Customer story
Fundamentally Changing How a Company Asks and Answers Data Questions
Fundamentally Changing How a Company Asks and Answers Data Questions
How Speak's Data Lead reimagined how a 200-person company gets answers from its
data, turning weeks-long queues into on-demand, self-serve analytics with Indent.
<2 min
<2 min
median response time
38.7K+
38.7K+
SQL queries generated
4.4K+
4.4K+
data visualizations created
Industry
AI Consumer Education
Team size
150+
Products used
Indent Code Handles tasks end-to-end on real codebases, opens PRs, self-heals on CI failures.
Indent Review Enforces nuanced code standards, proposes real tested changes, learns from ignored/rejected feedback.
Indent Data Get instant, accurate answers to any data question using AI that understands your data warehouse, BI tools, and code.
Industry
AI Consumer Education
Team size
150+
Products used
Indent Code Handles tasks end-to-end on real codebases, opens PRs, self-heals on CI failures.
Indent Review Enforces nuanced code standards, proposes real tested changes, learns from ignored/rejected feedback.
Indent Data Get instant, accurate answers to any data question using AI that understands your data warehouse, BI tools, and code.
About Speak
Speak is an AI-powered language learning platform that recently crossed $100M in ARR, with
over 15 million downloads globally. It is the top-grossing education app in South Korea's App
Store and was named to Forbes' 2025 AI 50 list. The company raised $160 million in total
funding at a $1 billion valuation, with backing from OpenAI, Accel, Founders Fund, and Khosla
Ventures.
Speak is an AI-powered language learning platform that recently crossed $100M in ARR, with over 15 million downloads globally. It is the top-grossing education app in South Korea's App Store and was named to Forbes' 2025 AI 50 list. The company raised $160 million in total
funding at a $1 billion valuation, with backing from OpenAI, Accel, Founders Fund, and Khosla Ventures.
The Problem
Jared Forman was Speak's Data Lead and the primary point of contact for business analytics
across growth, sales, finance, and monetization. Speak is a 200-person company, and over 70
of those people sat on the business teams Jared supported, most of them non-technical and
distributed globally.
Speak had dashboards, but the questions people needed answered were too specific for pre-
built views. Ad hoc requests stretched to weeks for lower-priority asks, and every hour spent
answering a one-off question was an hour not spent doing critical work to grow Speak's
business and improve the experience for users.
Jared Forman was Speak's Data Lead and the primary point of contact for business analytics across growth, sales, finance, and monetization. Speak is a 200-person company, and over 70 of those people sat on the business teams Jared supported, most of them non-technical and distributed globally.
Speak had dashboards, but the questions people needed answered were too specific for pre-built views. Ad hoc requests stretched to weeks for lower-priority asks, and every hour spent answering a one-off question was an hour not spent doing critical work to grow Speak's business and improve the experience for users.
"I was one person covering ingestion, modeling, experimentation, and strategic analysis,while also fielding questions from 70+ people on the business side. The dashboards we had couldn't keep up with the specificity of what people were asking."
Jared Forman, Data Lead at Speak
Finding Indent
Jared was looking for a natural language querying tool for ad hoc questions. He was connected to the Indent founders and started a pilot.
Working collaboratively with the Indent team, Jared integrated Indent with BigQuery, Airflow, Slack, GitHub, Notion, and Mode. He set up a public Slack channel so the whole company could
see questions going in and answers coming back. Within the first week, 16 people were
querying BigQuery in natural language.
Jared was looking for a natural language querying tool for ad hoc questions. He was connected to the Indent founders and started a pilot.
Working collaboratively with the Indent team, Jared integrated Indent with BigQuery, Airflow, Slack, GitHub, Notion, and Mode. He set up a public Slack channel so the whole company could see questions going in and answers coming back. Within the first week, 16 people were
querying BigQuery in natural language.
Jared was looking for a natural language querying tool for ad hoc questions. He was connected to the Indent founders and started a pilot.
Working collaboratively with the Indent team, Jared integrated Indent with BigQuery, Airflow, Slack, GitHub, Notion, and Mode. He set up a public Slack channel so the whole company could
see questions going in and answers coming back. Within the first week, 16 people were querying BigQuery in natural language.
