Angelica Gray Private
Protected case study

Need access? Contact me on LinkedIn

Incorrect password — please try again.
← Back to all work
Sinch Sinch · Messaging Compliance

Compliance Center & ACER

Turning a manual, expert-only compliance process into a self-serve product — and an AI review layer that helps customers earn approval before they ever hit submit.

ACER Review
Ready to submit
Approval likelihood
82%Likely to approve
Add opt-in consent language
Apply
Clarify your business use case
✓ Applied
My role
Lead Product Designer
Strategy → IA → flows → UI → AI interaction
Team
PM · Engineering
Compliance specialists · Carrier relations
Platform
Sinch Dashboard
Responsive web · B2B SaaS
Focus
Self-serve + AI
Trust · clarity · approval readiness
01 Overview

Messaging compliance is the gate every business has to pass before it can text its customers. At Sinch, getting through that gate meant relying on internal experts — slow, opaque, and impossible to scale.

Compliance Center is a self-serve experience inside the Sinch dashboard that lets customers understand what they need to do, complete verification steps, and improve their odds of approval — without waiting on a human. ACER is the AI-assisted review layer inside it: it flags issues before submission, predicts likelihood of approval, and turns vague rejection risk into specific, fixable recommendations.

My job was to make a genuinely hard, regulated domain feel learnable for people who are not compliance experts — and to design AI guidance they'd actually trust.

02 The challenge & why it mattered

A white-glove process that couldn't scale.

Every customer who wanted to send messages had to clear carrier and regulatory requirements. In practice, that meant a Sinch specialist walked them through it by hand — interpreting rules, reviewing submissions, and absorbing the back-and-forth with carriers. It worked, but it was expensive, inconsistent, and a bottleneck on growth.

Customers were frustrated too. They didn't know what was required, why they were rejected, or what to do next. Compliance felt like a black box they were failing for reasons no one would explain.

From manual service to self-serve product

The real shift wasn't redesigning a form. It was changing who holds the knowledge — moving expertise out of specialists' heads and into an interface customers could navigate themselves.

Before → After · the workflow
Before — Sinch-managed
Customer emails a requestDay 1
Specialist interprets the rules
Back-and-forth over emailfriction
Manual submission to carrier
Rejected — unclear whyDay 9+
After — Self-serve + ACER
Guided setup in-productDay 1
Plain-language requirements
ACER flags issues earlyfix first
Submit with confidence
Higher first-pass approvalDay 2–3
03 The product, explained simply

Three audiences, one impossible middle.

To text a customer in the US, a business has to satisfy rules set by carriers and industry bodies — proving who they are and that people opted in. The hard part of the design wasn't any single screen. It was that the experience had to serve three groups whose needs pull in different directions.

Users
Non-expert customers who just want to send messages — and have no patience for compliance jargon.
Carriers
The gatekeepers. Strict, evolving rules with real consequences for getting it wrong.
Sinch
The business — needs throughput, fewer support hours, and protected carrier relationships.
What made this complex
  • Rules change and vary by carrier, use case, and region.
  • The cost of a wrong answer is a real rejection, not a warning.
  • Users have no mental model for what "compliant" even means.
  • Expertise lived in people, not in any documented system.
04 The real design problem & goals

How do you make people feel confident about a decision they're not qualified to make — without pretending it's simpler than it is?

That framing set my goals. I wasn't optimizing a funnel — I was building the user's confidence and the system's trustworthiness at the same time. Four goals guided every decision:

1
Make the invisible visible
Show users exactly what's required, where they stand, and what to do next — at every step.
2
Catch problems before submission
Shift the moment of feedback from "rejected by a carrier" to "fixable, right now."
3
Earn trust in the AI
Make every recommendation explainable and reversible, so users stay in control.
4
Free the specialists
Reserve human expertise for genuinely hard cases — not routine hand-holding.
05 Discovery, alignment & mapping

First, I made the hidden process legible.

I interviewed compliance specialists to extract what they actually do, sat in on real submissions, and read carrier documentation until the patterns surfaced. Then I mapped the end-to-end system so the whole team — product, engineering, and compliance — could finally see the same picture and agree on where a self-serve product could safely take over.

That map became the backbone of the work: every node was either something we could guide a user through, automate, or escalate to a human.

