It was a Tuesday in November 2023. We were three days into a month-end close that should have taken five, and our controller — let's call her Priya, because that's her name — was on hour eleven of trying to figure out why our bank feed had silently stopped syncing on the 14th. Two weeks of transactions. Gone. Or rather, not gone — sitting on the bank's side, perfectly visible in the bank's own portal, but invisible to our accounting software because some auth token had expired and nobody had been emailed about it.
The software in question, which I will not name but which everyone reading this can guess, had a "Reconnect" button. Clicking it did nothing. Clicking it again opened a chat widget where a bot asked us to please describe our issue. We described our issue. The bot suggested clearing our cache. We cleared our cache. The bot suggested a Tier 2 agent would email us within 24-48 business hours.
Priya looked at me across the table and said, with the particular calm of someone who has stopped being angry and gone several stages past, "I don't understand why this is still how accounting works."
Neither did I. So we went and built VeloLedger.
VeloLedger is an AI-native, double-entry accounting platform for Canadian SMBs and mid-market companies: ERP-grade depth — multi-entity consolidation, IFRS 15 revenue recognition, IFRS 16 / ASC 842 leases, dimensional segments, and real Canadian tax — at a fraction of what the incumbents cost once you add up their seats and add-ons. The AI does the bookkeeping: it reads invoices, codes bank lines, drafts entries, and watches for anomalies. A person approves before anything posts. That last part is the whole design, and it's the rule we'll come back to: the AI literally cannot post a journal entry.
What we wanted to fix
This isn't going to be a list of every grievance we have with the incumbent accounting tools. That list is long and most of it is petty. But there are five specific things that we built VeloLedger to fix, and they're the things we'd like you to hold us to.
1. Brittle integrations that fail silently
Every accounting product on the market pitches itself on integrations. "Connects to 800+ apps." "Sync with your bank in real time." What none of them tell you is that those integrations are mostly held together by screen-scraping, undocumented APIs, and OAuth tokens that expire on a rolling basis that nobody at the vendor seems to track.
The result is the experience we had that Tuesday: a bank feed that "works" 96% of the time and that fails the other 4% in a way that is invisible until you go looking. Multiply that across payroll, expense management, payment processors, e-commerce platforms, and inventory — and you have an accounting system that is structurally untrustworthy. You can't actually believe what it tells you, because you can't be sure all the data made it in.
We wanted integrations that announce their own failure, that are deterministic, and that don't depend on us being heroes about our own software. That's why we bet on MCP, but I'll get to that.
2. Month-end that takes a week instead of a day
A clean month-end close for a 40-person SMB should take a day. Two if your AP volume is high. We've seen finance teams routinely take five, seven, sometimes ten days — and almost always the bottleneck is the same: people manually reconciling things the software should have reconciled, manually categorizing transactions the software should have categorized, manually chasing receipts the software should have already matched.
"AI-powered categorization" has been a feature checkbox for a decade. None of it actually works at the level you'd accept from a junior accountant. Which brings me to the next thing.
3. AI bolted on, instead of AI built in
Every legacy accounting tool launched an "AI" feature in 2024. Most of them are a chatbot that can summarize a report you already had, or an autocategorizer trained on a fixed taxonomy that doesn't match your chart of accounts. They're cosmetic.
What we wanted was something different: an accounting platform where AI doesn't sit on top of the ledger but actually reads from and writes to it, with proper guardrails. An AI that can draft a journal entry the way a junior would — and that the controller can review, edit, or reject — and that learns from being corrected. An AI that can run a variance analysis and write the commentary, instead of just generating the table. An AI that can fail gracefully and tell you why.
To do that, you need the AI to have a real interface to your data — not just a SQL pipe into a vector store, but a structured, tool-call-shaped interface that exposes accounts, customers, vendors, journals, and policies as first-class objects. Which, again, is why MCP.
4. Multi-entity treated like an afterthought
If you run more than one company — a holding co with operating subs, a US parent with a Canadian sub, a brand portfolio — you know the experience. Either your accounting software charges you per entity and makes you log in separately for each, or it has "multi-entity" support that consists of running them in parallel with no real consolidation.
We built VeloLedger with multi-entity in the core data model. Entities, business units, departments, classes, and locations are all proper dimensions, not bolt-ons. Intercompany journals net out automatically. Consolidation isn't a separate product — it's a view. Eliminations are tracked, not hand-jammed in Excel and re-keyed every month.
This wasn't a feature we added late. It's the reason the schema looks the way it does.
