Customer service automation on WhatsApp, web and voice
At an African mobile operator, 30 request types made up 78% of contact-centre volume. Letting customers complete them on WhatsApp, and handing the rest to agents with the conversation so far, cut inbound contact 44%. A pan-African bank deflects 67% of inbound contact on WhatsApp, and an Indian private bank cut handle time on assisted calls 38%.
What is customer service automation?
Customer service automation lets customers resolve routine requests themselves on WhatsApp, the web or the phone. A conversational AI understands the request in the customer's language, checks their identity when the request needs it and completes it in the systems behind it, such as core banking or billing. Whatever it cannot resolve reaches an agent with the conversation so far, and agent-assist shortens the calls people still take.
- A pan-African bank's WhatsApp channel deflects 67% of inbound contact in English and Swahili, against a target of 50%.
- An African mobile operator cut inbound contact 44%, and peak hold time fell from 23 minutes to 4 as customer activity grew.
- At an Indian private bank, agent-assist cut handle time 38% on assisted calls, and a voice agent handles 31% of inbound volume end to end.
- A UK energy utility resolves 78% of billing disputes the same day, and dispute-related complaints fell 84%.
Customer service, step by step
Map what customers actually ask
Start from real transcripts. At the pan-African bank, 3 months of chat and call transcripts produced 412 intents: 31 that could be fully self-served, 7 transactional ones needing step-up authentication, and the rest informational or for escalation. The classifier was trained on 280,000 historical messages in the mix of English and Swahili that customers use.
Connect the systems behind each request
Each intent is built end to end, so the bot completes the request instead of explaining it. At the African mobile operator it buys bundles, tops up airtime, moves mobile money and starts SIM swaps through audited connections to the billing, SIM and mobile-money systems. Money moves and SIM changes need a one-time passcode first.
Answer on WhatsApp, web and voice
ChatNext serves every channel from one intent model and knowledge base. At the pan-African bank, 95% of WhatsApp replies take 1.4 seconds or less. The Indian private bank's voice agent handles balance, statement and card-block calls in Hindi, English or Tamil. The UK utility runs the same dispute conversation on its website, app and WhatsApp.
Hand over with the context
Requests outside what the bot can do pass to a live agent with the full conversation, so the agent does not start from scratch. At the UK utility, disputes that need specialist judgement, involve several parties or show signs of a vulnerable customer go to a person with the case context attached.
Assist agents on the remaining calls
During live calls, the platform transcribes the conversation, identifies the intent and puts the relevant policy and a suggested answer on screen, with one-click actions that pre-fill the agent's workstation. Afterwards it drafts the summary, disposition and compliance-script check for one-click confirmation. At the Indian private bank, handle time on agent-assisted calls fell 38%.
People keep the conversations that need judgement: complex cases, specialist disputes and vulnerable customers. At the pan-African bank, the contact centre went from 110% of capacity to 71%, and the freed time went to outbound campaigns and complex cases. At the UK utility, people moved to complex disputes, vulnerable customers and preventing disputes in the first place.
Results from 4 deployments
Every figure below comes from the case study it links to.
Pan-African Tier-1 Bank
Sovereign WhatsApp Banking
- 6M+ customers served
- 2 languages (English + Swahili)
- 100% on-premises, no cloud calls
African Mobile Network Operator
Telecom WhatsApp Self-Service
- 2.3M monthly conversations
- 12s median resolution time
Indian Private Bank
Contact Centre AI
- 31% voice-AI deflection
- 3 languages (HI + EN + TA)
UK Energy Utility
Utility Billing Disputes
- 84% reduction in dispute-related complaints
- 5.2M customer accounts in scope
The accelerators behind it
Pre-built accelerators do the work, configured to your documents, rules and systems. Delivery took 16 to 26 weeks in the case studies above.
ChatNext →
Cutting-edge NLP chatbot with low-code studio, deployed across Africa, UAE and UK banks.
Voice Bot →
Telephony voice agent across IVR and outbound campaigns.
NLP Router →
Routes intents across human + AI agents based on confidence.
OnboardX →
Identity verification + onboarding for regulated industries.
Sentiment Analyzer →
Tone-aware ticket triage for support teams.
Contact centre ROI calculator
Enter your contact volumes, handle times and costs to see the three-year benefit of AI agents taking the routine calls, then email yourself the PDF.
Customer service: the questions buyers ask
How much contact can self-service take off the contact centre?
At a pan-African bank, WhatsApp self-service deflects 67% of inbound contact, against a 50% target; 73% of card blocks are self-served, averaging 35 seconds against 8 minutes on the old phone menu. An African mobile operator cut contact-centre volume 44%, and 81% of its bundle purchases now happen on WhatsApp. An Indian private bank's voice agent handles 31% of inbound volume.
How does it cut handle time on the calls agents still take?
At the Indian private bank, handle time on agent-assisted calls fell 38%. Agents no longer search for policy answers mid-call, because the platform puts them on screen; one-click actions replace switching between applications; and the call summary is drafted automatically, work that took roughly 4 minutes per call before. Training time for new agents dropped by roughly 40%.
What happens when the bot can't help?
It hands over with the context. At the African mobile operator, live agents receive the full conversation, so they don't start from scratch. At the UK utility, disputes needing specialist judgement, involving several parties or showing signs of a vulnerable customer go to a person with the full case context. At the pan-African bank, if core banking takes over 1.8 seconds, the bot sends a contextual escalation message.
Can it cope with customers who switch languages mid-conversation?
Yes, when it is trained on your own conversations. The pan-African bank's classifier learned from 280,000 historical messages in English, Swahili and regional blends. The African mobile operator's model, fine-tuned on 3 years of de-identified transcripts, handles urban street slang and replies in the customer's own mix. The Indian private bank's voice agent works in Hindi, English or Tamil, detected at the start of the call.
Do we have to replace our contact-centre platform or core systems?
No. The Indian private bank's existing contact-centre platform and routing stayed; the AI connects through standard telephony integration. The pan-African bank connected to core banking through its existing service bus, plus its card, loan and contact-centre systems. The UK utility's billing and CRM systems, mid-migration, were connected through their standard APIs, with the platform upstream of them.
Where does customer data stay?
It can stay entirely inside your network. The pan-African bank runs everything in its own data centres with no calls to any external model provider; WhatsApp messages never leave its network, and its regulator confirmed zero data egress before go-live. At the Indian private bank no call audio leaves the bank, as regulation requires processing in India. The UK utility runs in its own UK cloud tenant.
How long does a deployment take?
Delivery took 16 weeks at the pan-African bank: 6 weeks mapping intents from 3 months of chat and call transcripts, 6 building integrations, then 4 weeks of rollout from 5% of customers to bank-wide. The African mobile operator took 18 weeks; the Indian private bank and the UK utility 26 weeks each.
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.
More use cases
KYC and account opening →
4 hrs onboarding at a UK challenger bank, down from 5 days
AML and fraud detection →
71% fewer AML alerts to analysts at an Indian public-sector bank
Regulatory reporting and compliance →
7 days regulatory close at an ASEAN bank, down from 28 days
See it running on your own process
A 20-minute walkthrough with the engineers who build these deployments: your documents, your systems, the accelerators running.