Overview

Niles was a Y Combinator Winter 2017 company that built a Slack-native knowledge bot designed to replace the internal wikis that companies create but never use. Founded in late 2016 by Mack Lu and Andy Kim in San Francisco, the product allowed employees to ask questions directly in Slack and receive answers drawn from a continuously updated knowledge base built from existing chat conversations and documents. The core thesis was sound—institutional knowledge decay is a universal, costly problem—but the company shut down around early 2020 after approximately three years of operation, having raised only $120K in total funding and never growing beyond a two-person team. The most likely cause of failure was a combination of technical premature arrival and severe capital constraint: the natural language processing tools available between 2017 and 2020 were not reliable enough to deliver the seamless, accurate knowledge retrieval the product promised, and the team lacked the runway to iterate toward a solution. Niles' story is a case study in correct problem identification paired with insufficient resources and technology to execute the vision.

Founding Story

Niles was founded in November 2016 by Mack Lu and Andy Kim, operating out of San Francisco. ^[1]

Mack Lu brought a strong product management pedigree to the company. Before Niles, he had completed Google's Associate Product Manager program—one of the most competitive product training pipelines in the industry—and had risen to Director of Product at Apartment List, a venture-backed rental marketplace. He studied Computer Engineering at Columbia University. ^[2] This background gave Niles immediate credibility in enterprise productivity tooling and likely opened early doors with potential customers.

The founding insight was personal. Lu described the problem in direct terms on Product Hunt: "I've personally felt the pain of creating wikis that nobody used and struggling to find information across multiple tools, and wanted something better." ^[3] This was not a problem discovered through market research—it was a frustration Lu had lived through at scale, managing product teams at companies where institutional knowledge was constantly being lost to Slack threads, buried Google Docs, and outdated Confluence pages.

The founding insight was also structurally sound. The failure mode of internal wikis is well-documented: they require deliberate, sustained effort to maintain, and that effort competes with the immediate demands of daily work. The result is that most wikis become stale within months of creation. Lu's bet was that a tool embedded directly in Slack—where knowledge was already being created organically through conversation—could capture and surface that knowledge without requiring any additional behavioral change from employees.

Y Combinator accepted Niles into its Winter 2017 batch, providing early institutional validation of the thesis and $120K in seed funding alongside Eudaimonia Capital. ^{[4] [5]}

Andy Kim's specific background, technical role, and eventual departure from the company are not documented in available sources. The YC company profile lists only Mack Lu as founder, and Lu was the sole public-facing representative of the company across press coverage and Product Hunt. ^[6] Whether Kim departed early or remained through the shutdown is unknown.

Timeline

• November 2016 — Niles AI founded by Mack Lu and Andy Kim in San Francisco, CA. ^[1]

• Early 2017 — Niles accepted into Y Combinator Winter 2017 batch; raises $120K seed round from YC and Eudaimonia Capital. ^[5]

• March 17, 2017 — TechCrunch publishes a dedicated article on Niles, describing it as "an alternative to that internal wiki that every company has and no one uses." ^[7]

• March 21, 2017 — Niles presents at YC W17 Demo Day 2; TechCrunch covers it as "a wiki you can talk to in Slack." ^[8]

• March 2018 — Niles pricing tiers documented: Starter ($25/mo), Pro ($125/mo), Premium ($499/mo), Enterprise (custom). ^[9]

• Early 2020 — Mack Lu departs Niles AI and joins Snowflake as Senior Product Manager, indicating company shutdown around this date. ^[10]

• ~2020 — Niles publishes shutdown announcement stating execution was "more difficult than expected" and the team "fell short of our vision." ^[11]

• January 2021 — Latka database records an unverified revenue figure of approximately $196K for Niles (low confidence). ^[12]

What They Built

Niles was a Slack bot that functioned as a living, searchable knowledge base for internal teams. The core user experience was designed around a single principle: employees should be able to get answers to work questions without leaving Slack, opening a separate tool, or remembering where a document was filed. ^[13]

The Basic Interaction

A user would type a question directly into a Slack channel or direct message to the Niles bot—for example, "What is our discount policy for enterprise customers?" Niles would search its knowledge base and return an answer. If no answer existed, Niles would flag the question as unanswered and prompt a team member with the relevant expertise to provide one, which would then be stored for future queries. ^[14]

Knowledge Base Construction

Rather than requiring teams to manually populate a wiki from scratch, Niles was designed to learn from existing Slack activity. The bot could monitor conversations, identify questions that were asked repeatedly, and automatically suggest adding answers to the knowledge base. This addressed the core failure mode of traditional wikis: the blank-page problem, where teams never build the habit of documentation because starting from zero is too effortful. ^[15]

Integrations

Niles connected to Google Drive and Salesforce, allowing it to search across a team's existing document repositories and CRM data in addition to its own knowledge base. ^[16] This expanded the product's value proposition from "Slack wiki" to "unified knowledge search layer"—a meaningfully larger surface area.

