About Balancely
Balancely turns Canadian bank and credit-card PDFs into reconciled Excel workbooks in under a minute. We exist because spreadsheets shouldn't be transcribed by humans, and accountants shouldn't have to clean up after a parser.
Our story
The first version of Balancely was a Friday-night script. One of us had a stack of twelve credit-card PDFs to expense, a tax deadline two weeks out, and a quote from a local bookkeeper that priced “data entry” at four hours of work. The script parsed one issuer, exported a single CSV, and saved that weekend.
Canadian statements are an awkward in-between. They're not standardized like the US-style OFX feed, and they're not as locked-down as European PSD2 APIs. Every issuer renders their own PDF layout, and most of them quietly redesign it once or twice a year. The tools that work elsewhere don't work here, and the tools that work here only work for one bank at a time. Most people end up with a manila folder of PDFs and an evening of typing.
We replaced the per-bank parsers with a single Statement AI extractor in early 2026. The product changed shape: instead of advertising bank coverage, we advertise reconciliation guarantees. The output is the same three-sheet workbook every time, and the reconciliation step verifies the parser got it right whether the input came from a template parser or an LLM. That's the part we're proud of — not the AI, but the discipline of refusing to ship a workbook that doesn't balance.
Timeline
2025 — Friday night
One bank, one CSV, one tax deadline. The first version of Balancely was twelve hours of work over a long weekend in November.
Q1 2026
Hand-coded template parsers for RBC, TD, Scotia, BMO, CIBC, and Canadian Tire. They worked, until banks shipped layout updates and then they didn't.
Q2 2026
A single Statement AI parser with a tight system prompt and a strict response schema. Coverage went from "the six banks we wrote" to "any Canadian statement we've seen."
Q2 2026
Credit-card statements flip every amount to negative outflows; bank statements keep canonical signs. The reconciler and the preview tabs agree with the .xlsx, so what you see is what you download.
Today
A small Canadian team building Balancely end-to-end — parsing, reconciliation, dashboard, support. No outsourced AI middleman, no surprise vendors.
What we believe
Your PDFs are processed in memory and discarded the moment the workbook is built. We persist the audit summary and nothing else — no transactions, no images, no third-party trackers.
A workbook that doesn't balance is a workbook your accountant has to redo. We check section totals + opening + Σ amounts ≈ closing on every statement, with a 2¢ tolerance, and surface drift loudly when it happens.
Three sheets. Same column order. Same date format. Same vendor normalization. Once you build a pivot table on top of our output, it never breaks because we shipped a new feature.
The team
We're an indie team building Balancely end-to-end. No outsourced parsing, no third-party AI middleman, no surprise vendors.
Parser, reconciler, workbook writer
We obsess over rounding modes, sign conventions, and edge cases on Canadian Tire FX statements so you don't have to.
Workbook layout, dashboard UX
Three sheets. Same column order. Same date format. We sweat the boring stuff so accountants don't notice us.
Triage, onboarding, edge cases
When the AI gets confused by a new bank layout, we hear about it first and fix the prompt before the next batch.
Frequently asked
Yes. Built in Canada, for Canadian PDFs, with Canadian privacy law in mind (PIPEDA-aware). We process in Vercel's US East region for latency reasons — the privacy contract you're entering covers cross-border handling.
Brittleness. Per-bank parsers break the moment a bank tweaks its PDF layout. A single AI parser with a tight schema and a strong reconciler handles layout drift on its own and reconciliation catches anything that slips through.
Try it. Statement AI handles arbitrary PDF formats — RBC, TD, Scotia, BMO, CIBC, Canadian Tire, and American Express are the ones we test most aggressively, but smaller credit unions and unusual layouts usually work too. If reconciliation fails, that's the parser telling you it isn't confident; report it and we'll tune the prompt.
Yes — the same engine handles personal and business statements. Larger statements (50+ pages) may need to be split because of the 300-second Vercel function cap.
No credit card. Sign up, upload a PDF, download a workbook. Takes about a minute.