Case study · Forms platform

Merging seven form
builders into one

I was the only designer. I did the research, designed the new builder, and used what I learned to change the roadmap.

Watch the tour

CompanyKeap
RoleSole designer · end to end
Team1 PM · 1 PO · 1 tech manager · 1 development team
Research2 usability studies
Timeframe2025
FocusResearch · Systems · AI · Mobile
TL;DR

Keap had 7 form builders doing the same job, and no single team owned them. I argued for merging them into one builder on the existing backend instead of a 12+ month rebuild. That one builder has shipped in full, plus AI that drafts a complete form from a prompt. 79% of about 100K form users created at least one form in the new builder within three months of launch.

At a glance
12+
Month rebuild I argued against
7→1
Form builders merged into one, now shipped
79%
Of about 100K form users created a form in the new builder within three months
3 of 3
Top ranked survey features shipped
01 · The problem

Seven ways to build a form, and users still left

Four live builders and three legacy ones, each with its own UX, spread across four teams. Any improvement had to be built seven times, and no single team owned the whole problem, so I took it on.

I thought web forms ARE public forms. Do not know the difference.Survey respondent

Users couldn’t tell them apart

The names came from our product vocabulary, not theirs, so there was no obvious one to move to.

So they built forms elsewhere

Our survey found four other form tools people used instead, and a quarter named the same one.

One form got worse

Only one form hit its measured target, and the most used form scored lower after its rebuild than before.

02 · My role

The only designer on it

01

Pulled six years of research together

Five studies since 2020 had each sat in a folder. Side by side, they showed the problem was structural.

02

Ran new research

A survey of 25 people with two researchers, an audit of nine competitors, a prioritization workshop, and both usability studies.

03

Set the direction and the phases

Parity (matching every feature the old builders had before adding new ones) first, then consolidation, then appointments and payments. I got the team that owned checkout on board first.

04

Argued against the full rebuild

Engineering estimated a rebuild at twelve months or more. I showed the data models were sound and the fragmentation was in what users saw, so a rebuild meant regression risk for little payoff.

Timeline of nine form studies from 2020 to 2026. Five came before me: a form builder A/B test, two usability studies, form discovery and a parity feature priority study. I ran or designed the other four: a prioritization workshop, a survey of 25 people, a builder usability study with 7 participants, and a unified prototype study with 8 participants plus a Kano study with 11.
Nine studies from 2020 to 2026. I ran the four highlighted ones.
03 · The approach

Ask what the form is for

One builder with four use cases replaces seven products, and the use case decides where the form can appear.

One builder Lead captureInternalLanding pageCheckout
The model and the picker come from different stages, so their use case names differ.
Create form screen asking What form do you want to create, with four cards: Marketing (Public), Management (Internal), Payment (Checkout) and Appointment
The picker that replaced the seven product names, with the old names in brackets.
The unified builder in the interactive prototype: a Details, Fields, Style, Settings, Share stepper above the client intake form fields, with a live branded preview of the form beside them
The unified builder with its live preview. Interactive prototype, demo data.
04 · Testing

Round one went badly, and that was useful

Round one · 7 participants

None of the seven could add a step, and six got tripped up by the old categories. The live preview earned their trust.

Round two · 8 participants

A year later, the SUS score reached 89, an A grade. Manual creation scored a perfect 3.00 on the flow that broke for everyone, and most people didn’t notice the merge.

What still didn’t work

Finding Forms was the only real friction left. People looked in Automations, Contacts, Sales and Reporting first.

Round two, with eight participants and seven tasks. Manual creation scored a perfect 3.00 out of 3, on the flow that broke for all seven participants a year earlier. The SUS score reached 89 out of 100, an A grade, and most participants never noticed the merge. Finding Forms scored 5.88 out of 7 for ease of use, the only real friction. In round one, a year earlier, every participant hit friction and none of the seven could find how to add a step.
Round two results, a year after round one.
05 · AI

One builder made AI drafts possible

With one builder instead of seven, AI can produce a complete form that uses everything the builder supports.

Fields, copy, brand and template arrive together. Interactive prototype.

Thryv AI building a client intake form: a checklist ticks off fields, copy, brand and template while a live preview assembles beside it, then template chips switch the look
Thryv AI welcome screen: the prompt placeholder cycles through examples above commonly created forms, picking Client intake types the prompt, and Thryv AI answers with two quick questions
I won’t use AI for minor edits: typing instructions takes longer than just doing it. But I would for an entirely new form.Usability study participant

Nearly everyone tries AI once, so I added suggestion chips and templates so nobody starts from an empty prompt box.

