Upload Your Transactions

Drop a CSV from your bank. We’ll figure out the format automatically.

Drop CSV file here

or browse files

Works with Wealthsimple, RBC, TD, Scotiabank, or any CSV
Your money, actually understood
DemoGitHub
Scroll
~50 transactions a month.
Your bank gives you a list.
Same format since 2006. Date, merchant, amount, scroll. But your spending data isn't just line items — it's behavioral data. Correlations between categories, subscriptions quietly increasing, cycles tied to your paycheck. Patterns that exist in every bank statement but nobody has the tools to find.
Sift finds them.
Recent TransactionsAmount
UBER EATS 4F2K9Dining$34.50
LOBLAWS #1192Groceries$87.23
NETFLIX.COMSubscription$22.99
PRESTO TRANSITTransport$12.00
STARBUCKS #8834Dining$6.75
METRO INC ONGroceries$52.10
SPOTIFY P1A8F2Subscription$11.99
DOORDASH*PAD THAIDining$28.40
SHELL STN 0442Transport$65.00
3 categories identified · 2 patterns detected
The idea

Understanding your finances is personal. Where you spend, why you spend, what trade-offs feel right — that will always be human.

But scanning 600 transactions for price creep? Cross-referencing categories across 11 months? Detecting behavioral cycles tied to your paycheck? That's not human work. That's exactly what a system should do.

Sift handles the cognitive load — the statistical analysis, the pattern recognition, the anomaly detection. Then it shows you what it found. You decide what matters.

−0.72
Spending correlation
GroceriesDelivery
Spending connection
Groceries and delivery move in opposite directions
When one rises $80, the other drops $60. Cook more in a month, save roughly $80. Nobody spots these patterns by scrolling a bank statement. Sift can.
$84
Per year, unchosen
$15.99$16.99$17.99$22.9920232024
Price creep
Netflix: $15.99 to $22.99. You didn't notice.
Each bump was small enough to ignore. Annualized: $84 you never chose to spend. Multiply across 8 subscriptions and the number gets uncomfortable.
$340
Flagged instantly
$340
Anomaly detection
A charge you'd miss scrolling
New merchant. Three times your average transaction. Sift flags outliers, spending spikes, and merchants you've never used before — the things that hide in plain sight on a bank statement.
40%
Of discretionary budget
40%
Days 1–3
35%
Days 4–14
25%
Days 15–30
Spent within 3 days of payday. Every month.
9 of 11 months. Whether that's a problem is your call, not ours. The system counts. You interpret.
What you see
847 transactionsUpload CSV
Total Spent
$4,847
Jan — Nov 2024
Monthly Avg
$441
across 14 categories
Biggest Swing
Dining
32% of variance
Could Save
$2,250
per year identified
Ask Sift anything about your spending...
SubscriptionsTrendsSavings
Spending by Category
Dining
$1,420
Groceries
$1,140
Transport
$690
Shopping
$547
Subs
$375
1
$2,250/yr
Reliable
8 active subscriptions at $187/mo
Netflix has increased 43.7% since you subscribed. Three streaming services overlap at $53/mo.
2
$1,440/yr
Reliable
Payday spending concentration
40% of discretionary budget spent within 3 days of deposit, 9 of 11 months.
3
$960/yr
Likely
Groceries ↔ Delivery inverse
When grocery spending rises, delivery drops proportionally — a strong, consistent pattern.
How the agent works
01
Ingest
Auto-detects bank format. Cleans merchants, dedupes, scores quality.
Deterministic
02
Categorize
Rules handle 70% at zero cost. LLM only for the ambiguous rest.
RulesLLM
03
Plan
Not a fixed pipeline. Profiles your data — months, categories, income — and skips tools it can't trust.
Agent
04
Analyze
IQR outlier scoring, subscription vs. habit detection, payday cycles, FDR-corrected correlation, Monte Carlo stress testing. Every number computed on your data.
Statistical
05
Synthesize
Cross-references results across tools. Ranks insights by dollar impact. The LLM never invents a dollar amount.
LLM
06
Ask
"What if I cancel Netflix?" routes to a cancellation simulator. "How long would savings last?" runs Monte Carlo. Tools compute, LLM explains.
AgentLLM
Agent decision log
scanning 847 transactions
11 months of data · 14 categories · income detected
───────────────────────────────
✓ anomaly_detection — IQR outlier scoring, always enabled
✓ subscription_hunter — recurring vs. habit filtering
✓ temporal_patterns — payday cycles, 11 mo ≥ 3 mo min
✓ correlation_engine — FDR-corrected, 14 cats, 11 mo
✓ spending_impact — variance drivers, 11 mo ≥ 6 mo min
───────────────────────────────
full suite enabled · executing 5 tools
done in 2.3s
Not a fixed pipeline. Tools qualify or get skipped.
Each tool has hard data minimums. Correlation needs 3+ months and 3+ categories. Spending impact needs 6+. If your data can't support a reliable result, the tool is skipped — and you're told why.

Tools that qualify run in parallel: IQR-based anomaly detection, subscription hunting that distinguishes recurring charges from habits, payday cycle detection, cross-category correlation with false-discovery-rate correction, and Monte Carlo stress testing across 1,000 simulations. Every number is computed on your data. The LLM never invents a dollar amount.
Sift does the work.
You make the call.

Sift might flag that you spend 40% of your budget within 3 days of payday. But maybe those days are date night and your kid's soccer. That context? Only you have it. Sift finds the patterns. You decide what they mean.

The system does
You do
Find patterns across 12 months of data
Decide which patterns matter to your life
Quantify the dollar impact of each pattern
Decide which trade-offs are worth making
Rank insights by statistical confidence
Decide if $53/mo in streaming brings you joy
Detect behavioral triggers and cycles
Decide if the behavior is intentional

Your bank statement knows more than you think.

Drop a CSV. Get back the patterns, the correlations, the price creep — everything that was always there but never visible. Takes about 10 seconds.

Works withRBC·TD·BMO·Any CSV