D23 in 2026 So Far: 100 Days of Managed Apache Superset

The First 100 Days: Why This Moment Matters

When we launched D23 as a managed Apache Superset platform in early 2026, we had a straightforward thesis: teams evaluating Looker, Tableau, or Power BI shouldn’t have to choose between enterprise feature richness and the operational simplicity of open-source tooling. The first three months have validated that hypothesis in ways we didn’t entirely anticipate—and challenged us in others.

This reflection covers what we’ve learned from our first cohort of customers, the patterns we’re seeing in embedded analytics adoption, how AI-powered query generation is reshaping what self-serve BI actually means, and where we’re doubling down next.

For context: D23 is a managed hosting and consulting layer on top of Apache Superset, bundled with API-first architecture, MCP (Model Context Protocol) server integration for AI workflows, and hands-on data consulting. We target data leaders at scale-ups and mid-market companies, engineering teams embedding analytics into products, and portfolio companies in private equity portfolios that need standardized BI without the Looker or Tableau overhead.

What We Shipped: Core Capabilities That Resonated

Our initial product launch centered on three pillars: managed infrastructure for Apache Superset, embedded analytics APIs, and AI-assisted query generation via text-to-SQL.

Managed Superset Hosting was the table stakes. We handle provisioning, upgrades, security patches, and scaling—so customers don’t need a dedicated platform engineer just to keep dashboards running. This alone saved our first enterprise customer approximately 0.5 FTE annually compared to self-hosted Superset. The operational lift of maintaining Superset in-house is real; we’ve seen teams spend weeks troubleshooting metadata layer issues or database connectivity problems that our support team resolves in hours.

Embedded Analytics APIs became the unexpected winner. Teams building SaaS products or internal tools realized they could use D23 to power customer-facing dashboards without licensing per-user seats from Tableau or Looker. One customer—a venture capital portfolio tracking platform—embedded five dashboards into their LP reporting portal using our API in under two weeks. At Looker, that same integration would have cost 3–4x more in annual licensing and required 4–6 weeks of professional services.

The embedded model is fundamentally different from traditional BI licensing. Instead of paying per user, you pay for compute and API calls. For products with thousands of end users but relatively stable dashboard definitions, the unit economics flip dramatically in favor of open-source platforms like Superset.

Text-to-SQL and AI Query Generation launched as a beta feature in week six. We integrated an LLM (large language model) layer that lets users ask questions in plain English—“What was our ARR growth month-over-month in Q4?”—and automatically generates Superset SQL queries. This feature resonated most with non-technical stakeholders and business analysts who had previously relied on data engineers to write queries.

The accuracy is currently around 82% on first attempt, with human-in-the-loop refinement pushing that to 95%+. This is lower than what Looker or Tableau claim for their natural language interfaces, but our customers don’t care—they’re comparing it to “ask a data engineer,” not to perfection. The speed and iteration loop matter more than absolute accuracy in early-stage adoption.

Customer Wins: What Actually Closed Deals

We closed eight enterprise-tier customers in the first 100 days. Here’s what moved the needle.

Win 1: Scale-up with embedded product analytics. A B2B SaaS company with 200+ enterprise customers needed to embed usage analytics into their platform. Looker would have cost $50K+ annually in licensing plus 6–8 weeks of implementation. D23 came in at $8K annually for managed hosting plus embedded APIs, with go-live in two weeks. The decision was financial, not philosophical—they care about Superset as much as we do. They care about time-to-value and cost.

Win 2: Private equity portfolio standardization. A mid-market PE firm with 12 portfolio companies wanted a single BI platform for KPI reporting and value-creation dashboards across all companies. Power BI licensing was expensive at scale; Tableau required each company to maintain its own instance. D23 offered a multi-tenant architecture where each portfolio company has isolated dashboards and datasets, but the PE firm’s investment team has a consolidated view across all metrics. This deal showed us that managed platforms matter most when you have multiple teams that need BI but lack the infrastructure to self-host.

Win 3: Data consulting + platform. A fintech startup hired us not just for managed Superset, but for hands-on help designing their metrics layer, building dbt (data build tool) transformations, and training their analytics team. The platform itself was table stakes; the consulting was the differentiator. This win reinforced that data leaders don’t just want a tool—they want a partner who understands their domain.

