Skip to content
AwakastLet's talk

Fortum, support assistant

Fortum's support agents answer customer questions more than 30% faster

We built an assistant that finds the right clause across contracts and policies and shows where every answer comes from. It runs inside Fortum's own environment.

Systems for a team

faster answers to customer questions
30%+
support agents set up to use the assistant
20+
from the first workshop to daily use
6 weeks

Where Fortum started

Fortum's support agents answer customer questions that need exact references to contracts and policies, and those documents keep changing. Looking up the right clause by hand across large sets of documents slowed answers down and made it slow for new agents to get up to speed. Any solution had to stay inside Fortum's private environment, with no data sent outside, and every answer had to be easy to check against its source.

What we built

We started with a one-day AI workshop for segment leaders to agree on use cases, risks and where the system would run. Then we built an assistant that answers from Fortum's own documents (RAG), with separate document sets and search paths for business and retail customers. It searches by keyword and by meaning, re-ranks the results and shows each answer with its citations and a confidence level. We connected it to the support team's workflow, ran a pilot and tuned it for daily use.

Deliverables

  • One-day AI workshop for segment leaders

  • Separate RAG pipelines for business and retail customers

  • Answers that lead with their citations and show a confidence level

  • Integration into the support team's workflow

How it works

  1. Separate paths for business and retail.

    Each customer segment has its own document set and its own search path, so answers come from the right contracts and policies.

  2. It finds the passage that matters.

    Search combines keywords and meaning, then re-ranks the results so the most relevant clause comes first.

  3. Every answer shows its source.

    The agent sees the citation and a confidence level, and can open the original document before replying.

  4. Nothing leaves Fortum.

    The assistant runs on Fortum's private infrastructure, with no external data transfer.

Results

faster query resolution
30%+
support agents set up to use it
20+
from workshop to daily use
6 weeks

  • Handling time 30% lower after rollout

  • Faster answers for front-line support

  • More confidence in answers grounded in policy documents

  • New support agents get up to speed faster

Fortum's internal policy and contract documents are not shown publicly.

6 weeks from workshop to daily use

  1. Week 1

    Workshop and scoping

    Agreed use cases, risks and where the system may run.

  2. Weeks 2–4

    Building the assistant

    Set up the customer segments, indexing and the answer flow.

  3. Week 5

    Pilot

    Connected the assistant to the support workflow and checked answer quality.

  4. Week 6

    Tuning

    Adjusted search and confidence thresholds for daily use.

Technology

  • Python
  • Azure AI Search
  • Microsoft Azure
  • RAG
  • FastAPI
  • .NET
  • Kubernetes

Show us the questions your team still answers by digging through documents.

Tell us what you're building