Customer support · Production

A1 Slovenia drives 113-point NPS surge and automates 53% of interactions with AI agent Lumi

The problem

A1 Slovenia's previous chatbot used static conversation flows, causing customers to drop off before transfers and losing context during escalations, making it difficult to handle complex or context-dependent inquiries at scale.

First attempt

The previous chatbot's static flows and lack of contextual continuity during escalations resulted in customer drop-off and a degraded handoff experience, with the system acting as a query filter rather than a resolution tool.

Workflow diagram · grounded in source

Customer query received

trigger

“handling a wide range of inquiries – from technical support to account management across mobile, internet, and TV services”

NLU interprets user intent

ai_action

“boost.ai's sophisticated Natural Language Understanding (NLU) allows Lumi to accurately interpret user intent and integrates seamlessly with A1's knowledge base”

Hybrid routing: generative vs predefined

routing

“boost.ai's hybrid AI model allows Lumi to differentiate when a generative response is appropriate and when a predefined, vetted answer is necessary. This ensures accuracy and compliance, a key factor in regulated industries.”

Dynamic response delivered

output

“dynamically selecting between predefined answers—crafted by A1's team within boost.ai's robust intent hierarchy system—and context-aware responses generated by a powerful large language model”

Escalation to human agent

human_review

“Even when escalation is needed, Lumi provides a better starting point, contributing to a 35-point tNPS lift compared to escalations from the previous chatbot.”

Reported outcome

Since Lumi's March 2024 launch, A1 Slovenia achieved a 113-point tNPS increase for Lumi-only interactions, a 35-point tNPS improvement on escalated interactions over the prior chatbot, Lumi managing 53.5% of all interactions, and generative AI queries being 70% less likely to require human escalation.

Reported metrics

transactional NPS — Lumi-only interactions113-point increase

tNPS for escalated interactions vs prior chatbot35-point increase

share of customer interactions handled by Lumi53.3%

reduction in escalation likelihood for generative AI queries70% less likely

Show all 8 reported metrics

transactional NPS — Lumi-only interactions113-point increase

tNPS for escalated interactions vs prior chatbot35-point increase

share of customer interactions handled by Lumi53.3%

reduction in escalation likelihood for generative AI queries70% less likely

tNPS absolute score shiftfrom approximately -53 to 60

all interactions managed by Lumi53.5%

generative intent capacity vs traditional intentsroughly 12-15 traditional intents

topics trained with generative AIover 30 topics

Reported stack

boost.aiGenerative ActionNLUlarge language model2mobile

Source

https://www.boost.ai/case-studies/a1-slovenia-drives-major-customer-happiness-increase-with-generative-ai

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Since Lumi's March 2024 launch, A1 Slovenia achieved a 113-point tNPS increase for Lumi-only interactions, a 35-point tNPS improvement on escalated interactions over the prior chatbot, Lumi managing 53.5% of all inter…

What tools did this team use?

boost.ai, Generative Action, NLU, large language model, 2mobile.

What results were reported?

transactional NPS — Lumi-only interactions: 113-point increase; tNPS for escalated interactions vs prior chatbot: 35-point increase; share of customer interactions handled by Lumi: 53.3%; reduction in escalation likelihood for generative AI queries: 70% less likely (source-reported, not independently verified).

What failed first in this deployment?

How is this customer support AI workflow structured?

Customer query received → NLU interprets user intent → Hybrid routing: generative vs predefined → Dynamic response delivered → Escalation to human agent.