Client
Goals2Life
Timeline
March to August 2026
Team
4 designers
Role
UX Researcher & Interaction Designer
Our client asked for an AI coach inside her goal-setting app. Our team of four spent two quarters working out which parts AI should handle and which should stay human.
My role: I synthesized our interview research, mapped the app’s flows to choose where AI could help, and ran the co-design workshop with our client.
Scope of Work
Overview
The brief asked for an AI coach.
Goals2Life is an early-stage goal-setting app built on one idea: people follow through when someone keeps them honest. You set a goal, track it, and check in with an accountability partner.
Our client wanted to add an AI coach that would step in when people started doubting themselves or drifting off.
Opportunity
Where AI helps, and where it shouldn’t
01
People didn’t want AI acting like a coach
People wanted trust, empathy, and judgment to come from a person. AI made sense for the work between sessions: getting ready, looking back, and keeping going.
Design decision: Keep the coach human. Use AI between sessions.
02
The hardest moment came right after taking action
After finishing a task, people often didn’t know what to do next. None of the competitors we looked at helped with that moment. People needed a quick way to look back and pick a next step.
Design decision: Help people reflect and decide what comes next.
03
Being useful mattered more than sounding human
People judged the AI on whether it helped. How human it sounded mattered much less. So we made its job obvious and kept people in control of it.
Design decision: Make AI useful enough to come back to.
– Interview participant
The pivot
Does AI actually need to be the coach?
The brief put AI in the coach’s seat. Our interviews showed people wanted to keep that relationship human and needed help between sessions instead. So we changed direction.
AI as the coach, replacing the relationship
Original project brief
AI around the coach and partner, as support
New project direction
How might we use AI to support accountability without replacing human connection?
Updated design statement
Solution
We gave AI a smaller job.
We focused on three moments between sessions that already existed in the app. The coach keeps the relationship, and AI helps in between.
01 / Goal Creation
Give people a draft to react to
AI writes a first draft of the game plan. People can change whatever they want, and nothing is locked in until they approve it.
02 / Goal Progress
Keep AI in the background
When people check in, a small nudge reminds them why they set the goal. Reflecting is optional.
03 / Meeting Prep
Nothing goes to your partner until you’ve seen it
Before an accountability meeting, AI pulls recent progress into a draft agenda. You can read it and change it before your partner sees anything.
Research
How we got to the pivot
Over nine weeks, we ran a literature review, a competitive analysis, and a flow analysis of the existing app.
We also interviewed thirteen people: coaches, people being coached, and our client.
AI for the logistics, people for the rest
Participants were fine with AI handling tracking, reminders, and reflection prompts. Trust, empathy, and judgment had to come from a person.
The less people noticed the AI, the more they trusted it
Participants trusted it most when it quietly helped with something they were already doing.
Narrowing the scope
Picking where AI could help most
I mapped four flows in the app to find where people got stuck and where AI could help. Goal Creation and Goal Progress had the clearest gaps, and Meeting Prep tied planning to accountability, so we kept those three and cut Calendar Integration.
In each one, we focused on a single problem.
IN SCOPE
Goal Creation
Goal definition
→ Game plan creation
Review & commitment
IN SCOPE
Goal Progress
→ Logging goals
Viewing analytics
Updating the plan
IN SCOPE
Meeting Prep
Creating an accountability partner
Scheduling & logging meetings
→ Preparing an agenda
CUT
Calendar Integration
Integrating calendar
Setting milestones
Adding new events
The arrow marks the one problem we picked in each flow.
Design principles
Six rules the AI had to follow
We wrote these early and checked every design against them with our client.
Help, don’t do it for them
AI drafts the plan, but people edit it and decide what’s final.
Let people set the rules of contact
People choose how often the AI reaches out and where.
Personalize without anthropomorphizing
Learn about the person without pretending to be one.
Be honest, not a cheerleader
Reflect progress as it is, without empty praise.
Only show up where it’s invited
Stay inside moments people already use, and stay quiet otherwise.
Push people toward real people
Every AI moment should lead back to the coach or partner.
Collaboration
Our client sketched with us
Instead of presenting to our client, I ran a Crazy 8s workshop where she sketched, discussed, and voted alongside the team.
Her input showed us where AI helped and where it got in the way of coaching.
Testing & Design Iterations
Putting it in front of people
We ran concept and usability tests together. Participants used each flow with no intro, then rated how useful it was and how much they trusted it.
Goal Creation
The AI was planning too early.
Goal Creation scored lowest on both usefulness and trust. People felt the AI was building a plan before it knew enough about them.
What changed: The AI now asks about the person and their situation before suggesting anything.
Goal Progress
This flow was already working.
People barely noticed the AI here, and they trusted it most. So we kept it in the background.
What stayed: Small reminders that tie progress back to the reason behind the goal.
Meeting Prep
People wanted to see the agenda before their partner did.
Participants wanted to read and approve the AI’s draft agenda before it reached their accountability partner.
What changed: We added a preview and approval step. It shows exactly what the AI pulled in and what your partner will see before anything gets shared.
Results
Did people trust the proposed AI layer?
Participants rated each flow on how useful it was, how much they trusted it, and how naturally the AI fit into Goals2Life. All scores are out of 5.
4.5 / 5
How naturally the AI fit into Goals2Life. This was the highest score in the study.
3.5 / 4.1
Goal Creation: useful / trust
4.3 / 4.4
Goal Progress: useful / trust
4.1 / 4.2
Meeting Prep: useful / trust
n = 8
Sample Size
The gap was personalization. Goal Creation suggestions felt generic because the AI didn’t ask enough up front.
People wanted the AI to know their experience, schedule, and skill level before it suggested anything.
What happened next: Our client kept these designs for her team’s next phase of development. We graduated before that phase, so we prepared the files for handoff.
Goals2Life needed support between sessions more than it needed an AI coach.
The AI was most useful when it helped people prepare, look back, and keep moving until their next conversation with a person.
Reflection
What I learned
Trust and usefulness weren’t the same thing
People trusted the AI most where they barely noticed it. Where it was most visible, it needed more personalization to feel useful. Only the testing data showed us that.
Privacy has to be visible in the interface
Some participants assumed meeting prep meant their meetings were recorded. The screens have to show what the AI sees, uses, and shares.
Principles kept the AI in its lane
Our client had more ideas than we could build. Checking each one against the principles made it clear what stayed.