Aspen BankPolicy-driven
Declined
No offer
Income CV 0.96 exceeds policy maximum 0.35.
Built for Canadian credit unions and lenders
Approve the borrowers a credit score misses.
Farmers, gig workers, newcomers and seasonal businesses look risky on a credit score. Berry reads their cash flow, explains every decision in plain language, and runs the loan through funding, collections and reconciliation.
Try the live demoA portfolio project with fictional lenders. The live demo needs no sign-up and runs on synthetic data.
ONE APPLICANT · THREE RULEBOOKS
Hollis Family Farm
Grain farm, Saskatchewan · asking for a seasonal operating line
KiteAutomated
$8,000
Business line of credit
Limit reduced to $8,000 from $60,000: 2x median monthly net cash flow.
Northgate Credit UnionRelationship
$60,000
Seasonal operating line
Seasonal: 65% of income in September to November. Peak-season surplus $189,035 covers the line 3.15x (minimum 1.15x). Assessed on annual cash flow.
Northgate, Kite and Aspen are sample rulebooks. Each lender's policy runs as settings, not code.Try 13 more applicants in the demo →
Four tools. One loan record.
Use them together, or start with the one that fixes the biggest bottleneck.
DecideCash-flow underwriting on the lender's policy, with reasons a borrower can read and an approver can sign.See a decision →FundPay out with a sample AFT file or simulated Interac e-Transfer, after a name-and-account check.How funding works →CollectPre-authorized debits timed to when money lands, with safe retries when a payment fails.See a loan ledger →OperateWork queues with deadlines, second-person sign-off and an audit trail nobody can edit.See the console →
One system from application to ledger.
Replace the hand-offs between an origination tool, a payments file and a spreadsheet. Berry takes the loan from consent to the last reconciled payment.
Step 01 of 07
The applicant decides what Berry can see.
Before any data is read, the screen spells out what is shared, why, for how long and who receives it. Consent can be revoked later in one click.
Step 02 of 07
Identity and business checks, with reasons.
ID, name and date of birth, sanctions screening and business registration. Clear cases pass on their own. The rest go to a reviewer with a note on why.
Step 03 of 07
Three lenders read the same cash flow.
Each lender applies its own rules and explains the result in words a borrower can follow, along with what would change the answer.
Step 04 of 07
Money out on the right rail.
Pay out with a sample AFT batch file or simulated Interac e-Transfer, plus a simulation of the Real-Time Rail, which launches in phases from Q4 2026. Instant payouts run a name-and-account check first.
Step 05 of 07
Payments that land after the money does.
Pre-authorized debits scheduled around payday, or around harvest, with the Rule H1 agreement items checked off.
Step 06 of 07
A missed payment is a conversation.
One retry timed to the next deposit, a plain note to the borrower, then an outreach task. Hardship options go to a lender for approval.
Step 07 of 07
Every cent accounted for.
Each collection and return is matched to a double-entry loan ledger. Anything that doesn’t match lands in the reconciliation queue.
What a lending team would work in every day.
Cash flow, the decision, its reasons and who needs to sign off, on one screen.
berry / workbench / hollis-family-farm
Hollis Family Farm
Grain farm · Saskatchewan · Business
Most income arrives at harvest, input costs land in spring, and the months between are quiet.
- Income pattern
- Seasonal
- Months of history
- 12
- Returned payments, 12 mo
- 0
- Member since
- 2004
- Products held
- 4
Cash flow, last 12 monthsInOut
Northgate decisionApproved
$60,000
Seasonal operating line · interest-only, repaid after peak season
- Most of your income arrives September to November. We've set the line up so you draw on it in quieter months and repay it from those sales.
- You've been a member for 22 years and hold 4 products with us.
- None of your payments bounced in the last year.
Needs branch manager sign-off. Above the lending officer's $50,000 authority.
Compare all three lenders| Hollis Family Farm | Aspen Bank | Kite | Northgate Credit Union |
|---|---|---|---|
| Data used | Same 12 months of bank data | Same 12 months of bank data | Same 12 months of bank data |
| Reads income as | Income CV 0.96 exceeds policy maximum 0.35. | Seasonal product not offered; standard revolving line sized on median month $4,189. | Seasonal: 65% of income in September to November. Peak-season surplus $189,035 covers the line 3.15x (minimum 1.15x). Assessed on annual cash flow. |
| Outcome | Declined | Approved | Approved |
| Limit | — | $8,000 | $60,000 |
| Repayment | — | PAD · borrow and repay as needed | PAD · interest-only, repaid after peak season |
| Signed off by | Automatic | Automatic | Branch Manager |
| Compliance flags | None | None | None |
Shaded rows are where the lenders disagree. The data going in is identical.
Hollis Family Farm · operating line
$0.00
Repaid in full · final payment Nov 1, 2027 (simulated)
- Limit
- $60,000.00
- Peak drawn
- $29,012.03
- Season interest
- $816.07
- Repayment
- PAD · interest-only, repaid after peak season
- Ledger
- Balanced
| Date | Entry | Amount | Balance |
|---|---|---|---|
| 2027-05-03 | Draw | 11,452.05 | 11,452.05 |
| 2027-06-01 | Interest charged | 75.87 | 11,527.92 |
| 2027-06-01 | PAD collected | −75.87 | 11,452.05 |
| 2027-06-01 | Draw | 13,241.21 | 24,693.26 |
| 2027-07-02 | Interest charged | 163.59 | 24,856.85 |
| 2027-07-02 | PAD collected | −163.59 | 24,693.26 |
| 2027-07-02 | Draw | 4,318.77 | 29,012.03 |
| 2027-08-02 | Interest charged | 192.20 | 29,204.23 |
| 2027-08-02 | PAD collected | −192.20 | 29,012.03 |
| 2027-09-01 | Interest charged | 192.20 | 29,204.23 |
| 2027-09-01 | PAD collected | −9,862.88 | 19,341.35 |
| 2027-10-01 | Interest charged | 128.14 | 19,469.49 |
| 2027-10-01 | PAD collected | −9,798.82 | 9,670.67 |
| 2027-11-01 | Interest charged | 64.07 | 9,734.74 |
| 2027-11-01 | PAD collected | −9,734.74 | 0.00 |
Nothing waits in an inbox.