5. Canada as a second-class citizen
This one is personal. About 70% of the cap-S Serious accounting tools on the market were built in the US and treat Canada as an afterthought. GST and HST handling is hacked together with custom tax codes. PST and QST are barely supported. T4A, T5, T5018 generation is "coming soon" or requires a third-party add-on. The CRA's Business Number structure — with its program account suffixes — doesn't map cleanly to anyone's data model.
If you're a Canadian SMB, or a US company with a Canadian sub, you've spent more hours than you care to count working around these gaps. We've spent those hours too. VeloLedger ships with proper Canadian compliance from day one: per-province sales tax, ITCs tracked at the line level, BN program accounts modeled correctly, and full T4A / T5 / T5018 generation. There's a separate post on what that looks like in detail.
The bet we made
Building a new accounting product in 2026 is, on the surface, an absurd thing to do. The category is mature. The incumbents are entrenched. CFOs and controllers do not, generally speaking, wake up wanting to migrate ledgers.
But there's a structural shift happening underneath all of this, and we're betting on it.
The rule underneath everything: the AI cannot post
The single design decision VeloLedger is built around is that the AI cannot post a journal entry — its tool surface has no post action at all. The agent can read your books, draft a bill, code a bank line, propose an adjustment, run a variance, and write a close memo. What it cannot do is commit any of that to the ledger. Every AI-generated entry lands in a review queue as a draft, and a human approves before it posts.
This matters because in 2026 every accounting product ships "AI agents." That is no longer the differentiator; it's table stakes. What none of the incumbents guarantee is that their agent can't write to the ledger. For them, "the AI won't post without approval" is a policy — a setting, a prompt, a workflow someone can reconfigure. For us it's architectural: there is no code path for the agent to post, so it can't, ever, by any prompt or bug. That's the whitespace we're building in, and it's why we can hand the AI real work without handing it the keys to your books. The signature line we keep coming back to: agents draft, humans post.
ERP-grade depth without the ERP price
The other half of the bet is that a small or mid-market finance team shouldn't have to choose between "cheap and shallow" and "deep and six figures." The cheap tools (QuickBooks, Xero) don't do native multi-entity consolidation, native leases, or native revenue recognition. The deep ones (Sage Intacct, NetSuite) do — but they're quote-only, commonly land between five and six figures a year all-in, and take months to implement.
VeloLedger puts the deep feature set in the SMB tier: multi-entity consolidation with FX translation and intercompany eliminations, native IFRS 15 revenue recognition (deferred revenue schedules and RPO, with no SaaSOptics/Maxio bolt-on), a native IFRS 16 / ASC 842 lease sub-ledger, business-unit and department segments on every statement, and Canadian tax done properly. Pricing is per organization — from $29/month, with unlimited free viewer/auditor seats and a few poster/approver seats included per plan, and AI metered in plain credits — and materially cheaper than the incumbents once you add up their seats and add-ons.
MCP and an open API, so you can bring your own agent
Under the hood, every meaningful operation in the ledger — read a balance, draft a journal entry, generate a T4A, run a variance — is exposed through a native MCP server and a public REST API. That means an AI agent (ours, or yours, or your auditor's) can work against your books with the same guardrails a human would have: permissions, audit trails, and approval workflows all enforced server-side. And the same architectural limit applies to any agent you bring — the post action isn't on the tool surface for anyone. Everything an agent does is logged the same way a human's action would be, on an immutable, hash-chained audit trail.
This is unglamorous infrastructure work. It is also the difference between AI that occasionally hallucinates a credit and AI that you can actually trust in your books.
Sibling products as first-class data sources
VeloLedger does not exist in isolation. It's part of a family — VeloPulse (HRIS/payroll), Velo Expense Management, Velo CRM — wired to VeloLedger via MCP, not via brittle webhooks or nightly CSV exports. VeloPulse and Velo Expense Management MCP integration is available today; Velo CRM integration is rolling out now.
What that means in practice: when a contractor invoice is approved in Expense Management, the journal entry lands in the GL in seconds, with the right account, dimension, and tax code, and the source document attached. When payroll runs in VeloPulse, the burden allocation hits the right departments. And as Velo CRM integration rolls out, closing a deal will draft the deferred revenue schedule for your review.
You can use VeloLedger without the rest of the family. But if you use them together, you stop having "integration problems" because the integrations aren't integrations — they're shared infrastructure.
Built for an era where AI does the boring parts
We don't think AI is going to replace controllers. We think AI is going to replace the parts of a controller's job that nobody enjoys — the categorization, the chasing, the re-keying, the variance commentary that says "expenses up $14K, driven by software subscriptions" for the seventeenth month in a row.
