raw bank text next to the summary
Accountants trust what the bank actually sent, so the original description and trace number are shown verbatim.
2
An allocation that visibly adds up
Claims with variance are highlighted and totals sit at the bottom, so the gap is obvious before any explanation.
3
Explain, then offer options
The cause comes first, then three postings with their accounting consequence. The default is the most common choice, never an automatic one.
Considered insteadAuto-posting the difference — fast, but unsafe without review.
04 · Automate
Rules
Let people teach the system, safely.
1
Rules start from repeated decisions
The workbench suggests a rule after the same manual posting happens several times.
2
Preview against history first
A rule shows which past deposits it would have matched and whether a person did the same.
3
A second approver for ledger postings
Rules that create journal entries need four eyes — the same control auditors expect for manual entries.
05 · Close
Month-end close
Make “are we ready to close?” a yes/no question with reasons.
1
A checklist built from live data
Tasks complete themselves when the work is done — no manual ticking that drifts from reality.
2
Blockers say what and how much
“12 open · $86,420 (threshold $10,000)” tells the controller exactly what stands between the team and sign-off.
3
Locking explained before it happens
Closing changes how corrections work; saying so up front prevents panic edits after sign-off.
Considered insteadA confirmation dialog at the moment of closing.
06 · Prove
Audit trail
Show auditors who did what, when and why — people and automation alike.
1
People and automation in one timeline
A lightning mark identifies system actions and initials identify people, so every change has a visible author.
2
Before/after for every change
Reversals show exactly which fields changed, the journal entry created and the reason given.
3
Filters shaped like audit questions
“Everything automation did in October” or “all reversals” are one filter away, and exportable.
Trust
States & edge cases
First run, stale data, four-eyes approval and reversals: the moments where money could be posted wrong, designed to slow people down only when it matters.
System
Rules behind the interface
Small, explicit rules keep a dense interface consistent as it grows.
Confidence has three bands
≥ 90% green, 70–89% amber, below 70% red — always shown with the number.
96% · 82% · 61%
Money is monospace and exact
Working views never round; summaries can abbreviate but link to the exact figure.
$48,212.40
Every action has a visible author
Initials for people, a lightning mark for automation — in tables, timelines and approvals.
JK · ⚡
Outcome
Measuring it, and what’s next
As a concept, this has no live numbers — so here is what I would measure and test first.
How I’d measure success
Auto-match rate accepted without edits
Days to close the month
Unreconciled cash older than 30 days
Match reversals per month (a trust signal)
What I’d test next
At what confidence do accountants stop double-checking?
