Econet says its Shona AI now understands 80% of customer queries
Econet has upgraded its Shona language AI model to version 2.0.
Econet has upgraded its Shona language AI model to version 2.0. The company says the new model now understands and answers 80 percent of customer queries put to its Yamurai chatbot. This marks a jump from the 70 percent level the system reached only a few weeks ago.
The improvement comes from Econet AI, the wholly owned subsidiary set up this year to push the group's artificial intelligence projects. Yamurai itself first appeared in 2020 during the COVID lockdown. It was built to sort out simple tasks such as PIN resets, airtime top-ups and bundle queries, handing anything complex to a human agent.
If you call customer service in Shona, the chatbot should now resolve four out of five questions without passing you on. That cuts waiting time on the line and reduces the number of calls that reach the 600 contact-centre staff who still work alongside the bots.
Econet claims Shona 2.0 is the first African language model to hit this mark on a live helpline. It now copes with the mixed Shona and English that most Zimbabweans use in daily speech. The company is also training the system on six Shona dialects so that, once complete, Yamurai will reply in the dialect the caller uses.
This upgrade builds on the mass AI adoption campaign Econet launched last month and the earlier steps it took to improve customer experience with artificial intelligence, as Finlede reported in October. The firm has also created the ecoAI Token, which lets subscribers buy AI access in the same way they buy data bundles.
Questions remain about exactly what the 80 percent figure measures. It is not clear whether it covers all calls, only Shona ones, or a controlled test set. Nor has Econet said whether the chatbot actually fixes problems such as a missing bundle or simply understands the question before routing the caller to a person.
Econet AI chief executive Navdeep Kapur acknowledged that early versions of Yamurai felt robotic. The company is now working on a Ndebele version as well.
The next test will be whether the system can handle real-world problems such as a disputed EcoCash transaction in mixed language or a specific dialect. Econet has not given a date for the full dialect rollout or for the Ndebele launch.
- Econet launches campaign for mass AI adoption
- Econet launches ecoAI Token, taking the Data Bundles concept to AI
- Econet launches upgraded Shona AI model, targets 80% query comprehension
- Zimbabwean billionaire Strive Masiyiwa’s Econet launches Shona 2.0 AI model
- How Econet plans to get more Zimbabweans using AI
- Econet Targets Mass AI Adoption in Zimbabwe With CAIMEX
- Econet steps up campaign for mass AI adoption
- Is the phone you’re buying stolen? GSM Zimbabwe launches a WhatsApp IMEI checker