Uses AI
Finance · Key driver: Process
Content Velocity
When you call the bank, the specialist helping you reads from an internal article. Getting a new or updated article live took 10 to 12 weeks, so customers sometimes got yesterday's answer. We were asked to get it to two.
Organisation
JPMorganChase, New York
When
Jan – May 2024
My role
Lead service designer
Workshop
22 people · 40+ ideas · 6 concepts
12 → 2
weeks to get an article live: the goal we were set
4
root causes behind the delay, and only one was about writing
6
concepts shaped with the team, three of them using AI
01 · The problem
The article is what the customer actually hears.
Rules change, products change, and events like the pandemic change what customers need. Every change has to travel from the business, through writers and approvers, to the specialist's screen and the public FAQs. When that takes three months, specialists improvise, customers call back, and NPS takes the hit. The goal had two sides: shorter time to market for the people writing, and better answers for the people calling.
02 · Where the weeks went
The writing wasn't the slow part.
We mapped every role and step, from the request to the specialist's screen, and interviewed writers, schedulers, governance and business leads. Four causes kept feeding one cycle that kept time to market high.
What we found
Tools
Built for someone else's work.
Teams inherited tools from other parts of the firm and couldn't adapt them, so they worked around them.
“I need to go to four different places to do one impact.”
Process
Nobody could see the whole thing.
Writers chased requirements and versions, then double-checked everything because they couldn't see what happened next.
“We battle with it being outside of our control.”
Partners
Approvals were someone's side job.
Reviewers had other priorities, so analysts spent their time chasing replies.
“It's the biggest bottleneck we have in my experience.”
People
The extra effort didn't count.
Scorecards measured productivity and accuracy, not whether the content actually helped anyone.
“You can have numbers through the roof getting out procedures, but we don't see the other side.”
03 · What we designed
Six concepts, built with the people who'd use them.
In April 2024 we ran a co-creation workshop with 22 people from writing, scheduling, governance and the business: three focus areas, 12 themes and 40+ ideas, shaped into three concepts. A month later an AI work session added three more. Each concept came with how we'd measure it.
Workshop
Partnership 2.0
Training, shared dashboards and named contacts, so approvers see what a delay costs. Measure: time in approval, % meeting SLA by team.
Workshop
A workflow tool that fits
Views and fields set up by role, with live status, so nobody works around the tool. Measure: time to market.
Workshop
Streamlined feedback loop
Specialists tell writers directly when an article is wrong or unclear. Measure: rework cycles, handoffs, change requests.
AI
Anti-Word Word Club
One editing space instead of shuttling documents between tools, with AI pointing writers to what changed. Measure: corrections, time to approval.
AI
Gen-AI content creation
Summarises long requirements into plain language and drafts an article for a writer to edit. Measure: speed to market, average handle time.
AI
Eliminimail
Approvals and feedback happen in the tool, not email threads. Target: an article live within one two-week sprint.
Leaving it with the team
We handed over a playbook, How Might We ideation, concept mad-libs, a concept poster and a priority matrix, so the team could keep generating ideas without us.
What happened
We handed all six concepts to the product team, who took them into feasibility assessments.
What I'd take to public services
Sort out ownership and approvals before talking about AI. Otherwise you get a faster first draft and the same twelve weeks. Caseworkers rely on internal guidance in exactly the same way.
More projects
2024–25
Complex Customer Segmentation
JPMorganChase · Finance
About 5 million customers were tagged one way in one system and another way in the next. We designed a fix from the data up.
Artefact image, 4:3
2023–24
Lost Card Lifeline
JPMorganChase · Finance
Every month 800,000+ Chase customers tried to replace a lost card themselves, and around 40% gave up partway.
Artefact image, 4:3
Working on a regulated service?
I’m always glad to compare notes with people in finance, healthcare or the public sector, especially if you’re working out where AI belongs.