Rebuilding a marketing platform mistaken for a logistics tool
A legacy portal that read as logistics software — not marketing automation — was rebuilt on heuristic and usability data, with rage clicks and NPS both moving sharply in the right direction.
The problem
A marketing automation platform that the market kept reading as a logistics tool.
Enviou had shipped the same three products — cart recovery, boleto payment, and e-mail marketing — since 2018, without a consolidated design team to see a redesign through. Customers struggled just to access the Portal to configure their automations, and the interface worked against them at every step: a legacy visual style, broken interactions, even Portuguese-language copy errors scattered through the product.
Then the pandemic hit, and that legacy UI — paired with a brand presence that didn’t communicate the right category — did lasting damage: the company got read as a logistics business, when it was actually a marketing automation platform.
"We needed a design approach that consolidated complex information, aligned teams, and reflected both user needs and market expectations."
Felipe, CEO
Three earlier redesign attempts had already failed — not for lack of design effort, but because nothing was documented, so every new hire re-litigated the same debates. The interface problems were symptoms; the absence of a system was the cause.
My role
What I owned, and who I owned it with.
Scope
Lead Product Designer — sole designer for most of the project after an initial 3-person team rolled off mid-way. Owned discovery, interaction design, the design system, and cross-functional alignment with Engineering, Product, and the C-suite.
Deliverables
Launch of 3 new automations · brand redesign · Portal (core product) redesign · design system creation.
Goals
Reposition as a marketing automation company, not logistics-adjacent · implement design methodology into a company that had none · put users first, reducing dependency on support.
Discovery
With no budget for scheduled interviews, six weeks of stakeholder sessions and a 14-month ticket audit built the first hypotheses — put on hold for validation.
What the stakeholder sessions and artifact audit surfaced:
Repurchase: No easy way to trigger repurchase flows for recurring products — clients manually rebuilt the same automation.
Trust: Legacy visual style, broken interactions, and Portuguese copy errors undermined trust in clients’ own brands.
Editor: Email/SMS editor had no standardized templates — brands relied on manual workarounds for every campaign.
Credits: Credit top-up buried 4 clicks deep — a disproportionate share of support tickets.
Statistics: Statistics pages showed raw numbers with no interpretation — users couldn’t tell if a campaign was “good”.
Support tickets echoed the same complaints: credit purchase and renewal confusion, “how do I personalize my email,” and repeated repurchase logic requests. Ticket volume wasn’t a tracked metric — NPS was — so this is framed qualitatively, not with ticket percentages.
Method & database
Sessions with the CEO, CTO and other Chiefs · audit of old wireframes, support tickets and NPS free-text across ~14 months · full heuristic evaluation of the Portal.
Put on hold, not discarded
Several stakeholder points were genuinely relevant and were folded into the redesign. But we didn’t rely on them blindly — each hypothesis was put on hold pending real validation, once a working prototype existed. That validation, not the sessions alone, is what ultimately drove prioritization.
Validation
Testing after wireframing — not before — still caught what stakeholder input missed: 6 of 8 users misread the repurchase window entirely.
Round 1 — 8 users: Repurchase logic and statistics dashboard concept. 6 of 8 participants misunderstood the “suggested repurchase window” copy — leading to a full microcopy rewrite for the repurchase module, and a change from a slider to a stepped input.
Round 2 — 6 users: Redesigned credit top-up flow and navigation. Time-to-complete for “buy more credits” fell from 42s to 11s, validating the new flow before build.
Counterintuitive insight
Stakeholder input wasn’t wrong — it was incomplete. It named what could be complex, but not where that complexity would actually show up: copy and interaction failures in the Repurchase screens.
Process note
Wireframes existed around week 7, so discovery order was backwards from best practice — validation came after concept, not before. Even 3–4 unmoderated guerrilla interviews earlier would have caught this two weeks sooner.
Analysis & synthesis
A 40-page heuristic audit, traced to reusable components instead of one-off screens, is what got findings prioritized into sprints.
The CTO’s own sessions confirmed why the earlier attempts hadn’t stuck: nothing had ever been documented. This audit was built to close exactly that gap — instead of flagging one-off screens, every violation traced back to a specific, reusable component, so the debate couldn’t repeat itself with the next hire.
The catastrophic example: no error state on failed triggers — users believed sends had gone out when they hadn’t.
We structured the audit atomic-design-style — atoms, molecules, organisms — and each finding shipped with a recommended fix and an estimated engineering effort. That practice didn’t exist at Enviou before, and it’s the reason this report drove real sprint work instead of sitting in a drawer.
