Products
Built for the work nobody demos.
Our product line comes out of consulting engagements, not brainstorms. When the same expensive, evidence-heavy, repetitive process shows up at the fourth institution in a row, we build for it.
Five products: Passage, Loom, Redress, Attune and Relay.
Status is stated honestly on each product below. We would rather tell you something is at design-partner stage than imply a production install base we do not have.
Current portfolio
Five problems we are building against.
PAAR Passage
Migration accelerator
The problem. Migrations are the largest, riskiest programs a financial institution runs — off a core banking or lending platform, off a mainframe, off an on-premise warehouse. The logic that matters was written by people who left a decade ago, lives in COBOL and stored procedures, and is documented nowhere. Most of the budget goes on discovering what the current system actually does and on proving the new one agrees with it.
Our approach. Read the legacy estate and produce documentation a human can review: what each component does, what calls it, what is dead code. Recover the business rules into a specification your subject matter experts sign off on, so the rewrite is validated against agreed intent rather than reverse-engineered blind. Generate target schema, code scaffolding and tests instead of one-shot conversion. Map every field from source to target with its transformation rule and lineage. Then run source and target in parallel over real volumes and classify every difference — timing, rounding, genuine defect — rather than handing the team a million-line diff. Cutover ends with a reconciliation and evidence pack risk and audit can sign.
One accelerator, three migration types: core platform, legacy code, and data platform to cloud.
PAAR Loom
Structured & unstructured data processing
The problem. The answer to most questions in a bank sits half in a core system table and half in a PDF, an email thread, a scanned form or a call recording. Teams either ignore the unstructured half or pay people to retype it.
Our approach. One ingestion layer for both: classification and extraction from documents, emails, forms and transcripts; joins against core banking, CRM and warehouse tables; entity resolution across systems that spell the same customer four ways; and lineage on every field so a downstream number can be traced to its origin. This is the layer the rest of our products run on, and it can be deployed on its own.
PAAR Redress
Customer remediation acceleration
The problem. Remediation programs are slow, expensive and reputationally loaded. Identifying the affected cohort, recalculating what each customer is owed across years of product and rate changes, evidencing the method to a regulator, and communicating with hundreds of thousands of customers is done with spreadsheets, contractors and long timelines.
Our approach. Reconstruct the affected population from source systems with an auditable rule set, run recalculations with every input traceable, generate the evidence pack and methodology documentation, and draft customer correspondence for review. Exceptions and edge cases are surfaced rather than silently averaged. Every figure recomputes if the rules change, so a regulator challenge does not restart the program.
PAAR Attune
Voice AI for contact centres
The problem. Contact centres hold the highest-risk conversations in the bank — hardship, complaints, collections, disclosures — and review a tiny sample of calls after the fact. Agents spend a large share of every call typing notes instead of listening.
Our approach. Real-time transcription with agent assist that surfaces the right policy or script at the right moment, automatic post-call summaries and wrap-up notes written back to the CRM, and quality and compliance review across every call rather than a 2% sample — flagging missed disclosures, vulnerability and hardship signals, and complaint indicators for a human reviewer. Customers are told when AI is in use, and the assistant never speaks to the customer on the bank's behalf without explicit sign-off.
PAAR Relay
Workflow automation, approvals intact
The problem. Operational processes — onboarding, maintenance requests, disputes, hardship applications, account closures — span five systems and a queue of manual handoffs. Traditional RPA breaks the moment a screen changes or a case does not fit the template.
Our approach. Model the process as steps with explicit decision points, let the system execute the deterministic parts and prepare the judgement calls for a human, and route genuine exceptions rather than guessing. Integrates with the case and BPM tools you already run. Delegations, four-eyes approvals and segregation of duties are enforced, not bypassed, and every action is logged with who or what performed it.
Build principles
Non-negotiables in everything we ship.
Runs where your data lives
Your cloud tenant, your VPC, or on-premise. No training on your data. No silent egress. Deployment topology is agreed before the contract, not after.
Cited or it doesn't ship
Every generated statement links to the source document, record or system of origin. A reviewer can verify any claim in one click.
Human owns the decision
Our products draft, assemble, rank and flag. They do not approve credit, close alerts, or take customer-affecting action on their own.
Evaluated continuously
Golden datasets, regression gates before release, and live quality monitoring with alerting on drift. Results shared with you, not just with us.
Audit-ready by construction
Immutable logs of inputs, model version, prompt version, output and reviewer action — structured for the way examiners actually ask.
Model-portable
No single-vendor lock-in at the model layer. As the frontier moves, you move with it without a rebuild.
Design partners
We build with three institutions per product, not thirty.
What a design partnership involves
- You get preferential commercial terms, direct influence over the roadmap, and a working system in your environment months before general availability.
- We ask for access to the real workflow and its subject matter experts, a sponsor who can clear internal blockers, and honest feedback when something is not working.
- We start with a defined pilot scope, a measured baseline, and a written success threshold agreed before any build begins.
- Data handling is agreed first: where it sits, who can see it, how long it is retained, and how it is deleted at exit.