WhatsApp + USSD Self-Service at a Southern African Telco — Bridging the Smartphone-Feature-Phone Divide
ChatNext + Multi-Channel Agent + Voice Bot delivering unified self-service across WhatsApp, USSD and voice — covering the operator's full subscriber base.
MindMap OS · installed inside the customer's perimeter by forward deployed engineers · Private Cloud · billed on the result
The challenge
The operator — a major Southern African mobile network operator with approximately 38 million subscribers across multiple markets — had a structural self-service-coverage gap. The operator's WhatsApp self-service channel had been delivered to good effect for the smartphone-and-data subscriber segment, but a meaningful fraction of the subscriber base — particularly the lower-income and rural segments — remained primarily on feature-phones with USSD as the dominant non-voice interaction channel, and these segments were structurally excluded from the WhatsApp self-service uplift.
The USSD channel itself was operationally suboptimal. The USSD menu structure was static, deeply nested and updated rarely; the subscriber experience of navigating a 7-level deep menu structure to reach a specific service had been a long-standing dissatisfaction driver. The USSD volume was substantial — approximately 180 million USSD sessions per month across the operator's footprint — but the per-session productivity (the proportion of sessions that achieved the subscriber's intended outcome) was modest.
The constraints were structural. USSD as a channel imposes session-length and message-length constraints that conversational AI typically does not handle. The 11-language requirement applied across all channels equally — the operator's USSD subscribers spoke the same language mix as the operator's WhatsApp subscribers. The local-jurisdiction data-protection framework applied. And the operator's CCO had been emphatic that the new approach must not require the feature-phone subscriber base to upgrade to smartphones.
The approach
MindMap deployed a unified self-service platform composed of ChatNext (Cn) as the shared conversational intelligence layer across WhatsApp, USSD and voice; Multi-Channel Agent (Mh) for the channel-unification layer; Voice Bot (Vb) for the voice-channel coverage; and NLP Router (Nr) for the cross-channel intent routing.
Phase one was the USSD-conversational reimagining. The previous static USSD menu was replaced with a dynamic-conversation USSD experience — the subscriber types their intent in free-text (within the USSD session-length constraint) and the system routes to the appropriate response, rather than navigating a static menu structure. The dynamic-conversation USSD uses the same underlying ChatNext intent engine as the WhatsApp channel, ensuring consistency across channels.
Phase two was the language-coverage build. The 11-language requirement was met through a shared multilingual intent classifier and language-specific response generation. The most-used languages (English, Zulu, Xhosa, Afrikaans, Tswana, Sotho, Portuguese for the cross-border markets, Shona for the relevant markets) received the deepest model tuning; the longer-tail languages received the level of tuning that the subscriber-volume in each language justified.
Phase three was the channel-unification build. A subscriber who starts a transaction on USSD and continues on WhatsApp (or vice versa) is recognised across channels with conversation context preserved. Specific use cases — e.g. a SIM-swap that begins on USSD (the subscriber's only available channel from the feature phone) and continues on WhatsApp from a borrowed smartphone for the document-upload step — are now seamless rather than fragmented across separate channels.
Phase four was the voice-channel integration. The Voice Bot handles the voice-channel self-service with the same intent engine and the same language coverage, ensuring full coverage across the operator's three primary subscriber-interaction channels.
The pre-built building blocks
Rather than commission a ground-up build, the engagement leaned on MindMap's pre-built accelerator library — production-tested components that compress what would otherwise be a six-to-nine-month build into weeks.
The architecture
The platform runs on the operator's private cloud with full local data-residency. The USSD-integration layer uses the operator's existing USSD-gateway infrastructure with a new ChatNext-integrated handler replacing the previous static-menu handler.
ChatNext's conversational engine uses fine-tuned models per channel — a smaller Mistral 7B for the USSD path (latency and session-length critical) and Llama 3.1 70B for the more complex WhatsApp interactions. The intent classification is shared across channels, ensuring consistency; the response generation is channel-appropriate (the USSD response respects the message-length constraint; the WhatsApp response can be richer).
