Solving admin hell in AI Trust & Safety

The short version

Role: Product design and frontend lead. Client interviews, UX/UI and design system.
Problem: Trust & Safety teams wrote policies, then manually translated them into scattered rules and thresholds.
Decision: Put policies at the centre of the workflow, separating policy ownership from technical rule management.
Result: Checkstep gained 12 clients in the first six months, as reported in the original case study. This was a company outcome, alongside sales and conference activity.

A workflow built around the policy

Policies bring written guidance, thresholds, version history and insights together. Moderators get the context behind a decision; managers can change priorities without digging through every data type.

Checkstep policy management interface bringing policy guidance and rule thresholds together

A team handoff

Keep decisions aligned with policy.

Different roles share the workflow.

Trust & Safety
Set the policy
What should we enforce?
Put written guidance at the centre.
View the evidence
Policy + technical owners
Align the rules
How does guidance become a rule?
Separate policy thresholds from technical rule ownership.
View the evidence
Moderator
Review content
What should I act on, and why?
Keep policy context close to the review queue.
View the evidence
Manager + moderator
Track changes
What changed in the policy?
Make updates traceable with version history.

Tiny big detail

Shortcuts in plain sight.

The shortcut sits beside each policy, so moderators can learn the keys while they work.

Policy choices and their keyboard hints in the review screen.

01 / Policy workflow

Put the policy where decisions are made.

Checkstep moderation queues organising content for review

Checkstep moderator review interface showing content and decision controls

How the product was explained publicly

15-second recording from my archive, April 2023. This is the public marketing website, showing how analytics, moderator actions and policy conditions were presented. It is not a recording of the logged-in product.

Decisions, process and lessons

Find the mismatch

Client interviews showed that Trust & Safety teams reviewed policies roughly every two weeks. The product was organised around data types instead, which meant managers had to keep technical settings aligned manually.

Original Checkstep workflow organised around event and data types

Trust and Safety client interview findings informing the policy workflow

Separate responsibilities

Developers and data governors manage rules. Trust & Safety experts manage policy descriptions and thresholds. Policy updates notify moderators, while version history makes changes traceable.

What I would revisit

Small client requests could distract from the larger workflow. Another unresolved challenge was responding to sudden spikes in harmful content: creating a policy was quick, but reversing its previous actions was technically harder.

Role, tools & collaborators

Responsibilities

User interviews
UX/UI design
Design system
Frontend lead

Tools

Figma
Looker

Credits

Jonathan Manfield as CTO
Guillaume Bouchard as CEO

← Back to work