← All work Case study 02 / 16

Redesigning Seek AI

Seek AI · 2023

Product Design UX/UI Web

FIG. 01 — Redesigning Seek AI, web, 2023

Client

Seek AI

Year

2023

Scope

System Design · Product · UX/UI

Recognition

Acquired by IBM 2025

Redesigning Seek AI. From SQL tool to conversational data assistant. 4 months. Acquired by IBM in 2025.

Impact at a Glance

+Acquired by IBM, 2025
+4-month MVP: strategy through shipped product
+Replaced a SQL editor with a chat-first interface
+Expanded the addressable market from 1 persona to 4 — from executives to entire data teams
+Built the trust layer that made AI-generated answers safe to hand to stakeholders

FIG. 02 —

FIG. 03 —

The Problem

Seek AI 1.0 asked business users to write SQL to get an answer out of their own data — a tool built for engineers, sold to executives. Steep learning curve, no way to verify AI-generated answers, and nothing built for teams.

Seek needed to become a natural-language product fast enough to pitch investors and land partners like Snowflake. Four months, start to ship.

The Role

Principal Product Designer, Tony Voorhees Studio (engaged via AE Studio) — 4 months, design lead.

Brought in to pressure-test the product direction. Two weeks later the Head of Product left, and the CEO handed me the redesign outright: user interviews, UX strategy, every screen, engineering handoff — against an investor deadline that didn’t move.

FIG. 04 —

The Insights

Every interview surfaced the same workaround: ask ChatGPT to write the SQL, paste it into Seek, hope it runs. Users had already solved the interface problem — with a different product.

That reframed everything: Seek needed to be conversational, not transactional. A person you ask, not a tool you operate.

Redesign

What Changed

Chat-first interface: One input bar, plain English in, SQL and results out. Writing SQL became optional, not required — so the product could serve technical and non-technical users at once.
Four personas instead of one: Natural language opened Seek to a much wider range of users — engineers, SQL writers, data-literate askers, and people with no SQL at all. That shift let Seek sell into entire data teams, not just executives — a key part of the investor story.
Permissions built for real organizations: A four-tier system (Org Admin → Data Analyst), modeled on how Slack and Google Workspace handle shared access, so teams could collaborate on proprietary data safely.
A verification system — the actual trust fix: The real risk wasn't the UI, it was a wrong answer delivered with total confidence. Verified badges, an insights log, and smart resurfacing meant people could share a Seek answer knowing a human had already checked it.
Lighter onboarding: Try Seek on sample data first, connect a real database once you're convinced.
A defined agent personality: Documented how Seek should sound and behave, so its tone was a design choice handed to engineering — not a default.
Multi-user chat, resolved simply: @mentions to route questions between Seek and teammates in shared threads — the simplest solution that shipped on time.