Win 4: API-first architecture for internal tools. An engineering team at a logistics company was building an internal dashboard application using React. They needed a BI backend that exposed clean APIs rather than forcing engineers to reverse-engineer Superset’s UI. Our API-first approach meant they could build custom UX while we handled the query engine, caching, and metadata layer. This opened a new segment we hadn’t initially targeted: engineering teams that want BI capabilities but need to control the frontend.

Lessons Learned: Where Our Assumptions Were Wrong

We entered the market with several hypotheses. Some held up. Others didn’t.

Assumption 1: Customers would migrate from Tableau or Looker immediately if we were cheaper.

We were wrong. Cost alone doesn’t drive migration. Switching costs are real—existing dashboards, user training, integration rework, organizational inertia. We’ve had conversations with companies paying $200K annually for Looker who are genuinely interested in D23 but can’t justify the six-month migration project internally. What does work: positioning D23 as the platform for new analytics initiatives, not as a replacement for existing BI infrastructure. If you’re building embedded analytics or standing up BI from scratch, the math is compelling. If you’re already deep in Looker, the friction is too high.

Assumption 2: Self-serve BI would be the main use case.

We were partially wrong. Self-serve BI matters, but it’s not the primary driver. Our customers care more about embedded analytics (dashboards in products), standardized reporting (KPI dashboards across teams), and API-driven analytics (programmatic access to query results). Self-serve BI—the idea that business analysts can freely explore data without engineering—is real, but it’s downstream of solving the “how do we get BI into the hands of non-technical users at all” problem.

Assumption 3: Text-to-SQL would be a premium feature that only advanced users would adopt.

We were wrong. Non-technical stakeholders adopted it faster than we expected. The business folks who previously had to wait for data engineers to write queries suddenly had agency. The constraint shifted from “can I ask the question” to “is the LLM accurate enough.” We’ve had customers use text-to-SQL to prototype dashboards in hours rather than weeks.

Assumption 4: Open-source positioning would resonate with enterprise buyers.

Partially wrong. Enterprise buyers care that Superset is open-source—it means no vendor lock-in, no surprise licensing audits, and the ability to self-host if we go out of business. But they don’t care about open-source as a philosophy. They care about outcomes: faster time-to-dashboard, lower cost, better API support. Open-source is a means, not an end.

The Competitive Landscape: Where We Stand

After 100 days, we have a clearer view of how D23 fits in the market.

Versus Preset (Superset’s original commercial offering): Preset is the “pure Superset” play—hosted Superset with minimal additional features. D23 differentiates on embedded analytics APIs, text-to-SQL, and hands-on data consulting. Preset is stronger for teams that want vanilla Superset in the cloud. We’re stronger for teams that need BI integrated into products or processes.

Versus Looker and Tableau: These are still the market leaders. They have better UX, deeper integrations, and more mature AI features. But they’re also expensive and heavy. Our pitch isn’t “we’re better”—it’s “we’re lighter, cheaper, and open-source.” This works for teams that don’t need Looker’s advanced semantic layer or Tableau’s visual polish.

Versus Metabase and Mode: Metabase is simpler than Superset and cheaper than us. Mode is Superset-adjacent but focused on SQL exploration. Both are weaker on embedded analytics and lack the consulting layer we offer. Metabase is a real competitor for small teams; Mode is less of a threat in our target segment.

Versus building custom BI on top of Postgres or Snowflake: This is our biggest threat. Some engineering teams think, “Why pay for a BI platform at all? We’ll just build dashboards using Grafana or a custom React app.” This is rational for teams with strong engineering resources. Our answer: you’ll spend 6–12 months building what Superset gives you in weeks. The platform leverage is real.

The AI Opportunity: Text-to-SQL and Beyond

Our text-to-SQL feature has taught us a lot about where AI actually helps in analytics.

The core insight: AI is most valuable when it reduces friction for non-technical users to ask questions they already know how to ask. A business analyst who knows SQL can write queries. An executive who doesn’t know SQL can ask in English. The LLM bridges that gap.

Where text-to-SQL struggles: novel questions that require domain knowledge or complex logic. If an executive asks “What’s driving the variance in our gross margin,” the LLM can’t answer that without understanding your business model, cost structure, and historical context. A human analyst still needs to be involved.

Where text-to-SQL shines: routine, templated questions. “Show me revenue by region for the last 12 months.” “What’s our customer churn rate?” “List all deals closed in Q4.” These questions are asked repeatedly. Text-to-SQL lets you automate the query generation and let non-technical users self-serve.