What gets left behind is the judgment work: the close memo, the unusual transaction, the audit conversation, the decision about a reserve, the explanation to the CEO about why the gross margin moved. That's what accountants are good at. That's what they should be doing. VeloLedger is built to take the rest off their plate.
What's next — and what's not there yet
We want to be honest about the state of the product. Here's what we have today, and what's still coming.
What works today: Full double-entry general ledger with draft→approve→post and an immutable, hash-chained audit log; AP with bill capture, POs and 3-way match, and value-range approval tiers; AR with branded invoices, credit notes, recurring billing, and statements; bank feeds (Flinks live, Plaid adapter) with proper failure handling plus AI categorization and one-click book-and-reconcile; AI receipt/invoice extraction from an email inbox; Controller Copilot (chat your books, drafts adjustments into the review queue, never posts); an always-on anomaly scan; a natural-language report builder. On the depth side: multi-entity consolidation with FX translation and intercompany eliminations, native IFRS 15 revenue recognition, a native IFRS 16 / ASC 842 lease sub-ledger, business-unit and department segments on every statement, multi-currency with Bank-of-Canada daily FX and period-end revaluation, a fixed-asset register with CCA (T2 Schedule 8), and Canadian tax compliance including GST/HST/PST/QST, GST34 filing with GL reconciliation, T2 GIFI mapping, and T4A slips. Plus a native MCP server and public REST API, MCP integration with VeloPulse and Velo Expense Management, role-based access with segregation of duties on posting, Google SSO and TOTP MFA, and data residency in AWS Canada (ca-central-1).
What's coming next — and what's honestly not there yet: Native mobile apps (a PWA today), SOC 2 attestation (planned), SAML SSO (Google SSO ships today), a broader app marketplace, and native payroll (we partner with Humi/Wagepoint rather than build it). An inventory and COGS module and expanded e-commerce connectors are rolling. Velo CRM MCP integration is rolling out now (VeloPulse and Velo Expense Management MCP integration are already available).
What we're not building, on purpose: A 200-app marketplace of mediocre integrations. A "drag-and-drop dashboard builder." A custom report writer that requires three days of training. A separate "consolidation product" that costs extra.
If any of that disqualifies us for you, that's fair. If it sounds like a relief, that's the audience we're building for.
Frequently asked questions
What is VeloLedger?
VeloLedger is an AI-native, double-entry accounting platform for Canadian SMBs and mid-market companies. The AI reads invoices, codes bank lines, drafts entries, and watches for anomalies, and a person approves before anything posts. It pairs ERP-grade depth — multi-entity consolidation, IFRS 15 revenue recognition, IFRS 16 / ASC 842 leases, dimensional segments, and Canadian tax — with per-organization pricing (from $29/month) well below the incumbents.
Why did you build another accounting platform in 2026?
Because the category splits badly into "cheap and shallow" (QuickBooks, Xero — no native multi-entity, leases, or revenue recognition) and "deep and six figures" (Sage Intacct, NetSuite — quote-only, long implementations). VeloLedger puts the deep feature set in an SMB-priced tier, and does it AI-native from the schema up rather than bolting a chatbot onto a legacy ledger.
Can VeloLedger's AI post journal entries on its own?
No. The AI's tool surface has no post action, so it can draft entries into a review queue but cannot commit anything to the ledger — a human always approves first. This is an architectural guarantee, not a configurable setting, and it applies to any agent you connect through the MCP server or REST API as well.
How is VeloLedger cheaper than Sage Intacct or NetSuite?
Sage Intacct and NetSuite are quote-only and commonly run from roughly $9k into six figures a year once modules and implementation are added. VeloLedger uses transparent per-organization plans (Starter $29, Growth $79, Scale $199 per month, plus custom Enterprise) with unlimited free viewer/auditor seats and AI metered in plain credits, so multi-entity consolidation, leases, and revenue recognition come without the enterprise-ERP price tag.
Is VeloLedger built for Canadian accounting?
Yes. VeloLedger ships GST/HST/PST/QST handling, GST34 filing with GL reconciliation, automatic input tax credits, T2 GIFI mapping, CCA (T2 Schedule 8), and T4A slips, with data residency in AWS Canada (ca-central-1). Canadian compliance is native, not a US-built tool with tax codes hacked on top.
One last thing. If you've read this far, you probably have your own Priya-on-hour-eleven story. We'd genuinely like to hear it — not because we're going to sell you something, but because the things that frustrate you about accounting software are the things we want to keep fixing. Email us. Or just book a 15-minute conversation. No deck.
— The VeloLedger team