Retool AI Customer Research: How the Internal-Tools Platform Decides What to Build

TL;DR#

Retool, the developer platform for building internal tools, decides what to build by treating its own builders as a live research panel — combining hands-on usage of its own product, sales-led discovery, and large recurring survey reports like the State of Internal Tools and State of AI. Founder David Hsu reached roughly $2M in ARR before launch by personally running sales conversations with developers, a discovery method that still shapes the roadmap. Retool's 2026 Build vs. Buy Report surveyed 817 customers and found that 35% of enterprises have already replaced at least one SaaS tool with a custom build, with 78% planning more in 2026. That kind of structured listening is exactly where ai customer interviews change the economics: instead of a founder running every conversation by hand or a survey flattening developers into multiple-choice answers, conversational AI can interview hundreds of users at once and still follow up on the "why." This article breaks down how Retool actually learns from its users — and how product teams at any developer-tools company can apply the same approach with AI-moderated research.

How does Retool decide what to build?#

Retool decides what to build primarily by staying close to the developers and operations teams who use its platform every day, then validating direction through sales conversations and large-scale survey reports. The company is a low-code platform for assembling internal software — admin panels, support dashboards, customer portals — and is used by more than 10,000 companies including Amazon, Stripe, and OpenAI, according to Sequoia Capital's profile of founder David Hsu. Because Retool's users are themselves builders, the company has an unusually direct feedback loop: the people it sells to can articulate exactly what is broken in their workflow.

The founding story makes the discovery method explicit. Hsu pivoted into Retool with under 60 days of runway after recognizing a pattern in his own work — "I wish there was a faster way to build this software." As First Round Review documented, Retool reached roughly $2M in ARR before a public launch by running deep, founder-led sales and discovery calls with individual developers. That early discipline — learning what to build from direct conversation rather than assumption — is the through-line of everything Retool does today.

This is the same instinct behind how other category leaders run discovery. It mirrors the approach in how Linear builds its roadmap from customer feedback and how Notion decides what to build as a $10B company: stay close to power users, and treat their workflow friction as the roadmap.

Retool's research signals: usage, sales, and survey reports#

Retool's product discovery runs on three signals — its own product usage, sales-led conversations, and recurring industry survey reports. Each captures a different layer of customer truth, and together they form a discovery stack worth studying.

Signal 1: Dogfooding and direct usage. Retool builds internal tools, and it builds them on Retool. Hsu has described the product as "a fast way to build internal software," and the company's own teams are heavy users — meaning friction surfaces internally before customers ever report it.

Signal 2: Sales-led discovery. From the earliest days, Retool's go-to-market was sales-led specifically because complex technical buyers reveal more in conversation than in a form. The Acquired episode on Retool's path to product-market fit frames sales itself as the discovery engine: every deal is a structured interview about what the customer is actually trying to build.

Signal 3: Recurring survey reports. Retool publishes large annual reports — the State of Internal Tools (running since 2020) and the State of AI — that double as marketing and as primary research. These reports are how Retool listens at scale, and they reveal what the company believes about its market.

A developer-tools company that wants this same three-signal coverage without a giant research org can compress signals 2 and 3 using AI interviewer agents that run conversational interviews at survey scale. That is the gap between a static questionnaire and a real conversation — and it is the gap ai customer interviews are built to close.

What Retool's reports reveal about its market#

Retool's published reports show a company that listens to developers at scale and uses the findings to shape both product and positioning. The data points below are all drawn from Retool's own survey research and the press coverage of it.

The State of Internal Tools 2022 finding — that developers spend about a third of their time on internal tooling — is the entire thesis of Retool's product, validated by survey. The State of AI report adds nuance: as Wing Venture Capital's analysis notes, individual productivity gains are real even where organization-wide transformation lags. And the 2026 Build vs. Buy Report — based on 817 customers and reported by BusinessWire — found that 60% of those custom builds happened as shadow IT, outside official procurement.

Notice what these reports are: structured listening at scale, packaged so the findings feed both the roadmap and demand generation. It is the same playbook behind how Figma runs customer research and how Stripe approaches research as a payments leader serving 4M businesses.

Where surveys hit a ceiling for developer-tools research#

Surveys hit a ceiling because they capture what developers select, not what they mean — and developer feedback is full of "it depends." A multiple-choice question can tell Retool that 38.9% of respondents cite model accuracy and hallucinations as a pain point, but it cannot ask the obvious follow-up: which workflows, in which conditions, and what did you try instead? The richest signal in technical discovery lives in the follow-up, and a form has no follow-up.

This is the core limitation Perspective AI exists to fix. Forms flatten people into dropdowns and short fields; they front-load effort before the respondent feels understood; and they fail precisely at the messy, high-value moments where a developer says "well, it depends on the database." Static questionnaires turn a nuanced engineering reality into a tidy chart that hides the reasoning.