Staleness Management

One of Niles' more thoughtful features was its proactive staleness detection. The bot sent weekly notifications to the authors of knowledge base entries, prompting them to verify that answers were still accurate. ^[17] This directly addressed the second failure mode of wikis: entries that are created but never updated as company policies and processes change.

Security Architecture

All conversation logs processed by Niles were stored in an encrypted AWS instance, with data encrypted both in transit and at rest. ^[18] This was a deliberate signal to enterprise buyers who would require data security guarantees before allowing a third-party bot to read their internal communications.

Target Use Cases

The product was positioned horizontally across three primary team types: sales and support (e.g., pricing, objection handling, product specs), product and engineering (e.g., technical documentation, runbooks), and people operations (e.g., HR policies, onboarding materials). ^[19]

What distinguished Niles from a traditional wiki was the elimination of context-switching. Confluence, SharePoint, and Notion all require users to leave their primary work environment, navigate a separate interface, and search using exact terminology. Niles collapsed that workflow into a conversational interface that employees were already using. The bet was that reducing friction to near-zero would solve the adoption problem that plagued every prior knowledge management tool.

Market Position

Target Customers

Niles targeted knowledge workers at companies already using Slack—primarily small to mid-market businesses where institutional knowledge was growing faster than formal documentation practices could keep up. The product's horizontal positioning across sales, engineering, and HR teams suggested the founders were pursuing a bottom-up, team-by-team adoption model rather than a top-down enterprise sale.

The pricing structure reinforced this read. The $25/month Starter tier was accessible to individual teams without budget approval processes, while the $499/month Premium and custom Enterprise tiers indicated an aspiration to expand upmarket once initial adoption was established. ^[9]

The one named customer on record—Apartment List's Director of Sales Operations, Stephanie Soderborg—provided a strong testimonial: "Niles has been the most amazing tool for our sales and operations teams." ^[20] Apartment List was a well-funded startup with a large sales organization, suggesting Niles had at least some success in the sales team use case. Notably, Mack Lu had previously served as Director of Product at Apartment List, which likely facilitated this reference customer relationship.

Market Size

The knowledge management software market was large and growing during Niles' operating years. Enterprise wiki and intranet tools represented a multi-billion dollar category, with Atlassian's Confluence alone generating hundreds of millions in annual revenue by 2017. The adjacent market for team communication and collaboration tools was expanding rapidly, driven by Slack's growth from approximately 1.5 million daily active users in 2015 to over 8 million by 2017.

The Slack bot ecosystem was an emerging distribution channel during this period. Slack's App Directory launched in 2015, and by 2017 it hosted thousands of third-party integrations. Building natively on Slack gave Niles access to a fast-growing installed base without needing to build its own distribution. The risk was dependency: Slack controlled the platform, the discovery mechanism, and the terms of access.

Competition

Niles operated in a crowded adjacency. The direct competitors for knowledge management included:

Traditional wikis: Confluence (Atlassian), SharePoint (Microsoft), and MediaWiki-based tools dominated enterprise knowledge management. These products had deep enterprise relationships and large sales organizations but suffered from the exact adoption problem Niles was trying to solve.

Emerging wiki alternatives: Notion launched in 2016 and Tettra launched in 2015, both targeting the same frustration with traditional wikis. Guru, which raised a $9.3M Series A in 2018 and later a $30M Series B, was perhaps the most direct competitor—a knowledge management tool designed specifically for sales and support teams, with a browser extension that surfaced answers in context.

Other Slack bots: The Slack bot ecosystem included several knowledge management experiments during this period, though none achieved significant scale.

The critical competitive disadvantage Niles faced was capitalization. Guru raised over $40M in total funding, allowing it to build a dedicated sales team, invest in product quality, and sustain customer success operations. Niles raised $120K. ^{[5] [21]} That funding gap made it nearly impossible to compete on product depth, sales reach, or customer support quality.

Business Model

Niles operated on a SaaS subscription model with four pricing tiers as of March 2018: Starter at $25/month, Pro at $125/month, Premium at $499/month, and Enterprise at custom pricing. ^[9] The tiered structure suggests pricing was based on team size or feature access, following a standard SaaS expansion model.

The go-to-market motion appears to have been primarily self-serve, with customers discovering Niles through the Slack App Directory, Product Hunt, and press coverage. The Slack-native design reduced onboarding friction and eliminated the need for a dedicated sales motion at the lower tiers. Enterprise customers would have required direct sales engagement, but with a two-person team, sustained outbound sales was not a realistic option.

