Survey Loop

I wanted Survey Loop to feel good to use, whether you were building a survey, answering one, or trying to understand what customers were telling you. I also wanted an API we could build other things around. That decision became especially useful when AI assistants started being able to work with real tools and data.

Read the walkthrough

A scripted assistant request creates a draft post-event survey. At 8 seconds, the draft opens in the real builder with recommendation, event rating, and written follow-up questions. At 16 seconds, the appearance controls change its branding from violet to green. At 26 seconds, the respondent preview selects ratings, writes “More time for workshop questions,” and reaches the completion screen. No response is submitted to a live service.

Building around contract work

After Capacity acquired Denim Social in late 2023, Brandon Spriggs and I co-founded Survey Loop. I lead design and engineering; Brandon leads product and the commercial side. We set out to bootstrap a customer feedback product with a simpler interface and room to grow.

Bootstrapping is harder than people make it sound. The first stretch was moonlighting around contract work. We built an MVP in Vue, put it in front of customers, and brought several large fintech companies into the beta. A few months in, we rebuilt the web app in React as the AI tooling and libraries we needed developed around that ecosystem. Making that change while the product was still young meant taking on a substantial rebuild, but doing it before another year of features accumulated.

Designing both sides of a survey

A lot of my work has been in the details of creating a survey. You can add and rearrange question blocks, write directly on the canvas, adjust the branding, and preview what a customer will see. Ratings, open questions, conditional logic, and multiple pages give teams room to ask what they need. Autosave and version history make it easier to experiment and recover an earlier version.

The respondent experience gets the same attention. The survey carries the company's branding through to the questions and completion screen, with a layout that works on a phone. I built and refined both sides of that experience, using a shared design system to keep the editor, survey, and wider application consistent as we added features.

Read the respondent flow

The respondent selects a recommendation score of 9 out of 10 and an event rating of 5 out of 5, then writes “More time for workshop questions.” The preview ends at the branded completion screen.

Calculated evidence before an AI explanation

From there, teams can distribute surveys through email, links, and QR codes, with reminders and tracked links to understand where responses came from. Results sit alongside question-level charts, customer satisfaction scores, and the original answers. Organizations with multiple locations or teams can look at their feedback together and then narrow the view to a particular part of the business.

Making those results useful has become a substantial part of the product. I worked on the dashboard's AI summaries and rebuilt the survey analysis around calculated statistics. The analysis checks whether differences are meaningful before asking AI to explain them in plain language. If nothing meets that threshold, the interface says so. The underlying figures stay available alongside the explanation.

The same product through an assistant

The API gave us another way to work with all of this. We added support for MCP, the Model Context Protocol, so a connected AI assistant can use Survey Loop's tools to create and edit surveys, read responses, and query analytics. The assistant gets structured data and defined actions through the same API the application uses.

For example, you could ask an assistant to create a draft post-event survey with an NPS question and a written follow-up. It can create that draft in Survey Loop, ready for you to open in the builder and refine. Once responses arrive, you could ask it to compare NPS across two periods and pull the written feedback for review. Those requests use the surveys and responses already in your account, without exporting a spreadsheet for each conversation.

That connection also has to respect how the product works. MCP access is authenticated and follows the user's organization permissions. Read-only connections can inspect data without changing it. A survey created through an assistant is the same survey you open in the application, so you can move between the two as the work requires.

Connecting to everyday work

I've also built API integrations around everyday workflows. The monday.com integration connects a workflow action to survey delivery, using the recipient details from a board item. Google Business Profile integration brings public reviews into the product, with AI reply drafts, approvals, and escalation rules. That gives teams a way to respond to feedback as well as collect it.

Owning the implementation means coming back to earlier decisions, too. As the product has grown, I've removed obsolete features and duplicated code, consolidated API documentation, and checked that existing workflows still hold up. Survey Loop is live, and we're continuing to develop it with customers.