One screen for 'why is my bill higher?' · Vattenfall
Why is my bill higher? Answering it in one screen instead of seven.
A concept for Vattenfall's customer service agents. One workspace that explains a customer's annual settlement from their meter data and drafts a reply the agent edits and sends, with every fact linked to its source.
Self-initiated concept. Not affiliated with or endorsed by Vattenfall.
- Role
- Self-initiated concept, from interviews to tested prototype
- Company
- Vattenfall (concept)
- Timeline
- 2025 · 6 weeks, evenings
- Team
- Solo, with 7 customer service agents from Dutch energy suppliers
- Platforms
- Internal web app for customer service agents
1One workspace in place of seven separate tools
2The draft cites meter data and stays editable
3An estimated reading is flagged before anyone replies
9:40 → 6:10
Minutes to answer a billing email, in tests
7 → 1
Tools open during a billing question
6 of 7
Agents who would use it daily
0
Drafts that can be sent without an agent's edit
The problem
Every year, energy customers get an annual settlement, and many ask why it's higher than expected. The agents I interviewed answer that question with up to seven tools open, and write the explanation from scratch each time.
What I did
I interviewed seven customer service agents from Dutch energy suppliers, mapped the questions they get most, and designed a workspace in Vattenfall's visual identity: customer context, meter data explained in plain words, and AI drafts that cite their sources and are always edited and sent by the agent.
What changed
In moderated tests with six agents, answering a billing email took 6:10 minutes instead of 9:40, and six of seven said they would use it every day. The drafts' source links were the feature they trusted most.
01 · Context
Why I picked Vattenfall.
I spent five years on energy software at Schneider Electric, mostly for building operators. I was curious about the other side: the people who explain energy bills to households. Vattenfall serves about two million customers in the Netherlands, and annual settlement season brings a wave of questions.
This is a self-initiated concept. I don't work for Vattenfall and had no access to their systems. The research is with agents from several Dutch energy suppliers, and the design follows Vattenfall's public visual identity.
| Question type | Share of contacts | How long it takes today |
|---|---|---|
| Why is my bill higher? | About a quarter | The longest by far |
| Change my monthly amount | About a fifth | Short |
| Meter reading problems | About one in seven | Long, often a callback |
| Moving house | About one in ten | Short, mostly forms |
02 · Problem
Seven tools and a blank reply box.
The question I worked on
How might an agent get the full picture and a good first answer in one place, while staying responsible for what the customer receives?
03 · Research
Seven agents, three suppliers.
I interviewed seven agents and two team leads from three Dutch energy suppliers, recruited through LinkedIn. I also read Vattenfall's public help pages on the annual settlement and 400 public customer questions about bills on forums and review sites.
Most higher bills have three causes
A price change, a colder winter, or an estimated meter reading. Agents work out which one from scratch every time.
“It's almost always the same three things. I just have to prove which one.”
Estimated readings cause the angriest calls
Customers feel charged for nothing when a reading was estimated, and agents often spot the estimate late.
“Once you see it's an estimate, the whole call changes.”
Agents want help writing, and to stay in charge
Agents were open to AI drafts, but only if they could see why the draft said what it said, and change it.
“If it writes for me, I want to check its homework.”
04 · Process
Six weeks, tested with the people who'd use it.
Chatbot for customers
Bill questions are emotional and specific. In the forum posts, people wanted a person who understood their bill.
Templates library
Faster typing, but agents still had to find and interpret the data themselves.
Workspace with explained data and drafts
The agent sees the explanation and a draft built from the customer's own data, then edits and sends it.
05 · Design decisions
Context first, then a draft.
A summary of the customer before the first word
When a contact comes in, the workspace shows who the customer is, what changed on their contract and what they asked before, in three lines with links to the sources. Agents in the tests stopped asking customers to repeat themselves.
The bill explained from the meter data
The workspace compares usage with the previous year, marks estimated readings, and lists the likely causes of a higher bill in plain language. One click asks the customer for a meter photo, which settles most estimate disputes.
Drafts cite their sources and are always sent by a person
The AI drafts a reply using only the customer's data and published tariffs. Every fact in the draft is highlighted and links to where it came from. Nothing is sent automatically.
- 01QuestionEmail, chat or call notes
- 02ContextContract, meter data, tariffs
- 03DraftFacts linked to sources
- 04Agent editsRequired before sending
- 05SentEdits feed the quality review
No draft reaches a customer without this step
Team leads see where drafts go wrong
Agents can flag a draft as wrong with one click. Team leads see how often drafts are used, how much agents change them, and every flagged draft. The numbers in this screen are illustrative.
06 · Results
What the tests showed.
Six agents answered the same three billing emails twice: once with their usual tools recreated as screenshots, once with the prototype.
“Half the work is done when I open the email. And I can see where every number comes from.”
“Seeing the estimated reading straight away would calm down so many calls.”
07 · Reflection
What I learned.
The principles were the same as in Atlas Copilot at Schneider: show the sources, keep a person in charge, and measure what people change. They worked just as well for billing emails as for building data.
Internal tools deserve the same care as customer products. Agents use them eight hours a day, and every saved click repeats thousands of times.
Tools on this project
Figma, FigJam, ProtoPie, Claude API for draft prototypes, Lookback
Names of the people quoted have been changed to protect their privacy.
Quoting a roof before the ladder comes out →