AI with no playbook

Mike ran Authentic, an agency whose fundraising emails had won awards and raised millions for Democratic campaigns. He saw the gap clearly: the candidates who need that writing most — school board, city council, small nonprofits — are exactly the ones who can't afford an agency. His idea was to train a model on his agency's library of high-performing emails, so anyone could generate that quality on their own. That meant solving the same bottleneck from two sides: campaign teams that could not afford professional help, and agency strategists buried in repetitive drafting. Authentic hired Able, the product agency where I worked, to turn the idea into a usable MVP in less than six months.

I created the early prototype featured in an MSNBC segment about AI's growing role in political advertising. It made the idea tangible before we had solved the much harder problem of turning it into a product campaigns could actually trust. The prototype sold the possibility. The work that followed was figuring out how much context, guidance, and control the real product needed.

We had a bank of high-quality fundraising content, but we had to tag it ourselves, iterating on the parameters of tone and urgency as we worked. Since we were new to the fundraising space, we made sure to consult with Authentic's experts on both what tones were most important and what the different levels of urgency meant. The dataset was a design surface. We treated it like one.

My design colleague set up regular output review with Authentic's digital strategists, which we synthesized into themes together. They caught problems we never could have identified from the dataset alone: language a real fundraiser would never use, appeals that felt too formal or generic, and subtle differences between urgency that motivated someone and urgency that sounded desperate.

I led interviews with two very different groups: agency strategists writing fundraising emails at volume, and grassroots candidates often writing alone. Both needed a first draft that wasn't overly generic before they would adopt Quiller into their existing workflow: candidates were already hesitant to fundraise if it didn't feel authentic, while strategists could write a decent first draft themselves.

Campaign profiles carried each candidate's voice, positions, and talking points into every generation. Inside the writing experience, users could adjust tone and urgency, add more specific context, rate the result, explain what felt wrong, and keep refining the draft without starting over.

Because most users were new to generative AI, the interface also taught them what the model needed to produce something useful.

Before handoff to the Quiller team, we helped shape the roadmap proposal for what came next. We accomplished all of this in 6 months by prioritizing research, quick iteration, using existing design and UI kits, and working nimbly as a team.