The process combined a full UI and user-journey analysis, run in sessions alongside stakeholders and the engineering team — each represented by its own post-it color. Each purple circle marks a point analyzed and the specific heuristic being broken; a companion written document paired every circle with its severity, the nature of the problem, the reasoning behind it, and a recommended fix.
Worked example
One banner, six broken heuristics — and a fix for each one.
A worked sample from the synthesis: the Ted Banner area, broken down by primary/secondary problem, barrier or noise, and severity — with a justification and a recommended adjustment for every issue found.
Catastrophic: Deleting the card makes it disappear entirely until the page is refreshed; it can’t be minimized, only deleted; its warnings rarely match the tool’s current state; and there is no trace of the action and no undo.
Serious: Experienced users may recover, but the missing undo still costs them. For novice users, losing this card is catastrophic — and the card is so minimal in color that it reads as negligible in the first place.
Cosmetic: Titles and buttons use a different automation color from the rest of the product, and the pop-up appears far from the click that triggered it.
How to adjust
Make the card minimizable instead of deletable · use color to give it real prominence · standardize a solid palette (positive, error, inactive) · bring the pop-up next to its trigger, or use an overlay carrying the full variation content.
Strategy
Fixing the system before the screens — a design system as a decision log, not a component library.
In the new governance model I owned the design system as design lead. Any new component request went through a lightweight RFC — a one-page doc plus async review — before being added. That prevented the “everyone builds their own button” problem that had caused the original fragmentation.
Brand redesign
New logo, color system and typography — extended into the product UI so brand and product stopped looking like two companies.
Redesigned interfaces
Rebuilt automation creation, credit management and the statistics dashboard. Channels expanded beyond email and SMS to push and WhatsApp, unified under one creation flow instead of three.
By the numbers
Codesign changed the dashboard: Statistics were reframed around “last updated X min ago”, because engineering’s pipeline was hourly-batch, not real-time.
Codesign changed the system’s scope: Copy and content standards moved into the design system, which is what let us finally fix the Portuguese-language errors scattered through the old product.
Why this fit in 9 months
Before any interface work began, I partnered directly with the Tech Lead to study what was structurally feasible — jointly defining scope, instead of discovering the limits after design was already finished.
New automations
Repurchase, wired to real Shopify purchase data instead of manual guesswork, drove automations per account from ~4 to 15.
Three new automations shipped alongside the redesign, each answering a different pressure point uncovered along the way.
Repurchase
Built directly from a discovery finding — clients were manually recreating the same automation for recurring products. It connects to Shopify, pulling both product catalog and actual purchase data, so the repurchase window reflects real buying behavior rather than generic timing.
Custom Trigger
Essentially a “dev area” inside the product, letting companies structure fully custom personalization logic. Intentionally not broadly accessible — for larger clients with an in-house technical team, it’s a sustainable advantage over requesting custom work.
Retargeting
The third automation launched in this project, bringing Enviou’s catalog closer to what competitors already offered.
Interface showcase
Two screens that turned support tickets into self-service.
Why this screen exists: Credit purchase and renewal were among the most frequent support requests — the top-up itself sat four clicks deep.
Design decision: My Subscription — one of the most requested additions — lets clients see credit usage per product without leaving the flow or opening a ticket.
Why this screen exists: Clients had no reliable way to tell which automations were actually running, on which channel.
Design decision: My Automation lists every product’s automations in one place, with a live signal for which channel — email, SMS, WhatsApp or push — is active on each.
Outcomes
A redesign guided by heuristic and usability data — not aesthetic preference — cut rage clicks by 80% and lifted NPS from 44 to 60.
Read together, these numbers say the same thing: independence went up. Fewer support-driven clicks, faster self-service, more automations built without asking Enviou for help.
New client signups grew 75% in the same quarter too, though that reflects several initiatives landing together — redesign, pricing, and marketing — not design alone.
Reflection
The system — not any single screen — is what outlasted the team that built it.
What I’d do differently
Reorder discovery and validation — test before wireframing, not after · push harder for a real accessibility audit, not just contrast and keyboard defaults · push harder for dedicated research earlier, rather than running discovery solo · instrument attribution from day one, isolating design impact from pricing and marketing changes.
Design lead ↔ engineering lead: Joint ownership of the architecture study and scope definition from the outset — not a downstream handoff, but a shared foundation for what was buildable before any interface work started. Once the team was reduced to just me, that partnership expanded from architecture into interface construction itself.
Design ↔ engineering: Biweekly codesign sessions with the Tech Lead from week 3 onward — constraints surfaced before designs were finalized, not discovered later in sprint planning.
Design ↔ customer support: The CS team recruited usability participants and supplied raw ticket data for the retroactive analysis.
"Three redesigns failed because nothing was written down. This one held because the decisions were."
Closing thought
Let's talk