The alternative is not "more surveys" — it is conversation at scale. AI customer interviews let a developer describe a workflow in their own words while an AI interviewer probes the specifics, the same way Hsu probed in those early sales calls. Teams building this kind of always-on, continuous discovery practice get the depth of an interview with the reach of a survey.

How AI customer interviews apply to developer-tools discovery#

AI customer interviews apply to developer-tools discovery by automating the conversation Retool runs manually — interviewing hundreds of builders at once while still following up on each answer. The mechanics map directly onto Retool's three-signal model.

1. Replace the survey instrument with a conversation. Instead of the State-of-X questionnaire, an AI interviewer agent asks open questions and adapts. A developer who says they "build mostly on Postgres" gets asked why, what breaks, and what they wish existed — automatically.
2. Scale the founder's discovery call. Hsu's early advantage was that he personally interviewed customers. AI conversations reproduce that depth across thousands of users, so a product team gets founder-quality discovery without the founder's calendar.
3. Turn intake into research. When developers sign up, hit a wall, or request a feature, a concierge agent can capture the "why now" in the moment instead of routing them to a static form.
4. Make it continuous. Annual reports are snapshots. Conversational research can run as an always-on stream, so the roadmap reflects this quarter's reality, not last year's survey.

This is the methodology behind product discovery research that replaces surveys and scripts and the continuous discovery stack for AI-first product teams. The continuous-touchpoint model — at least one customer conversation per week — is the backbone of modern discovery, as Product School's continuous discovery guide explains, drawing on Teresa Torres's framework. AI is what makes "one conversation a week" turn into "a hundred conversations a week."

A discovery playbook for developer-tools teams#

A developer-tools team can adapt Retool's approach with a four-part playbook that pairs structured listening with conversational depth. Use it whether you sell to engineers, operations teams, or both.

• Run a recurring "state of" study. Like Retool's annual reports, commit to a repeatable research cadence so you can measure change over time, not just snapshots. Stand one up quickly with Perspective AI studies.
• Interview the workflow, not the feature request. Developers ask for features but need outcomes. Probe the underlying job — the same instinct that took Hsu from FileMaker developers to React developers.
• Capture shadow usage. Retool's 2026 report found 60% of custom builds happen as shadow IT. The most honest signal often lives outside official channels — design intake that invites the unofficial story.
• Close the loop continuously. Don't wait for the annual survey. Launch a new research study whenever a release ships, and let conversations run between launches.

Companies across the developer-tooling and horizontal-SaaS landscape are converging on this model — see how it plays out in Atlassian's AI customer discovery across Jira and Confluence, GitLab's strategy for listening to 30M users, and Vercel's AI-native onboarding for developer teams.

Frequently Asked Questions#

How does Retool do product discovery?#

Retool does product discovery by combining heavy internal usage of its own platform, founder- and sales-led customer conversations, and recurring large-scale survey reports like the State of Internal Tools and State of AI. Founder David Hsu reached roughly $2M in ARR before launch through direct discovery calls with developers, and that conversation-first approach still shapes the roadmap today.

What is Retool's State of Internal Tools report?#

Retool's State of Internal Tools report is an annual survey of tech professionals, running since 2020, that measures how companies build and maintain internal software. A widely cited finding from the 2022 edition is that developers spend roughly 33% of their time building internal tools — a statistic that validates Retool's entire reason for existing.

What did Retool's 2026 Build vs. Buy Report find?#

Retool's 2026 Build vs. Buy Report found that 35% of enterprises have already replaced at least one SaaS tool with a custom build, and 78% expect to build more custom internal tools in 2026. Based on a survey of 817 customers conducted in late 2025, it also reported that 60% of those custom builds happened as shadow IT, outside official procurement.

Are AI customer interviews better than surveys for developer research?#

AI customer interviews are generally better than surveys for developer research because they follow up on vague or conditional answers, which dominate technical feedback. A survey can record that a developer "builds on Postgres," but only a conversation can ask why, what breaks, and what they wish existed — capturing the reasoning a multiple-choice field discards.

How can a developer-tools company scale customer interviews with AI?#

A developer-tools company can scale customer interviews with AI by deploying an AI interviewer agent that runs open-ended, adaptive conversations with hundreds of users simultaneously. This reproduces the depth of a founder-led sales call at survey scale, and can run continuously rather than as a once-a-year report, so the roadmap reflects current reality.

Conclusion#

Retool decides what to build by listening to developers in three layers — using its own product, running sales-led discovery, and publishing recurring survey reports that turn structured listening into both roadmap fuel and market data. The pattern is clear: the companies that learn fastest are the ones that stay closest to the actual words of their users, the way founder David Hsu did in those first $2M-in-ARR sales calls. Surveys alone hit a ceiling because developer truth lives in the follow-up, not the dropdown. That is exactly where ai customer interviews change the math — giving product teams the depth of a one-on-one conversation at the scale of a survey, continuously. If you build for developers and want to learn what to build next, start a research study with Perspective AI and let AI interview your users in their own words.