Building Context From Existing Code
Speak maintained two codebases that encoded institutional knowledge about their data: dbt
models orchestrated through Airflow, and Mode dashboard SQL. Indent used both as a
makeshift semantic layer, seeding the agent with context about how tables relate, how to join
them, and where to look for specific metrics.
For questions that mapped to existing pipelines and definitions, this context made answers
significantly more accurate. But Indent's agent was also capable of going deeper, sifting through
thousands of tables to answer entirely new questions that no one had built a pipeline for.
Speak maintained two codebases that encoded institutional knowledge about their data: dbt
models orchestrated through Airflow, and Mode dashboard SQL. Indent used both as a
makeshift semantic layer, seeding the agent with context about how tables relate, how to join
them, and where to look for specific metrics.
For questions that mapped to existing pipelines and definitions, this context made answers
significantly more accurate. But Indent's agent was also capable of going deeper, sifting through thousands of tables to answer entirely new questions that no one had built a pipeline for.
Speak maintained two codebases that encoded institutional knowledge about their data: dbt models orchestrated through Airflow, and Mode dashboard SQL. Indent used both as a makeshift semantic layer, seeding the agent with context about how tables relate, how to join
them, and where to look for specific metrics.
For questions that mapped to existing pipelines and definitions, this context made answers significantly more accurate. But Indent's agent was also capable of going deeper, sifting through thousands of tables to answer entirely new questions that no one had built a pipeline for.
"We had core tables and pipelines, but not everything was modeled. Indent used what we had as a foundation and was capable enough to explore beyond it when the question required it."
Jared Forman
The aha moment came when the team realized that the more context they fed Indent, the more
it behaved like a true copilot. Answers got more nuanced and more thoughtful as Indent
absorbed more of Speak's institutional knowledge.
The aha moment came when the team realized that the more context they fed Indent, the more it behaved like a true copilot. Answers got more nuanced and more thoughtful as Indent
absorbed more of Speak's institutional knowledge.
"The better the context we provide, the more nuanced and thoughtful the answers get. That was the real unlock. We went from treating Indent as a query tool to investing in it as a true copilot."
Jared Forman
How It Works
Someone asks a question in Slack. Indent pulls the relevant data from BigQuery, writes and
runs SQL, and layers in context from Airflow logs, dbt lineage, and existing BI definitions. It
returns an answer with charts, the underlying data, and a clear trace of how it got there.
What sets Indent apart is the reasoning loop. Instead of asking the user a series of clarifying
questions, Indent works through the problem autonomously: pulling data, reasoning about what
it found, deciding where to look next, and iterating until it reaches an answer. One question in,
one answer out, with the full chain of logic visible.
Someone asks a question in Slack. Indent pulls the relevant data from BigQuery, writes and runs SQL, and layers in context from Airflow logs, dbt lineage, and existing BI definitions. It
returns an answer with charts, the underlying data, and a clear trace of how it got there.
What sets Indent apart is the reasoning loop. Instead of asking the user a series of clarifying questions, Indent works through the problem autonomously: pulling data, reasoning about what
it found, deciding where to look next, and iterating until it reaches an answer. One question in, one answer out, with the full chain of logic visible.
Someone asks a question in Slack. Indent pulls the relevant data from BigQuery, writes and
runs SQL, and layers in context from Airflow logs, dbt lineage, and existing BI definitions. It
returns an answer with charts, the underlying data, and a clear trace of how it got there.
What sets Indent apart is the reasoning loop. Instead of asking the user a series of clarifying
questions, Indent works through the problem autonomously: pulling data, reasoning about what it found, deciding where to look next, and iterating until it reaches an answer. One question in, one answer out, with the full chain of logic visible.
Someone asks a question in Slack. Indent pulls the relevant data from BigQuery, writes and runs SQL, and layers in context from Airflow logs, dbt lineage, and existing BI definitions. It returns an answer with charts, the underlying data, and a clear trace of how it got there.
What sets Indent apart is the reasoning loop. Instead of asking the user a series of clarifying questions, Indent works through the problem autonomously: pulling data, reasoning about what it found, deciding where to look next, and iterating until it reaches an answer. One question in, one answer out, with the full chain of logic visible.