We’re seeing adoption patterns that align with this. Teams using text-to-SQL most effectively are those that have already standardized their metrics—they have a dbt project with well-documented models, clear naming conventions, and a documented business glossary. The LLM can then map natural language to these models reliably.

Our next iteration is integrating MCP servers for analytics—a standardized protocol that lets LLMs interact with your BI platform as a tool. Instead of just generating SQL, the LLM can execute queries, interpret results, and suggest follow-up questions. This is where the real leverage is. You move from “LLM as query writer” to “LLM as analytical assistant.”

Infrastructure and Scaling Lessons

Managing Superset at scale has taught us things that aren’t obvious from the open-source docs.

Database connection pooling is critical. Early on, we had a customer whose dashboards were creating hundreds of database connections per minute. This overwhelmed their Postgres instance. The fix: implement connection pooling in the Superset metadata layer. This is table stakes for any managed Superset offering, but it’s not obvious from the vanilla open-source setup.

Query caching is asymmetric. Some dashboards get viewed thousands of times per day; others are viewed once monthly. Caching strategies that work for one don’t work for the other. We’ve built tiered caching—in-memory for hot queries, Redis for warm queries, database-direct for cold queries. This reduced average query latency from 2.3 seconds to 0.6 seconds for our most-used dashboards.

Metadata layer governance matters early. As customers add more datasets, the metadata layer becomes a bottleneck. Teams need clear naming conventions, ownership, and deprecation policies. We’ve started offering metadata governance consulting as part of our onboarding—helping teams structure their datasets so they scale.

Security and multi-tenancy are harder than they look. Superset’s native multi-tenancy is good, but it’s not designed for SaaS-level isolation. We’ve had to add additional row-level security (RLS) logic, encrypt sensitive query results, and implement audit logging. This is the kind of work that doesn’t show up in marketing materials but is essential for enterprise adoption.

What We’re Doing Next: The 100-Day Roadmap

Based on customer feedback and our own learning, here’s where we’re investing in the next phase.

1. Semantic Layer Improvements. Superset’s semantic layer (the translation from user-facing metrics to database queries) is functional but not as polished as Looker’s or Tableau’s. We’re investing in a cleaner UX for defining metrics, dimensions, and measures. We’re also building connectors to dbt Cloud so teams can define metrics in dbt and automatically surface them in Superset.

2. Advanced AI Features. Beyond text-to-SQL, we’re building anomaly detection (“alert me when revenue dips below forecast”), automated insights (“your churn rate increased 2% this month—here’s why”), and predictive analytics (“based on current trends, here’s next quarter’s forecast”). These are LLM-powered features that go beyond query generation.

3. Embedded Analytics Marketplace. We’re building a library of pre-built dashboards and charts for common use cases—SaaS metrics, e-commerce KPIs, portfolio performance tracking. This lets customers get to value faster and gives us a channel to showcase what’s possible with D23.

4. Deeper Data Consulting. We’re expanding our consulting team to help customers with metrics definition, data modeling, and dbt implementation. The platform is the delivery mechanism, but the consulting is where we add the most value.

5. MCP Server Standardization. We’re investing heavily in MCP (Model Context Protocol) integration—both as a server that exposes Superset to LLMs and as a client that lets Superset interact with other AI tools. This is where the real AI-native BI future is headed.

The Competitive Advantage We’ve Built

After 100 days, here’s what we think actually differentiates D23 in a crowded market.

Open-source foundation. Superset is mature, battle-tested, and free. We don’t have to build a query engine from scratch. We inherit years of open-source development and can focus on the management layer and AI integration.

API-first architecture. Unlike Tableau or Looker, which treat APIs as an afterthought, D23 is built with APIs as a first-class citizen. This makes embedded analytics natural, not forced.

Hands-on consulting. We’re not just a platform vendor. We’re a consulting firm that happens to sell a platform. This means we understand the hard part of BI—not the tool, but the metrics, data modeling, and organizational change management.

AI integration from day one. We’re not bolting AI onto an existing product. We’re building AI into the core workflow—text-to-SQL, anomaly detection, automated insights. This gives us a 12–18 month advantage over competitors playing catch-up.

Cost structure. Our pricing is transparent and predictable. You pay for compute and API calls, not per-user seats. For teams with many users or embedded use cases, this is 3–10x cheaper than Looker or Tableau.

