IOanyT (our own product) (India · AWS Mumbai)
The Heartbeat: why we sell call QA as a service, not as software
We built a call-review product, then had to decide how a contact centre should actually use it. These are the choices we made, what we rejected, and what we would change.
Quick Facts
Consulting case
At a glance
- Who
- IOanyT, deciding for its own product: The Heartbeat, call QA for Indian collections and contact centres.
- Decision at stake
- Ship AI call scoring as self-serve software, or as a service in which people verify every flag.
- What we chose
- A managed service: AI reviews every call, a trained analyst verifies each flag, and the client receives a weekly report.
- Path
- Built in-house. Offered as a two-week pilot on the client’s own calls, then a subscription.
The business problem
Contact centres review a small sample of their calls by hand, so most risks are never seen. Conversation-analytics tools exist, but they are sold as software and run as a project: someone has to configure the rubric, tune the flags, triage the alerts and build the reports. Most contact centres have nobody to do that. The software works; the programme does not.
Diagnosis, including what we decided against
What we found while building it:
- Unverified AI flags are not trusted. AI-only tools flood operations with false positives, and a flag nobody believes gets ignored.
- Another dashboard is another unused tool. Buyers do not want to train staff or change workflows to get value from a review.
So, for our own product, we decided against three things:
- Sampling. Coverage is the product: every call is reviewed.
- Handing over raw AI output. Every flag is verified by an analyst before it reaches the client.
- A dashboard for clients to operate. The client reads a report; we do the work in between.
The decision
Decided by: IOanyT, as the product’s owner.
Options rejected: self-serve analytics software, AI-only scoring, and sampled review.
Governance: nothing reaches a client until an analyst has verified it. Access to a client’s data is limited to the analysts assigned to that account, and the data stays in India, in AWS Mumbai.
What we built to prove it
The client grants read access to its call-recording storage once, in about 20 minutes. Calls are processed overnight and transcribed in Hindi, English and Hinglish. AI flags conduct and compliance risks across more than 670 behaviour categories, analysts verify each flag, and a prioritised PDF arrives every Monday at 11:00 IST. It is available on AWS Marketplace.
Technical detail
Whisper speech-to-text with speaker diarization, a speech-emotion model, and a Node.js back end on AWS Lambda, with S3 and RDS for storage. Hosted in AWS Mumbai (ap-south-1).
Evidence the client receives
A weekly report: an executive summary, a severity breakdown, the week’s most critical calls and an agent leaderboard. Every flag in it has been verified by an analyst.
What the client owns
This is a service, not a build. The client keeps its recordings in its own storage, and we read them with the access the client grants. The client does not receive the code: The Heartbeat is our product. (When we build a system for a client, the client owns the code.)
What we would do differently
The service model ties quality to analyst headcount, so analyst capacity, not software, sets how fast the product can grow. If we started again, we would plan analyst capacity alongside the software from day one rather than after launch.
Have a decision like this in front of you?
Bring one AI pilot or workflow to a free Reliability Review: 60–90 minutes and a 2-page note you keep.
Book a Reliability ReviewFree. You keep the findings.