Reviews, referrals, missed payments and reconciliation breaks each become a task with an owner and a deadline. The back office sees what's late before a member does.
Work queuesFeb 4, 2027, 09:05 (simulated) · 3 open · sorted by SLA
- Fennick Hardware: returned PAD not appliedBreachedReconciliation breaks · M. Tran
- Quillfeather Plumbing & Heating: stopped or refused debit2 business days leftCollections outreach · J. Okafor
- Petalwick Florals: approve a 3-month deferral3 business days leftHardship approvals · R. Bisset
You prepared this override, so you can't approve it. A second person has to sign off. Your attempt is in the audit trail.
Time to funding, by rail
Simulated medians across 30 funded loans
- AFT batch
- 23 hours
- e-Transfer
- 20 minutes
- Real-Time Rail (simulated)
- 10 seconds
The dashboard also tracks cure rate, delinquency, SLA attainment and cost per loan.
- Maker-checker. Overrides and hardship approvals always need a second person.
- Policy changes with a preview. “This change would have approved 4 more of the 51 applications in the demo.”
- An audit trail you can't edit. Every decision, override and action, logged and exportable.
Concept
How it would plug in
Berry would sit between a lender's core banking system and its payments provider. It wouldn't replace either one, and the lender's team would keep the final say on every decision.
- Core banking
- Would take member and account data in and send booked loans out, by API or nightly file.
- Applications
- Could use Berry’s application flow, or take files from the origination tool a lender already has.
- Bank data
- Would start with uploaded statements, then use consent-based data feeds as the open banking framework goes live.
- Payments
- Would produce CPA 005 files for a lender’s clearing provider, plus Interac e-Transfer.
- Sign-in
- Could use single sign-on with a lender’s identity provider, with roles that match its approval authority.
Built for Canadian rules and rails.
Designed to cut custom work for compliance and payments teams. Each item links to the rule it follows and its current status.
- Consent-first data sharingBuilt around the consumer-driven banking framework’s consent principles. The Act is passed but not yet in force; its regulations are in draft.
- PAD repaymentDebit schedules and agreements modelled on Payments Canada Rule H1.
- AFT and instant fundingSample CPA 005 batch files, with simulated Interac e-Transfer and Real-Time Rail payouts.
- FINTRAC checksCash of $10,000 or more within 24 hours goes to review. It never touches the credit decision.
- March 2026 NSF rulesRetries are timed so a borrower is never charged a second NSF fee inside two business days.
- Newcomer credit pathA starter limit and clear milestones that fill the gap foreign credit reports don’t cover.
Decisions you can check.
Rules, not guesses. Limits, prices and approvals come from tested code. The same file gets the same answer every time.
Blind to who you are. Name, age, gender, nationality and postal code are never used. A test shuffles them on every build to prove it.
AI drafts the memo. The engine and a human make the decision. Every figure in the draft is checked against the engine before anyone sees it.
Tested in public. Every release runs against a set of applicants with known right answers. See the results, including the misses →
How it's built
- Synthetic data onlyYes
- No personal data sent to AIYes
- Append-only, hash-chained audit trailYes
- Maker-checker on overrides and sign-offsYes
- Deterministic, tested decisionsYes
- Rate-limited AI routeYes
Find the good borrowers you're turning away. First findings in a week.
Berry is a portfolio project. This is how I'd run a pilot with a lender.
A 30-day shadow pilot on one product. No borrower is affected, and you finish with a go or no-go.
- Week 1Load your rulesYour limits, thresholds and approval authority, plus last year's declined and referred applications for one product.
- Week 2Find the missed approvalsRe-run those files on cash flow. See who your policy declined that could likely have repaid: seasonal businesses, gig workers, newcomers with thin files.
- Week 3Run in the shadowBerry decides alongside your team on new applications. Your staff still make every decision; nothing reaches the borrower.
- Week 4Decide with evidenceA short readout: approvals you're missing, the risk behind them, time saved per file, and a go or no-go for a live trial.
Common questions
The lender's team. Berry recommends a decision with its reasons, and the lender's approval authority decides who signs off. Overrides are allowed, logged and checked by a second person.
In Berry, limits, thresholds, rates and approval tiers are settings, not code. Every change is versioned, and its effect on past applications can be previewed before it goes live.
First findings within a week of receiving past applications; a shadow pilot on one product would take about 30 days.
A plain-language reason, what would change the answer, and a way to ask for a review. The lender's team sees the same decision in lender terms.
Berry is a working portfolio project by Joshua Osunbor. The rules, ledger and operations console are real and tested; the lenders and borrowers are fictional, and all data is synthetic.
No. Limits, prices and approvals come from tested rules. AI only drafts the credit memo, and every figure in it is checked against the engine.
About this project
Berry is a portfolio project by Joshua Osunbor: fictional lenders and synthetic data, with real, tested rules, ledger and operations console. Questions about how it works, or how it would run with a real lender? I'm happy to talk it through.