"Most tools ask you five questions before they start and require a complete semantic
model. Indent chugs through to an answer in one shot and shows you how it got there and the data behind it."
"Most tools ask you five questions before they start and require a complete semantic model. Indent chugs through to an answer in one shot and shows you how it got there and the data behind it."
Jared Forman
What Actually Changed
The biggest shift wasn't that Jared's queue disappeared. People started asking more questions.
Indent acts as a first line of defense: it handles the volume, gives people immediate answers,
and surfaces the investigations that actually need a human analyst to follow up on.
The public Slack channel changed the culture. Seeing questions asked and answered in the
open promoted data-driven thinking across teams. People learned what questions to ask by
watching what others were asking.
Product managers now run retention cohort analyses right after shipping changes. Growth
marketers analyze campaign performance across regions and markets in minutes instead of
waiting days. Finance investigates subscription and revenue trends directly in Slack.
The biggest shift wasn't that Jared's queue disappeared. People started asking more questions. Indent acts as a first line of defense: it handles the volume, gives people immediate answers, and surfaces the investigations that actually need a human analyst to follow up on.
The public Slack channel changed the culture. Seeing questions asked and answered in the open promoted data-driven thinking across teams. People learned what questions to ask by watching what others were asking.
Product managers now run retention cohort analyses right after shipping changes. Growth marketers analyze campaign performance across regions and markets in minutes instead of
waiting days. Finance investigates subscription and revenue trends directly in Slack.
The biggest shift wasn't that Jared's queue disappeared. People started asking more questions. Indent acts as a first line of defense: it handles the volume, gives people immediate answers, and surfaces the investigations that actually need a human analyst to follow up on.
The public Slack channel changed the culture. Seeing questions asked and answered in the
open promoted data-driven thinking across teams. People learned what questions to ask by
watching what others were asking.
Product managers now run retention cohort analyses right after shipping changes. Growth
marketers analyze campaign performance across regions and markets in minutes instead of
waiting days. Finance investigates subscription and revenue trends directly in Slack.
The biggest shift wasn't that Jared's queue disappeared. People started asking more questions. Indent acts as a first line of defense: it handles the volume, gives people immediate answers, and surfaces the investigations that actually need a human analyst to follow up on.
The public Slack channel changed the culture. Seeing questions asked and answered in the open promoted data-driven thinking across teams. People learned what questions to ask by
watching what others were asking.
Product managers now run retention cohort analyses right after shipping changes. Growth marketers analyze campaign performance across regions and markets in minutes instead of waiting days. Finance investigates subscription and revenue trends directly in Slack.
Indent Across the Organization
Indent has penetrated every team at Speak. The rest of the analytics team and stakeholders
across the company use it to run analyses, investigate alerts and bugs, and prioritize and ship
impactful features.
SJ Cha, Speak's AI Product Manager, uses Indent as part of his core workflows to scale
Speak's content generation across languages and levels.
Indent has penetrated every team at Speak. The rest of the analytics team and stakeholders across the company use it to run analyses, investigate alerts and bugs, and prioritize and ship impactful features.
SJ Cha, Speak's AI Product Manager, uses Indent as part of his core workflows to scale Speak's content generation across languages and levels.
"I've built custom MCPs on top of Indent to generate and evaluate content, manage tickets, and run analysis. It went from an analytics tool to core infrastructure for how I build product."
SJ Cha, AI Product Manager at Speak
Khunal Parkash, GM of New Bets, uses Indent to drive Speak's enterprise language learning
offering, Speak for Business.
Khunal Parkash, GM of New Bets, uses Indent to drive Speak's enterprise language learning offering, Speak for Business.
"I'm building a new business line without a dedicated analyst. Indent lets me dig into
pipeline data, understand how enterprise customers are using the product, and make decisions without waiting in a queue."
"I'm building a new business line without a dedicated analyst. Indent lets me dig into pipeline data, understand how enterprise customers are using the product, and make decisions without waiting in a queue."
Khunal Parkash, GM of New Bets at Speak
Some answers aren't perfect, but getting a directionally right answer in two minutes is
enormously more valuable than waiting days for a polished one. And when Jared needs to focus
on infrastructure, modeling, or experimentation, the team has somewhere to go.
