One thread
Ask in your own words and get the answer in the same sitting. No ticket, no queue.
The answer, and the move it implies.
Ask in plain language. aiRA reads the warehouse you already have, shows its reasoning, and builds the move for a person to approve.
What it does
aiRA does not reply. It builds you something.
Not a report, and not a chat. The answer arrives with the reasoning that produced it and the move it implies, built against your own data and held for a decision.
The move
The segment and the journey are built in the same thread, then held until a person approves.
Your business context
Playbooks ship pre-built, and your own rules layer on in days rather than months of modelling.
Its working
The query, the assumptions and the sources come back with every answer.
Your domain
It reads loyalty terms like lapsed, liability and tier the way your programme means them.
How it fits
aiRA sits between your
data and your channels.
You already own the warehouse. You already own the channels. What is missing is the judgement in between, and today that runs on analyst tickets and Monday meetings.
System of record
Reads where your data already sits
The intelligence layer
System of action
Executes anywhere, in any channel
-
Zero copy
aiRA reads your data in place. No migration, no rebuild.
-
Nothing to rip out
It plugs into what you already own and pushes to what you already run.
-
A person approves
Every move waits on your approval before it goes live.
Security
Your data stays where it is, and nothing writes without you.
- The model is Anthropic Claude on AWS Bedrock, running inside your own region, stateless and with zero retention. Capillary never trains on brand data.
- Four layers before the model sees a row: database access control, RBAC scoping, masked-by-default views and payload truncation.
- Retrieval is tenant and brand isolated, so cross-brand analysis is architecturally impossible rather than disallowed by policy.
- Queries run SELECT-only in a sandbox with no outbound network, and every customer-impacting write stops at a person, on an audit trail.
-
SOC 2 Type II
-
CCPA
-
GDPR
Also ISO 27001 and PCI DSS certified, and PDPA compliant. HIPAA-capable under a BAA. Aligned to the NIST AI Risk Management Framework and the EU AI Act.
Getting started
Live on your own data in two to four weeks.
Four phases with a named output each, run by a Capillary success manager. Your team is needed in the first and the third.
-
Business requirements
Your business context, your tiers and the questions your team actually asks.
OutputA signed-off requirements document
-
Training and testing
aiRA is pointed at your warehouse and trained on your context. Read access is all it needs.
OutputA configured instance on your data
-
Brand testing
Around thirty representative questions, checked against your own numbers until the gap closes.
OutputA parity report you sign off
-
Go live
Seats get access, the approval gate is wired to the people who own it, and the credit ledger starts.
OutputaiRA live for your teams
The parity gate
99% agreement, minimum
aiRA is not switched on for a brand until it reproduces your own reporting at least that closely, so the first answer your team sees is one you can check against a number you already trust.
The question your team is still waiting on.
Thirty minutes, one real question, answered live. We reply within one business day with a time and the security brief.