To reach $1M in annual recurring revenue at the Starter price point, Niles would have needed approximately 3,333 paying teams. At the Pro tier, roughly 667 teams. No customer count data is available to assess how close the company came to either threshold. The unverified Latka figure of approximately $196K in revenue, if taken at face value, would suggest the company was generating meaningful but pre-scale revenue at the time of shutdown. ^[12]

Traction

Niles received meaningful early press attention. TechCrunch published a dedicated article on March 17, 2017—before Demo Day—describing the product in detail and framing it as a solution to the universal problem of unused internal wikis. ^[7] Four days later, TechCrunch covered Niles again as part of its YC W17 Demo Day 2 roundup. ^[8] Double TechCrunch coverage within a week of Demo Day was a signal of genuine journalist interest in the product concept.

The product also attracted at least one enterprise reference customer. Apartment List's Director of Sales Operations provided a named testimonial, indicating the product delivered measurable value in a real sales organization deployment. ^[20]

Beyond these data points, traction metrics are sparse. The company never grew beyond two employees, ^[6] and no customer count, churn rate, or net revenue retention figures are available in public sources. The $120K total funding figure—representing only the standard YC batch check—is itself a traction signal: investors who reviewed Niles after Demo Day did not fund a follow-on round, suggesting the growth metrics presented during fundraising conversations were insufficient to clear the bar for seed investment.

The Latka-reported revenue figure of approximately $196K should be treated with caution. Latka's data collection methodology for small, private companies is inconsistent, and the figure may represent cumulative lifetime revenue, annualized run rate, or an estimate rather than a verified point-in-time measurement. ^[12]

Post-Mortem

Niles shut down around early 2020, approximately three years after its founding. The company's farewell statement was brief: "While we're proud of what we've accomplished, executing on that was more difficult than expected and we fell short of our vision." ^[11] No detailed post-mortem, founder interview, or investor commentary exists in the public record. What follows is an analysis of the most probable failure causes, ordered by likely impact, drawn from the available evidence.

1. The AI Technology Was Not Ready to Deliver the Core Promise

The central value proposition of Niles was that it could reliably answer questions by understanding natural language queries and matching them to relevant knowledge stored in unstructured Slack conversations and documents. In 2017–2019, this was a genuinely hard problem.

The NLP tools available during Niles' operating years—pre-GPT-3, pre-BERT at scale—were significantly less capable than what became available after 2020. Extracting structured, accurate, searchable answers from conversational text required either extensive rule-based engineering (brittle and expensive to maintain) or machine learning models that needed large amounts of training data to perform reliably. A two-person team without ML engineering resources could not build or maintain either approach at the quality level required to make the product feel magical rather than frustrating.

The founders acknowledged this gap directly in their shutdown statement: "executing on that was more difficult than expected." ^[11] For a product where the core interaction is asking a question and getting a correct answer, accuracy is not a nice-to-have—it is the product. A knowledge bot that returns wrong or irrelevant answers trains users to stop asking questions, which collapses the entire value loop. The team had no documented path to closing this accuracy gap with the resources available.

The subsequent success of AI-native knowledge tools built on GPT-4 and similar models—Glean raised $200M at a $1B valuation in 2022, and Guru relaunched with AI features in 2023—validates that the problem Niles identified was real. The technology to solve it reliably simply did not exist at the scale and cost accessible to a two-person startup in 2017.

2. $120K in Total Funding Created an Insurmountable Execution Constraint

Niles raised $120K across its entire operating life—the standard YC batch check, with no follow-on funding. ^{[5] [21]} In San Francisco, $120K covers approximately three to four months of runway for two founders paying themselves modest salaries, or longer if the founders were not drawing salaries at all.

The company operated for approximately 38 months on this capital. This implies either that the founders were generating enough revenue to sustain operations, were working without meaningful compensation, or some combination of both. None of these scenarios is conducive to the kind of rapid product iteration, sales investment, and customer success work required to build a B2B SaaS company.

The failure to raise a follow-on seed round is the clearest external signal of the traction problem. YC Demo Day in March 2017 put Niles in front of hundreds of seed investors. The fact that none of them wrote a check—or that the founders chose not to pursue additional funding—indicates that the growth metrics at that point did not support a conventional seed round. Competitors in the knowledge management space who did raise meaningful capital—Guru's $9.3M Series A in 2018, Tettra's funding rounds—were able to hire sales teams, invest in product quality, and sustain customer relationships in ways Niles structurally could not.

3. A Two-Person Team Could Not Simultaneously Build, Sell, and Support an AI Product

The company never grew beyond two employees. ^[6] For a product that required continuous AI/ML improvement, active customer onboarding, ongoing customer success to prevent churn, and a sales motion to acquire new customers, two people was not a viable operating configuration.

