EmailFlow Pro

At Serasa Experian (the Brazilian arm of the global credit bureau), I was the designer responsible for producing campaign emails in Salesforce Marketing Cloud. The friction wasn't theoretical — I felt it every day: changes that should take minutes meant diving back into code, and every campaign repeated the same manual ritual.

My approach was pragmatic: if I can't use UX and AI to fix my own daily workflow, how can I design efficient solutions for millions of people? EmailFlow Pro wasn't born as a portfolio piece. It was a survival tool.

Fig. 01

The users weren't a market — they were my own team: the internal CRM and email-marketing staff at Serasa Experian who finish campaign emails in Salesforce Marketing Cloud. People fluent in HTML and design, but not interested in defending AMPScript edge cases on a deadline.

Designing for internal colleagues changed the priorities. There's no marketing funnel, no onboarding, no sign-up — just a tool that has to earn trust on the first paste. Because the stakes are millions of recipients, their deepest need wasn't speed, it was verifiable correctness. Speed was the headline; trust was the spec.

Jobs-to-be-done — framed as the outcomes my colleagues were "hiring" the tool to achieve, not features I wanted to build.

This was not a handoff exercise. I defined the workflow around the job I needed to complete, designed the interface that made the steps visible, and built the processing logic that made the result repeatable.

Developing for Salesforce Marketing Cloud (SFMC) requires absolute precision — these emails went out to millions of Serasa Experian customers. My manual workflow was a massive bottleneck:

• Time-consuming: 80 minutes per email.
• Image re-linking: Hand-swapping every local image src for its SFMC content URL, one at a time, from an upload log.
• High risk: Manually injecting AMPScript and checking 20+ links per campaign for millions of customers.

I was acting like a "human compiler." My creative energy was being drained by tasks that technology should have solved—but didn't.

My first instinct was to use ChatGPT. I tried custom GPTs and complex prompts, but it revealed the limits of conversational AI:

• Technical Hallucinations: GPT often invented AMPScript syntax, creating code that crashed inside Salesforce.
• Inconsistency: Being a probabilistic model, the HTML structure changed every time. I spent more time reviewing the AI's work than doing it myself.
• Lack of Workflow: A chat is a conversation, not a tool. Copying and pasting code back and forth is just another bottleneck. I needed a precision tool, not a chat window.

If the tool I needed didn't exist, I had to build it — without asking for budget, a dev team, or permission. I ran an AI-assisted development cycle in two stages:

• Functional ideation (Lovable): I prototyped the interface around utility UX. The design had to "disappear" so the function could shine: raw code in, polished code out.
• Precision engineering (Cursor): I moved the project to Cursor to code rigid business rules — deterministic sanitization that an LLM couldn't maintain consistently, ensuring the HTML met global email-rendering standards.

A tool my team would trust with millions of sends can't be validated by "it looks fine." The bar was strict: the assembled email had to match the version we already produced by hand — same images, same personalization, same links. So I tested it against real campaigns, not happy-path samples.

Testing against real data surfaced three failures that shaped the final product:

None of these showed up in the original prototype. They only appeared under real logs and real campaigns — which is exactly why shipping a working tool to my own team and stress-testing it beat polishing a mockup.

The result is a client-side single-page app with three inputs and one trustworthy output. You paste the raw email, hand it the data SFMC gives you, and it assembles a production-ready version — nothing leaves the browser.

Fig. 02

• HTML Editor: Paste, upload a .html file, preview the rendered email, and format the markup. The starting point for every campaign.
• Image Log Processor: The core. SFMC spits out an upload log like Nome: banner.gif, URL: …sfmc-content.com/…. The tool parses it (that format, filename → URL, or JSON) and automatically rewrites every image src to its final SFMC content URL — the most error-prone manual step, gone.
• AMPScript Block + Link Analysis: Paste your %%[ … ]%% block and it's injected in the right place; "Analisar Links" scans every link so you can verify protocols and tracking before the send reaches millions.

The measurable result was a reduction in campaign production time from approximately 80 minutes to 10–20 minutes per campaign. The tool also moved the risky steps into visible, repeatable checks.

• Production time: Approximately 80 minutes to 10–20 minutes per campaign.
• Verification: Image URLs, links, AMPScript placement, and processed output became visible checks in the workflow.
• Processing: The app runs client-side, so customer HTML stays in the browser during processing.

EmailFlow Pro changed how I work: when a tool limits the team, I treat that as a design problem and fix it. The same loop — understand, design, ship, measure — now runs through everything I build.