Virtual Assistant That Actually Solves Issues
At a glance
CLIENT
A regional telecommunications provider
SERVICE
- AI strategy and design, ai software development services, conversational ai for customer service, machine learning consulting services, knowledge management, cloud based devops, custom software development services
INDUSTRY
- Telecom, Media & Entertainment / Telecommunications
A regional telecommunications provider operated a multi-channel contact center handling billing, tariffs, roaming and home internet support. Call volumes grew faster than revenue, average handling time was increasing, and customers waited too long for answers to simple questions such as balance checks or modem resets. Several earlier chatbot attempts had failed: rule-based scripts covered only a narrow set of scenarios and were quickly abandoned by both customers and agents.
LeanCoded partnered with the provider to design and build a new AI-powered virtual assistant that actually resolved customer issues. Over 12 months, we implemented a conversational AI platform for web and mobile, added real-time guidance for agents, and integrated analytics that continuously improved the knowledge base. Within six months of full rollout, the assistant was resolving 52% of eligible requests end-to-end, total call volume dropped by 38%, and average handle time in the remaining calls fell by 24%.
When “Chatbot” Becomes a Four-Letter Word
Customer surveys and call recordings showed a clear pattern: previous chatbots were slow, rigid, and often redirected customers to phone or email after several unhelpful steps. Agents did not trust the suggested answers and preferred to search manually through outdated knowledge articles. The contact center processed more than 1.5 million interactions per year, with over 60% related to a predictable set of topics: tariffs, roaming, SIM activation, and home internet troubleshooting. Yet less than 5% of these interactions were resolved in digital self-service.
LeanCoded started with a diagnostic phase, combining interaction analytics and targeted interviews. We clustered intents, mapped volumes per topic and channel, and identified where automation would deliver the highest impact without harming customer experience. Instead of trying to automate every use case at once, we focused on a core set of high-volume scenarios and designed the assistant and agent tools around them, using our experience in conversational AI for customer service and machine learning consulting services.
Designing a Virtual Assistant People Actually Want to Use
The new solution had three pillars: a customer-facing virtual assistant, an in-agent assistant, and an analytics layer. For customers, we designed natural, multi-turn dialogs that handled authentication, balance and usage questions, tariff changes, roaming configuration, and guided troubleshooting for common modem issues. For agents, we embedded an AI-powered side panel in the existing desktop: when a call or chat started, it suggested the most likely intents, pulled relevant knowledge snippets, and proposed next best actions.
Behind the scenes, LeanCoded implemented a cloud-native AI stack with intent recognition, entity extraction, policy orchestration and integration into billing, CRM and network systems. Using AI software development services and custom software development services, we ensured that both the virtual assistant and agent tools could execute end-to-end actions—like changing a tariff or resetting a device—rather than just answering FAQs.
From Proof of Concept to Production Workhorse
Previous chatbot initiatives had stalled after proof-of-concept because they were not built for operations. LeanCoded approached the rollout as a product, not a one-off project. We defined a roadmap with three releases: pilot on a limited set of intents and channels, expansion to more topics and full web/mobile integration, and finally deployment of the in-agent assistant. Each release had clear adoption and performance targets, tied to real operational metrics such as containment rate, call deflection and AHT.
To keep the assistant relevant, we put continuous improvement at the core of the operating model. The analytics layer captured every interaction—successful or not—and fed it into dashboards for product owners, data scientists and knowledge managers. This allowed the team to spot new intents, identify gaps in answers, and tune models regularly. Cloud based devops practices automated deployment, monitoring and rollback, so updates could be shipped weekly without disrupting service.
Intent and opportunity discovery
Conversational design and knowledge restructuring
AI and integration architecture
Agent assistant and next best actions
Analytics, feedback and training loops
How LeanCoded Made AI Part of Everyday Support
The telco had tried automation before. The difference this time was a practical focus on outcomes, tight integration with core systems, and an operating model that kept improving the assistant after launch.
Self-service that really closes tickets
The virtual assistant can authenticate customers, retrieve account data and apply changes, meaning many issues are resolved without any agent involvement.
- Agents supported, not replaced
The in-agent assistant provides real-time guidance and content, reducing handle time and error rates, especially for newer staff handling complex issues. - One knowledge backbone across channels
Both the virtual assistant and agents use the same restructured knowledge base, ensuring consistent answers whether the customer starts in chat, app or phone. - AI managed as a product
With clear KPIs, release cycles and ownership, the assistant evolves based on real interaction data instead of being a static project that fades after go-live.
Impact on Volumes, Costs and Experience
Six months after full rollout across web, mobile app and key call queues, the virtual assistant was handling an average of 52% of eligible service requests end-to-end in digital channels. Overall inbound call volume to the contact center decreased by 38%, as customers chose to resolve simple issues via self-service. For calls that still reached agents, the combination of better routing and real-time guidance reduced average handle time by 24%.
On the cost side, the provider achieved a 22% reduction in annual operating expenses related to first-line support, driven by lower call volumes and improved productivity. Customer satisfaction with digital support increased: CSAT for assistant-handled sessions was comparable to human-assisted cases, and overall NPS for support interactions rose by 12 points. By combining AI software development services, conversational AI for customer service, machine learning consulting services, cloud based devops and custom software development services, LeanCoded helped the regional telco turn a failing chatbot history into a sustainable AI-powered support model.