Their QA team read one grievance ticket in twenty and scored it carefully. The other nineteen were never examined by anyone. Audira now scores 840,000 conversations a month against the customer's own scorecard, at a rupee each. In the refund queue it found that a quarter of the value paid out should not have gone, and a further tenth was owed to a customer and withheld.
What changed
Coverage is an input, not a result, so it is not the headline here. What coverage bought is the headline, and it arrived in the refund queue within the first month.
What it cost
We publish the rates, so the arithmetic is short enough to do in your head.
| Workstream | Published rate | Share of volume | |
|---|---|---|---|
| 750,000+ grievance chat tickets | Re 1 per ticket evaluated | 90% of volume | |
| 3,200+ hours of fleet call audio | Rs 25 per hour transcribed | 10% of volume | |
| Blended across both | Re 1.00 per interaction | Not an average |
Not a blended average, which is what makes it credible. A chat ticket is a rupee because that is the rate. A fleet call is a rupee because it runs 2.4 minutes, and 2.4 minutes at twenty-five rupees an hour is a rupee.
Short calls are cheap to audit, which is the thing that will or will not be true of your floor. A team averaging eight-minute calls pays about three rupees thirty a call instead, and we would rather say that now than have you find it on the invoice. The figures above are at published list rates; at scale, list is where a conversation starts rather than where it finishes. Live order support is billed on the same two rates by channel mix and sits on top.
On coverage, precisely: total grievance chat volume runs near 80,000 tickets a day, of which about 25,000 a day are in scope for Audira today. Within that scope, coverage runs at 99 percent against 5 percent under manual sampling. The rest is headroom, not coverage, and we would rather you heard that from us.
Where it leaks
These are shares of refund value processed, not of tickets. A quarter going out wrongly is a margin number and a finance director recognises it on sight. The tenth wrongly withheld is the one nobody looks for, because a customer who was underpaid does not always complain, they just score you lower and order less. Both halves are invisible at a one-in-twenty sample, because no single ticket looks wrong on its own.
Surfaced, not recovered. Audira identifies the error and names the ticket. Whether it is clawed back, coached out or written off is the customer's process, and these figures deliberately do not claim otherwise.
What is actually running
They did not begin with all three. They began with one, and the other two followed once the first was obviously working, which is the sequence we would suggest to anyone starting.
Internal. Not agent to customer, but manager to field worker. Scored for tone, instruction clarity, escalation and conduct. No audit vendor in this market sells this.
Quality, refunds, order status. Scored against the customer's own scorecard across several SOPs at once. This is the queue the refund finding came out of.
The live queue where an order is already in trouble. Both channels of one conversation scored together rather than as two separate records.
The one to notice is the first. Fleet manager to delivery executive calls are not customer conversations at all. They are internal, manager to field worker, and no audit vendor in this market sells that. If you run a field force, a gig fleet, a collections team or a rider network, the conversations that decide your operating quality are the ones your own managers are having, and today nobody is listening to any of them.
Where the findings go
Most audit tools end at a dashboard, and the best of them end at a coaching note. This one ended somewhere else, and it took a month.
Tickets raised during a weather surge were failing differently from ordinary tickets. At a one-in-twenty sample this is statistically invisible; at ninety-nine percent it is obvious.
Instead of becoming a coaching note for one agent, the pattern became test content. Weather-surge scenarios were fed into Kalibr as customised assessment data.
A 90 percent threshold on that assessment now decides who may handle weather-surge tickets. Every competitor's loop closes into training, which fixes one agent. This one closed into who is allowed on the queue.
An audit finding became assessment content, and assessment content became a hiring gate, inside a single month. It needs both halves live on the same customer at the same time, which is why no competitor can produce an equivalent. That other half is Kalibr, and its own case study is here.
How it passed review
Customer anonymised pending approval. Figures are one calendar month, supplied by the customer. Competitor rates are list prices supplied by Singularium and used only for cost comparison at this volume.
Talk to us
The fastest way to know whether any of this transfers to your floor is to point it at your floor. Send us a sample, we run it, and you tell us where we got it wrong.