AVA Credit Inc. builds Ava, an accelerated virtual analyst for Canadian financial services work.
The company exists to put the best available AI in the hands of the licensed professional, on infrastructure the industry trusts. The operator brings intent, judgment, context and accountability. Ava brings speed, precision and trust to the work.
Ava 1.0 is the first release, and it is the governance layer. It removes client identity from a financial document so the work can be done in the best AI available, then puts the identity back into the finished work. The document leaves. The client’s name does not.
Bank-grade data governance, paired with the best enterprise AI models on the market.
An institution solves this by building. Private deployment on infrastructure it controls, retrieval built in-house, a model risk function, specialist vendors, and people whose job is to keep the whole thing inside policy. That path works, and the banks are well down it. It is also priced for an institution, and it is the reason a bank does not need to send anything anywhere.
A three-lawyer practice, a ten-person accounting firm, a sole mortgage broker. The same duty of confidentiality, the same client files, none of that capacity. Two options are actually on the table. Build an apparatus they cannot afford and could not match anyway, or leave the tools alone and spread the financials by hand while the institutions accelerate.
Ava is a third option. An enterprise account with a frontier provider, and Ava in front of it. Govern the data rather than the infrastructure. The identity stays inside the firm, the model does the work, and nobody has to stand up a platform to get there.
For the era of governed AI in financial services.
Where domain expertise meets AI acceleration
Identity is masked before it crosses into any outside system, and restored on the way back.
What identifies a client and what makes a deal analyzable are two different things sitting in the same document. Names, addresses, social insurance numbers, business numbers and account details identify. Loan amounts, appraised values, coverage ratios, rates and three years of margins are what make the file analyzable. Ava removes the first and leaves the second exactly as written, so the model still sees the whole financial picture. It simply does not know whose deal it is.
The key that maps a token back to a real name stays inside the firm, beside the client documents it describes. It is never transmitted. Every restore produces a report that is read before the file goes anywhere.
Ava is not an alternative to an enterprise AI agreement. It is what you send through one.
Two controls, doing two different jobs. A team or enterprise agreement with a frontier model provider governs the vendor: SOC 2 attestation, contractual terms on retention and training, administrative control over accounts. Ava governs the content. The first is a promise about handling. The second is a fact about what was sent.
The enterprise agreement Covers the vendor. Security attestation, terms on retention and model training, administrative control, and the audit surface a firm needs when it is asked about third-party risk.
Ava Covers the content. Direct identifiers are removed before the document crosses the boundary, and the key that maps them back never leaves the firm at all.
Neither one replaces the other. Under an enterprise agreement alone, the client’s name, address and identifiers still leave the firm and sit in another company’s system under another company’s terms, however good those terms are. Ava means they never make the trip. Run together, a firm can hold a contract covering the vendor and still say, as a matter of record, what did not leave.
Residency is the clearest case. An enterprise agreement is a promise about handling, not a change of address: the servers are somewhere else, usually the United States, and no contract moves them. A bank avoids the question by building its own system on infrastructure it controls. A ten-person firm cannot. Ava does not stop the data crossing the border. It changes what crosses.
Because Ava sits before the boundary rather than behind it, it does not care which model is on the other side. Whatever proves best next year is a configuration change, not a rebuild. The pairing is with the tier, not the brand.
Ava reduces exposure on any account, including a consumer one. We do not present it that way. Client work does not belong on a consumer account in the first place, and a tool that made that arrangement feel acceptable would be doing harm rather than good.
The deadline is not arriving to an empty field. It is arriving to a habit.
OSFI’s Guideline E-23 on model risk management takes effect on 1 May 2027, and it defines a model broadly enough to include AI tools that materially affect decisions. It binds federally regulated institutions directly. Independent firms are reached a different way: through B-10 third-party risk, when the institutions they serve begin asking their channel what AI touches client data, and through the privacy and professional obligations that already apply.
Those questions arrive before the deadline, not after. The firms that will answer them well are not the ones that waited, and they are not the ones that told their people to stop. They are the ones that governed the data instead of the behaviour.
Five years of live operational research, and one product.
The work behind Ava was done on real transactions under real conditions, not in a lab.
A fuller site is being built. Until then, the door is a mailbox.
info@avacredit.ca