The Style step in the interactive prototype: four template cards, Classic, Spotlight, Cover and Minimal, above a Branding card synced to the Branding Center, with the live preview beside them
Templates as the fallback, branded from the Branding Center. Interactive prototype.
06 · Mobile

Built on a phone, filled on a phone

I designed both sides on a phone, the owner who makes the form and the customer who fills it out, with the same builder and Thryv AI as desktop.

Tap, don’t type

Answers you can tap, so a draft can come from two taps instead of a paragraph.

Fields sized for fingers

52 pixel fields with 16 pixel text, so iOS doesn’t zoom in. Submit stays pinned in reach.

Thryv AI on a phone asking two questions about a client intake form, with three answers to tap: new customers, my internal team, or add a preferred contact method
The built form in the Thryv AI chat on a phone: every checklist step done, a preview with the Otter and Rocket logo, template chips for Classic, Spotlight, Cover and Minimal, and an Open in builder button
The share sheet on a phone over the builder: the form is published, the link has a Copy button, and there are buttons for Text message, Email, QR code and More
The owner’s side, ending in a share sheet with text, email and QR code. Interactive prototype, demo data.
The client intake form at phone width in the Classic template: logo on top, outlined fields, and Submit pinned to the bottom
The same form in the Spotlight template: a navy header band with the headline in white
The same form in the Cover template: an artwork banner, a round logo and soft filled fields
The same form in the Minimal template: a large headline on white with underlined fields
The customer’s side: the same form in all four templates. Interactive prototype, demo data.
07 · Decisions

Where I changed direction

Bar chart of the share of survey respondents who ranked each feature in their top three, from 13 of the 25: hidden fields 46%, copy and embed unstyled HTML 46%, apply tags from field selections 39%, conditional fields 31%, thank you page from a landing page 23%, and insert merge fields in long text 15%. The top three are emphasized.
How 13 of 25 survey respondents ranked features. The darker bars are the three parity features I shipped.
The full rebuild
I kept the existing backend, merged the UI on top of it, and shipped it in stages.
Merging all form types at once
The backend was still split, and checkout had its own payment logic. I merged the two closest first and designed the rest as future states.
Full AI form creation
People wanted AI to start a form, not for small edits. I shipped AI drafting with manual editing after it, and templates as the fallback.
Selling it with a deck
Slides couldn’t show what a three year plan would feel like. I built a working prototype instead. Leadership got it within days, and I got the sign off.

A Kano study (a survey method that sorts features by how much users notice their presence or absence) didn’t match our plan. Conditional logic ranked highest in both it and the survey, but wasn’t on the roadmap. I took that to leadership: conditional logic moved up, and the payments phase came out.

Who decided what

Me, with the PM
I shaped the scope and product calls with the PM: merge instead of rebuild, what parity covered, and the AI direction.
The PM and PO
Owned the timeline, delivery and story writing.
Leadership
Signed off on the roadmap change.
Kano study with 11 participants, showing satisfaction if present and dissatisfaction if absent. Conditional logic: 60% and 30%, attractive, with the highest stated importance. Appointments in forms: 60% and 10%, attractive. Payments in forms: 22% and 33%, indifferent, and two respondents did not want it.
The Kano study that changed the roadmap. Payments was scheduled, and users were indifferent.
Conditional fields in the interactive prototype: a panel reading If Service type is Web design, Then show Website budget, open beside a three step Service Request Form
Conditional logic, as designed. Interactive prototype, demo data.
08 · Outcome

What shipped and what’s next

ShippedAll three of the survey’s top ranked parity features.
ShippedThe unified builder, in full. Every stage of the merge has shipped.
In progressA single form service replacing 32 legacy endpoints (old backend APIs the previous builders depended on). Twelve are done, and the remaining gaps are mapped.
On holdFuture improvements are designed and on hold.

Once the form service is finished, one team can own it, and the tech debt of seven implementations goes away. Forms also work across the platform, with AI and MCP (Model Context Protocol, a standard way to let AI tools call the product’s functions) layers on top. It took four teams, two months of design research and six months of development. So far, 79% of about 100K form users created at least one form in the new builder within three months of launch, and support reported fewer form-builder tickets after launch.

One thing I’d do differently

I fixed the builder before the front door
I put navigation off because the survey ranked parity features higher. I’d flip that now. The navigation fix was much cheaper than the features I shipped first.

Let’s talk