Some answers aren't perfect, but getting a directionally right answer in two minutes is enormously more valuable than waiting days for a polished one. And when Jared needs to focus on infrastructure, modeling, or experimentation, the team has somewhere to go.
Some answers aren't perfect, but getting a directionally right answer in two minutes is
enormously more valuable than waiting days for a polished one. And when Jared needs to focus on infrastructure, modeling, or experimentation, the team has somewhere to go.
"People ask more questions now, not fewer. Indent handles the first pass, and the things
that need real investigation get escalated to me. It helps me do everything I need to do across the full stack."
"People ask more questions now, not fewer. Indent handles the first pass, and the things that need real investigation get escalated to me. It helps me do everything I need to do across the full stack."
Jared Forman
Beyond Analytics
Jared's role at Speak spans the full data stack: ingestion, modeling, analytics, experimentation,
and strategic analysis. Indent is the driver that lets him work across all of it in one place.
The full loop runs in a single chat: investigate a data question, write code, create a PR, get it
reviewed, and push it. What used to be a multi-tool, multi-step process happens in one
conversation, sometimes from a single human prompt.
Jared's role at Speak spans the full data stack: ingestion, modeling, analytics, experimentation, and strategic analysis. Indent is the driver that lets him work across all of it in one place.
The full loop runs in a single chat: investigate a data question, write code, create a PR, get it reviewed, and push it. What used to be a multi-tool, multi-step process happens in one
conversation, sometimes from a single human prompt.
Jared's role at Speak spans the full data stack: ingestion, modeling, analytics, experimentation, and strategic analysis. Indent is the driver that lets him work across all of it in one place.
The full loop runs in a single chat: investigate a data question, write code, create a PR, get it
reviewed, and push it. What used to be a multi-tool, multi-step process happens in one
conversation, sometimes from a single human prompt.
Jared's role at Speak spans the full data stack: ingestion, modeling, analytics, experimentation, and strategic analysis. Indent is the driver that lets him work across all of it in one place.
The full loop runs in a single chat: investigate a data question, write code, create a PR, get it reviewed, and push it. What used to be a multi-tool, multi-step process happens in one conversation, sometimes from a single human prompt.
Results
Indent gave Speak on-demand analytical support across every team without adding headcount.
With a median response time under two minutes, over 38,700 SQL queries generated, and more than 4,400 data visualizations created, Indent has become the first place people go when they have a data question.
The data team moves faster, the people asking questions move faster, and the questions that matter most get the attention they deserve.
Indent gave Speak on-demand analytical support across every team without adding headcount. With a median response time under two minutes, over 38,700 SQL queries generated, and more than 4,400 data visualizations created, Indent has become the first place people go when they have a data question.
The data team moves faster, the people asking questions move faster, and the questions that matter most get the attention they deserve.
Indent gave Speak on-demand analytical support across every team without adding headcount.
With a median response time under two minutes, over 38,700 SQL queries generated, and more than 4,400 data visualizations created, Indent has become the first place people go when they have a data question.
The data team moves faster, the people asking questions move faster, and the questions that matter most get the attention they deserve.
Hear it from our customers
Hear it from our customers
"Before Indent, getting a quick data answer meant waiting in a weeks-long queue. Now, product managers, growth marketers, and even our finance team can gather and analyze data in minutes. We went from a bottleneck workflow to true self-serve analytics."
Jared Forman, Data Lead at Speak
"Before Indent, getting a quick data answer meant waiting in a weeks-long queue. Now, product managers, growth marketers, and even our finance team can gather and analyze data in minutes. We went from a bottleneck workflow to true self-serve analytics."
Jared Forman, Data Lead at Speak
"Before Indent, getting a quick data answer meant waiting in a weeks-long queue. Now, product managers, growth marketers, and even our finance team can gather and analyze data in minutes. We went from a bottleneck workflow to true self-serve analytics."
Jared Forman, Data Lead at Speak
"Before Indent, getting a quick data answer meant waiting in a weeks-long queue. Now, product managers, growth marketers, and even our finance team can gather and analyze data in minutes. We went from a bottleneck workflow to true self-serve analytics."
Jared Forman, Data Lead at Speak
© 2026 Indent
© 2026 Indent
© 2026 Indent
© 2026 Indent