Projects we have delivered
Advised, built, tested, shipped and operated:
Reference projects for banks, insurers and telecom providers.
Swiss telecom provider (anonymised)
Telecommunications
Automation & AI, software engineering, testing & release
Chat and voice
Telecom: AI service bot in production
The starting point:
- A large share of enquiries concerned recurring standard topics.
- At peak times, for example during outages, the load on customer service rose sharply.
- An earlier rule-based bot did not understand many requests.
- The bot had to access customer data without compromising data protection.
How we worked:
- Analysis of the most frequent requests and how they are resolved.
- AI engineering with a clearly defined scope and guardrails.
- Integration with customer systems through secured interfaces.
- Evaluation before every release, operations and improvement after go-live.
Project course:
- Analyse: evaluate frequent requests and existing processes.
- Architecture: bot, integrations, data protection and handovers.
- Implement: development, connectivity and evaluation with realistic dialogues.
- Operate: monitoring, analysis and step-by-step extension.
What we delivered:
- An AI bot for frequent requests in chat and on the phone.
- Access to relevant customer data through secured interfaces.
- An evaluation set of realistic dialogues as a quality check before every release.
- Handover to staff with full context, plus analytics on topics and resolution rate.
The result:
In production
Resolved by the bot
Checked before every release
By gridz
The challenge
The telecom provider received many enquiries on the same topics every day: invoices, contract details, outages, device settings. During outages the volume spiked.
An existing rule-based bot only helped to a limited extent. It failed to recognise many phrasings and often passed customers on to staff without context, who then had to start from scratch. What was needed was AI that works not only in a pilot but reliably in production.
How we worked
First we analysed which requests are most frequent and how they are resolved today. From that came a clearly delimited scope for the new bot.
The AI bot answers those requests on its own and accesses the relevant customer systems through secured interfaces. If a request becomes complex, or the customer wants to speak to a person, the bot hands the conversation to the service team – with a summary and everything already captured.
We made quality measurable: before every release the bot runs against a set of realistic conversation flows. Data protection and logging were part of the architecture from the start.
The result
- The bot runs in production and resolves standard requests directly, including outside office hours.
- The service team is noticeably relieved at peak times.
- Staff receive conversations with full context and can help straight away.
- Analytics show which topics are growing and where the bot can be improved.
gridz continues to operate the bot, reviews the conversations regularly and extends its scope step by step.
More on these services: automation & AI, software engineering and testing, quality & release.