The specific execution demands of an AI-powered knowledge bot are particularly labor-intensive. Every customer deployment required training the bot on that team's specific knowledge, monitoring answer quality, handling edge cases where the bot failed, and iterating on the model based on feedback. These activities scale with the number of customers, not with the number of features shipped. A two-person team could handle a small number of customers well, but could not scale customer success operations without additional headcount—which required funding that was not available.

The horizontal positioning across sales, engineering, and HR teams compounded this problem. Rather than building deep expertise in one use case—for example, becoming the definitive knowledge tool for sales teams—Niles spread its limited product development capacity across three distinct personas with different workflows, different knowledge structures, and different success metrics. This made it harder to build a product that was excellent for any single customer type.

4. Slack Platform Dependency Created Structural Risk

Niles' entire distribution and product experience was built on Slack's platform. This was a deliberate strategic choice—Slack's rapid growth in 2016–2017 made it an attractive distribution channel, and the Slack-native design was the product's primary UX differentiator.

The risk of this dependency is that Slack controls the terms of access, the discovery mechanism through its App Directory, and the user experience constraints within which Niles had to operate. Any change to Slack's bot API, App Directory ranking algorithm, or competitive strategy could materially affect Niles' business. Slack itself launched Workflow Builder in 2019, which enabled some basic automation and knowledge-sharing workflows natively, reducing the differentiation of third-party bots. While there is no direct evidence that Slack's platform changes specifically harmed Niles, the structural risk of building a company entirely on a third-party platform—without the resources to pivot if that platform shifted—was a meaningful vulnerability.

Key Lessons

• Correct problem identification is necessary but not sufficient. Niles identified a real, durable pain point—institutional knowledge decay—that has since been validated by the success of Notion, Guru, Glean, and AI-native knowledge tools. But identifying the right problem does not guarantee the ability to solve it. The gap between the vision (a wiki that learns from your team) and the technical reality (a bot that sometimes answers questions correctly) was too large to close with the tools and capital available in 2017–2020. Founders should assess not just whether a problem is real, but whether the technology exists to solve it reliably enough to retain customers.

• Failing to raise a follow-on round is a forcing function, not just a setback. Niles' inability to raise beyond the initial $120K YC check meant the company was permanently constrained in its ability to hire, iterate, and compete. When a B2B SaaS company cannot raise a seed round after Demo Day, it is a signal that investors do not see sufficient evidence of repeatable growth. The appropriate response is either to dramatically narrow scope to find a wedge that works, or to wind down before burning founder time on a path that capital markets have already evaluated skeptically. Operating for three years on $120K without a clear path to additional funding is a long time to sustain a structurally undercapitalized position.

• Horizontal positioning delays the discovery of product-market fit. By targeting sales teams, engineering teams, and HR teams simultaneously, Niles spread its limited product development capacity across three distinct use cases. A narrower initial focus—for example, exclusively targeting sales teams with a knowledge tool for pricing, objection handling, and competitive intelligence—would have allowed the team to build deeper value for one persona, generate stronger retention signals, and create a more compelling fundraising narrative. Horizontal positioning is a strategy that requires capital to execute; without it, it becomes a liability.

• Platform dependency requires a contingency plan. Building entirely on Slack's platform gave Niles distribution leverage but created a single point of failure. Companies that build on third-party platforms need either the resources to pivot if the platform shifts, or a plan to develop direct distribution channels over time. For Niles, the Slack-first design was both the product's greatest strength and its most significant structural risk.

• Team size must match the operational demands of the product. An AI-powered knowledge bot requires continuous model improvement, active customer onboarding, and ongoing quality monitoring—all of which scale with customer count. Two founders cannot simultaneously build the product, acquire customers, and deliver the customer success work required to retain them. The team size constraint was both a symptom of the funding problem and an independent cause of the execution gap the founders acknowledged at shutdown.

Sources

1. Tracxn — Niles company profile
2. Clay.earth — Mack Lu profile
3. Product Hunt — Niles launch page
4. Y Combinator — Niles company page
5. TechCrunch — Niles dedicated article, March 17, 2017
6. TechCrunch — YC W17 Demo Day 2 coverage, March 21, 2017
7. Niles.ai — Archived product homepage
8. Niles.webflow.io — Shutdown announcement page
9. Crunchbase — Mack Lu profile
10. ContactOut — Mack Lu profile
11. The Org — Mack Lu at Retool
12. Latka — Niles revenue data
13. PitchBook — Niles AI funding profile
14. Wayback Machine — niles.ai archive history
15. Wayback Machine — niles.ai homepage snapshot, 2018
16. Hongkiat — Meet Niles, Slack's Very Own Chat Assistant