Maika prices every transporter's messages and minutes, logs each token at the published rate, follows every customer through the booking funnel, and splits the margin 50/50 after cost.
People only approve — everything else runs on schedule.
Live figures from the MavenTech WhatsApp bot. Voice AI and KSRTC join from their first billed month.
Each runs on a schedule against the same numbers, so the operator's report, Maven's invoice and the settlement can never drift apart.
Prices each operator's messages and minutes, drafts Maven's invoice to the operator for approval, and raises the settlement invoice to Maven.
Logs every LLM token, server, telephony and speech cost per operator and product, at published rates, reconciled to the actual bills.
Tracks every session and call through the booking funnel, month on month, and shows where customers stop.
Splits margin after cost 50/50 and shows what each partner receives, pays and nets.
Tracks each transporter's capacity, booking sources and AI seat share, its level on the five-step ladder and its next actions.
Their report and invoice are built from a data view that does not contain cost, margin or settlement fields at all.
We carry the cost up front, so Maven pays it back plus half the margin and keeps the other half. Both partners pay for cost, so both can check every line of it.
Invoice to Maven = C + (R − C)/2 = (R + C)/2| Revenue Maven bills the operator (R) | ₹100 |
| Our cost to serve (C) | ₹20 |
| Margin after cost | ₹80 |
| Each partner's half | ₹40 |
| Invoice to Maven | ₹60 |
| Maven keeps | ₹40 |
| We keep after paying cost | ₹40 |
There are no loss months. If cost ever exceeds revenue for an operator, the engine treats it as a data error and holds that line for review rather than invoicing it.
Published rates explain each operator's cost; the provider's bill proves the total. The gap between them is its own reconciliation line, so nothing is hidden in either direction.
| LLM tokens | Both | Input, output and cached tokens per call, by model | Direct |
| Speech-to-text | Voice | Audio minutes per call, from the provider | Direct |
| Text-to-speech | Voice | Characters or minutes per call | Direct |
| Telephony | Voice | Connected minutes, from the telephony bill | Direct |
| Servers and hosting | Both | Monthly cloud bill, Maika's agreed share | By token share |
| WhatsApp messaging | — | Maven's own vendor cost, no margin for us | excluded |
(Tin·rin + Tout·rout + Tcached·rcached) × FXEvery call priced at its model's published rate, converted at the FBIL reference rate — today ₹88 to the dollar. Cached input bills at a fraction of fresh input, which is why the cache rate moves the bill more than the token count does.
Operators judge the AI on cost per booking, so the funnel and the unit economics sit on the same page. A drop at one stage shows even when total bookings rise.
Every transporter sits on one of five levels, and a level counts only when every item on it has verified evidence. The platform always shows what to do next.
Evidence: Template approved and sending
Evidence: Screenshots, offer start date
Evidence: Dated depot photos, ticket sample
Evidence: Depot photos, buses stickered
Evidence: Depots and buses, out of the total
Up to three per transporter each week, ranked by expected extra bookings. Each has an owner, a due date and a status — and the platform records AI booking share before and after, so the playbook learns which actions work.
One record and one owner RM per transporter, prospect to full adoption. Stages 1–5 move on uploaded evidence; 6 onwards move themselves from live data.
Scored monthly: outcome 40%, adoption progress 25%, mission adherence 20%, sales journey 15%. Outcome uses Maven's measured bookings only — never transporter-reported figures.
Each to a named audience, on a fixed cadence. The operator's report is built so it can be forwarded unchanged.
| Output | Audience | Cadence | Carries cost or margin? |
|---|---|---|---|
| Maika Success and Billing Report | Operator, through Maven | Monthly, one per operator | safe to forward |
| Maven's invoice to each operator | Operator, issued by Maven | Monthly | safe to forward |
| JV Cost and Margin Report | PMB and Maven, never operators | Monthly | partners only |
| Settlement invoice to Maven | Maven finance | Monthly, one invoice | partners only |
| Adoption tracker and RM view | Maven RMs, Adoption Head, PMB | Live, refreshed daily | safe to forward |
| AI trend insights | RMs, Adoption Head, PMB | Weekly | safe to forward |
| RM performance tracker | Adoption Head, Maven admin, PMB | Monthly score, live progress | safe to forward |
Usage freezes; every operator is checked for a price.
Reports, Maven's invoices, the JV cost report and the settlement invoice, with flags.
Each RM's score, rating and rank; scorecards sent.
Flags cleared, the settlement invoice approved.
Reports out. An operator with an open flag waits until it clears.
Both enforced on the server. Fourteen design decisions keep the numbers trusted and the data safe — these are the four that shape every screen.
Their report and invoice are built from a data view that does not contain cost, margin or settlement fields.
Screens, totals, search, exports, the API and scheduled email. A scoped user's totals add up only their operators.
Hashed in analytics, last four digits on screens. Full numbers appear only on the payment recovery list.
An invoice cannot be approved by whoever last edited that operator's price or discount that month.
WhatsApp AI bills per message, Voice AI per connected minute. A product that is not connected yet says so — it never shows a zero, because a zero reads as “nobody used it” when it means “we are not reading it”.
Twelve transporters answering customers and taking bookings on WhatsApp. Usage, tokens and bookings read straight from the bot database.
Joins from its first billed month. Only connected talk time bills; unanswered, busy and failed calls never do.
One more Maika operator on the same settlement. Usage will be read over an SSH tunnel to KSRTC's own server.
Usage, tokens at published rates, the funnel and the 50/50 split — per operator, per product, per month.