The multilingual layer uses a multilingual base model with per-language fine-tuning on the operator's actual subscriber-conversation corpus. The model handles the language-detection-and-switching that is common in the operator's customer base (a subscriber might initiate in English and switch to Zulu mid-conversation).
Multi-Channel Agent maintains the unified subscriber-interaction state across USSD, WhatsApp and voice. The cross-channel handoff preserves the conversation context, with the state stored against the subscriber's MSISDN for identity continuity.
Voice Bot's voice-channel integration uses the operator's existing telephony infrastructure with the speech-to-text and text-to-speech running in the operator's tenant for the local-language coverage.
Integration with the operator's BSS stack uses the operator's existing BSS-integration framework, with the platform handling the cross-system orchestration the various subscriber-interaction intents require.
The numbers behind the story
Inbound contact-centre deflection rate is 52% across the subscriber base, with the deflection roughly evenly distributed across the WhatsApp, USSD and voice self-service channels. The USSD-channel deflection — the segment that the previous WhatsApp-only self-service had structurally excluded — has been the largest source of incremental deflection volume.
USSD-session productivity (the proportion of sessions that achieve the subscriber's intent) has risen materially over the previous static-menu baseline. The free-text USSD experience eliminates the menu-navigation friction that had driven much of the previous session-abandonment.
Coverage of the feature-phone subscriber segments has been the strategic outcome. The previous WhatsApp-only self-service had inadvertently widened the operator's customer-experience gap between the smartphone-data subscribers and the feature-phone subscribers; the new approach has closed that gap and provided self-service uplift across the full subscriber base.
Cross-channel customer-experience has improved as well. The SIM-swap flow (initiated on USSD, completed on WhatsApp) is the most-cited example of cross-channel uplift — subscribers can complete the full flow without channel-restart even though no single channel could handle the full flow on its own.
An unexpected outcome: the dynamic-conversation USSD has become a competitive asset for the operator beyond the self-service use case. The operator's enterprise-services arm has begun offering USSD-conversational integration to enterprise customers (banks, retailers, government services) that want to reach the feature-phone subscriber base, with the licensing revenue contributing to the platform's business case.
“Our WhatsApp self-service had served the smartphone-data segment but had widened the gap with the feature-phone segments that needed self-service uplift most. MindMap delivered unified self-service across WhatsApp, USSD and voice with shared intelligence and shared language coverage, deflecting fifty-two per cent of contact-centre volume and closing the self-service gap across our full subscriber base.”— Chief Customer Officer· Southern African Telco
Why MindMap was chosen
The operator had evaluated several self-service platforms and concluded that none of them genuinely handled the USSD-channel uplift. The leading vendors either treated USSD as a legacy channel not worth investment, or proposed USSD-specific approaches that did not share intelligence with the WhatsApp and voice channels.
MindMap's accelerator-composition approach — bringing ChatNext, Multi-Channel Agent, Voice Bot and NLP Router together into a unified self-service platform with first-class USSD support and shared intelligence across all channels — was structurally unique. We could demonstrate the unified approach at a peer African operator.
Our embedded African-telecom expertise on the delivery team (two former subscriber-experience heads from peer African operators) was the third factor. The operator's CCO felt that the team understood the operational and customer-experience reality of African telecom — including the structural importance of USSD as a channel — in a way the global vendors did not.
Get the MindMap capability deck
The platform, the accelerator library and how an engagement runs, in one deck. Our team emails it within one business day.
Related deployments
Telecom WhatsApp Self-Service
ChatNext deployed across SIM, billing, and bundle management — bilingual and integrated with the carrier's BSS stack — deflecting 44% of inbound contact.
AI Voice Agent for Collections
An Arabic-English voice agent replaced the outbound collections dialler, reducing cost per contact by 58% while increasing same-call promise-to-pay rate.
Voice-First African Self-Service
Voice Bot + ChatNext reached 22M subscribers in 6 local African languages — 48% of voice-based customer interaction now AI-handled.
Want an outcome like this?
Start with a 2-week AI Readiness Sprint. We deliver a prioritised use-case backlog and business case grounded in what's actually buildable with our accelerator library.