System map · simplified
Business identity
Brand verification
Campaign & use case
ACER · AI review & readiness
Ready
Submit to carrier
Needs work
Apply fixes, re-check
Edge case
Escalate to specialist
06 Designing the self-serve Center

A setup guide that always answers "what now?"

I broke an intimidating process into a guided sequence with persistent progress. Users always know what's done, what's next, and why each step matters. Requirements are written in plain language, with the compliance reasoning available — but never in the way.

Compliance setup 3 of 4 complete
Tell us about your business
Legal entity, EIN, contact
Done
Verify your brand
Identity & trust score
Done
3
Build your campaign
Use case, sample messages, opt-in
Continue
4
Review with ACER & submit
Check approval readiness
Locked
How I reduced cognitive load

One decision per step. Progressive disclosure for the legal detail. Plain-language labels with expandable "why this matters." And a single source of truth for status — so users never had to hold the whole compliance model in their head to make the next move.

07 Designing ACER · AI-assisted review

AI that recommends, never overrules.

ACER reviews a submission the way a specialist would — but instantly, and in the open. It surfaces a single readiness signal, then breaks it into specific recommendations the user can act on. I used AI throughout the design process too: prototyping recommendation copy, stress-testing edge cases, and exploring how much explanation users actually needed before they'd trust a suggestion.

ACER Review
2 to fix
Approval likelihood
64%Needs attention
High impact
Sample message is missing opt-out language
Medium
Use case description is too vague for this carrier
Recommendation detail
High impact
Add opt-out language to your sample message
Why
Carriers require every promotional message to tell recipients how to stop. Without it, this campaign is very likely to be rejected.
Suggested fix
…reply STOP to unsubscribe at any time.
Apply fix Ignore
Designing AI that earns trust

Trust isn't a tone of voice. It's whether the user can see the reasoning, undo the action, and stay the decision-maker.

Explainable
Every flag shows the "why" and the carrier rule behind it.
Reversible
Apply and ignore are both one click — and both undoable.
Honest
Likelihood, not a guarantee. ACER never promises approval.
08 Key decisions, states & edge cases

The product lives in its states.

An AI review experience is only as good as how it handles disagreement, uncertainty, and "what if I'm wrong?" I designed for the full range — including the moments users push back.

Approval likelihood · three states
88%
Ready to submit
Encouraging, but never a guarantee.
64%
Fixable issues
Prioritized by impact, not volume.
31%
Likely rejection
Clear path: fix, or talk to a specialist.
When a user applies
Opt-out language added✓ Applied
Likelihood rose +14% · undo available
The fix is shown in context, the score updates live, and the change can always be reverted.
When a user ignores
Use-case wordingIgnored
Noted as a known risk · re-open anytime
Ignoring is respected, not punished — but ACER keeps the risk visible so the choice stays informed.
09 Testing strategy

Validating confidence, not just clicks.

The risk with AI guidance is that people either over-trust it or dismiss it. I built a research plan to test whether users understood the recommendations, felt in control, and made better-informed decisions — measuring comprehension and confidence, not only task completion.

Research plan · snapshot Moderated · unmoderated
We're testing whether…
  • users grasp what each recommendation means
  • the likelihood score builds or breaks trust
  • people feel in control of apply / ignore
Who
Non-expert customers across business sizes who have never completed compliance before, plus a few who've been rejected.
Method
Task-based usability sessions on the prototype, comprehension probes, and a confidence rating before vs. after ACER.
Success looks like
Users can explain a recommendation in their own words, and self-reported confidence rises without blind over-trust.
10 Outcomes & expected impact

What it moved.

+16pt
increase in first-pass approval readiness
37%
faster average time to approval
61%
reduction in manual review time
93%
more weekly verifications per specialist

Figures reflect projected and early-measured impact of the self-serve + ACER experience against the prior manual baseline.

11 Reflections

What I'd validate next.

Compliance Center proved that expert-only work can become a product — if the interface carries the knowledge and the AI stays honest about uncertainty. If I kept going, I'd watch three things closely.

Does confidence translate to outcomes?
Track whether higher self-reported confidence actually correlates with real approval rates over time.
When do people ignore good advice?
Study the ignore patterns to learn where ACER's reasoning isn't landing — and where the rules themselves need explaining.
How does trust hold as rules change?
Carrier requirements evolve — I'd monitor whether users keep trusting ACER when its guidance shifts beneath them.
Next case study · 02
AI-powered assignment suggestions — Anthology
Read next