Professional Training / EdTech
AI Voice & Automation Platform for Counselling Operations
Two AI voice lines, a WhatsApp channel and a live counsellor co-pilot on one knowledge base, wired into the ERP.
- AI Voice Bot
- Agent Assist
- Auto-QA
- WhatsApp Bot
- ERP Integration
- MongoDB
100%
of calls QA-scored, with a live co-pilot on every call
24/7
inbound lead line, qualified leads logged to the ERP
1 core
knowledge base serving two voice lines and WhatsApp
4
monthly milestones from discovery to handover
- Full handover of source code, credentials and infrastructure at the final milestone, with a 60-day defect warranty.
- Client data is never used to train models, and the client owns the platform outright.
Coverage is the metric that matters in quality management — a sample tells you about the calls you sampled, and says nothing about the ones that went wrong.
Enquiry-led advisory operations still run on the telephone, and the demand does not follow business hours. Someone deciding between courses, policies or providers rings when the decision is in front of them, which is frequently outside the window in which anyone is available to answer — and an unanswered enquiry in a competitive category is usually a lost one.
What was said on those calls is also the organisation's least accessible record. It exists in the conversation and nowhere else, so quality management is conducted on a sample: a supervisor reviews a handful of calls per agent per month. Most of what an organisation says to its customers is therefore never observed by the organisation at all.
Why it is hard
The obvious responses both fail in predictable ways. Adding reviewers scales linearly with call volume and never catches up, and the coaching it produces arrives weeks after the conversation it refers to. Pushing callers to self-service deflects the straightforward contacts and leaves the difficult ones, which raises average handling time rather than lowering it.
The harder constraint is timing. Guidance is only useful while the conversation is still open, and an adviser mid-call cannot go and read a policy document. So the intelligence has to be present in the call, in under a second, and grounded in the same material the organisation would be willing to see quoted back to it — which is a different engineering problem from summarising the call afterwards.
What we built
Two voice lines run on a shared core: one taking inbound enquiries around the clock and placing follow-up calls, one answering customer and onboarding queries. Both are grounded in a single knowledge base, and both can hand a conversation to a person with its context intact, so the caller does not start again.
An intelligence layer sits across both lines and the WhatsApp channel. It prompts the counsellor while the call is still live, then scores every completed call against the organisation's own quality metrics rather than a generic rubric, and writes dispositions and outcomes back into the ERP so the record updates itself rather than waiting on someone to update it.
- Two voice lines on one shared knowledge base — inbound enquiries and customer care
- Context-preserving handoff to a person, so the caller does not repeat themselves
- A live co-pilot prompting the counsellor while the conversation is still open
- Every completed call scored against the organisation's own metrics, not a generic rubric
- WhatsApp on the same context, with outcomes written back into the ERP
Questions about this engagement.
Does this replace the counsellors?
No, and the co-pilot only makes sense if it does not. The voice lines absorb enquiry traffic and repetitive follow-up calling, which is the volume that was going unanswered; conversations that need judgement hand off to a person with the context already assembled. The live co-pilot exists specifically to make a human call go better, which presumes there is a human on it.
How is automated scoring kept consistent with how quality is already measured?
It runs against the metric set the organisation already uses, not a rubric supplied with the system. The purpose of automating quality review is not to introduce a new standard that nobody has agreed to — it is to apply the existing standard to every call instead of to a sample, which is the part human review cannot do at volume.
What stops a voice agent from saying something wrong?
Answers are grounded in a defined knowledge base rather than generated from general knowledge, and questions outside that scope are handed to a person instead of answered. Every conversation is retained, so anything the system said can be reviewed afterwards. That last part matters more than it sounds: a voice agent whose statements cannot be audited is not deployable in an advisory setting regardless of how well it performs.
What this was built on.
Tell us what you're trying to solve.
Describe the workflow, the data behind it, and what a good outcome looks like. We read every message and reply personally.