AI Product Design / IC Craft
Direct Debit Explainer
"One of our biggest contact reasons - why has my bill gone up? The goal was to reduce Direct Debit cancellations and be transparent with our customers. I shaped what that looked like across customer and advisor surfaces simultaneously."
My Role
Staff Designer on OVO's customer-facing AI squad. Responsible for both surfaces — customer and advisor — from problem framing through to shipped product. Led research across both user groups and used code-based prototyping to explore conversational UI at pace.
The Problem
Three users.
Three problems.
The DD explainer wasn't a single user problem. Business, customer and advisor each had a distinct version of the same breakdown — and the design had to address all three simultaneously.
Business
30% of all inbound contact was billing-related. Advisors were explaining balance inconsistently across thousands of calls. Target: cut contacts by 145k/year and prevent 1% of DD cancellations — worth £4m.
Customer
Direct debit is 'set and forget' — until it changes. Customers have no visibility into why. Generic explanations don't cut it. They want their specific data.
"I feel like it's a bit of a slap in the face... after I've been telling my family to turn lights off."
Advisor
No single tool. Advisors juggled Salesforce, Kaluza and Google Sheets on live calls — high mental load, inconsistent output, no way to scale quality.
"It typically takes me 5–10 minutes from notes and history to find what I need to tell the customer."
The Hypothesis
Before anthing
was designed.
"Because customers lack credible, in-app explanations of why their balance changes — a key driver of DD cancellations — we believe an AI agent will prevent 1% of cancellations among digitally active customers, avoiding £2m in bad debt."
Research
Two tracks,
simultaneously.
Because this was a dual-surface product, discovery ran in parallel — customer needs and AI sentiment on one side, advisor usability on the other.
Customer Research
DD calculations are opaque
Customers accept prices go up. They rarely understand how OVO arrived at their specific new figure — and that gap erodes trust.
Generic isn't good enough
They want granular, personalised proof — charts, percentages, before/after. Their bill, not a template.
AI as analyst, not chatbot
Open to AI — but only if it answers specific questions and never blocks access to a human.
"With the power of AI, I would like a tool that explains bills automatically for me."
Advisor Research
Usability study with 6 advisors — evaluating entry point, content quality and perceived utility on live calls.
4.4/5
Usefulness
5–10m
Saved per call
83.3%
Positive feedback
"This is massively beneficial... it goes into the customer's usage rather than 'oh Ofgem announced a price increase.' It saves me time, saves the customer time."
Advisor — usability study
Design Approach
A phased roadmap,
not a single feature.
Starting simple, building confidence in AI incrementally, expanding capability based on what each phase taught us.
Explain
Single-turn. Personalised balance explanation via modal. No actions — just clarity.
Explain + Recommend
Explanation plus next best action. Phase 2.1: the AI agent carries out the action directly for the customer.
Multi-turn
AI embedded in the core journey. Customers interrogate their account conversationally and ask follow-up questions.
Key Decisions
Calls worth naming.
01 — Content format
Redesigning after advisor testing
All 6 advisor sessions surfaced the same problem: block paragraphs were too dense for live calls. One participant with dyspraxia couldn't track across lines at all. Advisors were skipping the summary entirely.
I pushed back on the product preference for comprehensive text — replacing paragraphs with bulleted summaries and adding an affirming opening statement.
02 — The entry point
Proactive, not reactive
Instead of expecting customers to hunt for an explanation, we designed a bottom sheet that triggers immediately on app launch — the AI agent has already answered "Why has my DD changed?" before the customer even has to ask.
1 in 5
customers visit app within 31 days
48%
visit within 24 hours
3%
cancel DD after visiting the app
24%
of those cancel within 24 hours
03 — Engineering collaboration
Designing within AI constraints
The LLM returns up to 19 possible reasons why a DD has changed. Working with engineering, we defined the rules that shaped the whole information architecture.
Max 7 cards
— prevent information overload
Min 1 card
— no blank or unhelpful responses
Cards expand on tap
— progressive disclosure builds trust
Data bolded
— personalised figures stand out
The Solution
Both surfaces.
The customer-facing AI conversation and the advisor balance agent — designed simultaneously, with different contexts, constraints and needs.
Customer surface
Advisor surface
Outcome
Key OKRs.
Primary metric
−0%
Reduction in direct debit cancellation rate
£2.3m pa estimated impact
−0s
AHT per call
£190k pa
−0%
Complaint rate
£200k pa
+0.00pp
CSAT topbox
0.0/5
Advisor usefulness
One honest reflection
A single design for both surfaces would not have worked here. The advisor and customer have fundamentally different contexts, pressures and needs — one is on a live call under time pressure, the other is at home trying to understand their bill. The design had to be genuinely flexible depending on who was looking at it, not just reskinned. That tension shaped every decision on this project.