Lessons for Data Leaders Evaluating BI Platforms

If you’re reading this and evaluating BI options—whether D23 or otherwise—here’s what we’ve learned matters.

Define your use case first. Self-serve BI, embedded analytics, and standardized reporting have very different requirements. A platform great for one might be terrible for another. We initially thought we’d be strong at all three; we’ve learned we’re strongest at embedded and standardized reporting.

Evaluate the total cost of ownership, not just licensing. Looker costs $50K annually, but it might cost $30K in implementation and $20K annually in maintenance. Superset costs $0, but self-hosting costs $100K in engineering time. D23 costs $8K annually, but it might cost $15K in consulting. The math is complex, and most teams get it wrong.

Assess your data maturity. If you don’t have a clean data warehouse, a good data model, and documented metrics, no BI platform will save you. Fix the data layer first. The platform is downstream.

Consider your growth trajectory. If you’re growing 100% annually, your BI needs will change dramatically. A platform that works for 10 dashboards might break at 1,000. Superset’s architecture scales; Metabase’s doesn’t. This matters if you’re planning for multi-year growth.

Think about integrations. Your BI platform will need to talk to your data warehouse, dbt, your product analytics tool, your CRM, and your data catalog. Looker has deep integrations; Superset has fewer, but the API-first approach means you can build them yourself.

The Road Ahead: What We’re Thinking About

As we move into the next 100 days, we’re thinking about several big questions.

Can we build a BI platform that’s genuinely AI-native? Most BI tools treat AI as a feature. We want to build a platform where AI is the foundation—where the primary interface is natural language, and SQL is the fallback. This is a different product than Superset+AI. It’s a different product than Looker+Copilot.

How do we compete on UX without rebuilding Superset’s entire frontend? Superset’s UX is functional but not delightful. Tableau’s is delightful but expensive. Can we find the middle ground? We’re experimenting with UI customization and white-label options so customers can build the UX they want on top of our backend.

What’s the right pricing model for embedded analytics? Per-user pricing doesn’t work. Per-query pricing creates perverse incentives (teams optimize for fewer queries rather than better insights). Per-compute pricing is transparent but unpredictable. We’re still figuring this out.

How do we scale consulting without becoming a services company? We want to help customers succeed, but we can’t hire consultants at the rate we’re signing customers. The answer is probably a mix: hands-on consulting for enterprise deals, community support for mid-market, and self-service resources for everyone else.

Reflections: What We Got Right and Wrong

If I’m being honest about the first 100 days, here’s what stands out.

We got the market timing right. There’s a real shift happening away from expensive, monolithic BI platforms toward lighter, API-driven, open-source alternatives. We’re riding that wave, and it’s real.

We underestimated the importance of data consulting. We thought the platform would be the main value prop. It’s not. The consulting—helping teams define metrics, build data models, implement dbt—is where we actually move the needle for customers.

We overestimated how much customers care about open-source. Open-source is a feature, not a benefit. Customers care about outcomes: faster dashboards, lower cost, no vendor lock-in. Open-source enables those outcomes, but it’s not why they buy.

We got embedded analytics right. The product-analytics use case is real and underserved. Teams are tired of licensing Amplitude or Mixpanel. They want to embed analytics into their products. Superset + our API layer is a compelling alternative.

We’re still figuring out AI. Text-to-SQL is cool, but it’s not transformative. The real opportunity is in using AI to help teams ask better questions—anomaly detection, automated insights, predictive analytics. We’re building toward that, but we’re not there yet.

Conclusion: The Next 100 Days

We’re three months into D23, and we’re more convinced than ever that there’s a massive market opportunity at the intersection of open-source BI, embedded analytics, and AI-powered query generation.

Our first 100 days have been about learning: learning what customers actually need (not what we thought they needed), learning where Superset is strong (embedded analytics, standardized reporting) and where it’s weak (UX, semantic layer), and learning that the real value we add is not the platform—it’s the consulting, the architecture guidance, and the operational excellence.

The next 100 days are about scaling: scaling our customer base, scaling our consulting team, and scaling our AI capabilities. We’re hiring engineers, consultants, and customer success folks. We’re investing in semantic layer improvements, advanced AI features, and the MCP infrastructure that will let Superset integrate deeply with AI workflows.

If you’re a data leader evaluating BI platforms, we’d love to talk. Visit D23 to learn more about how managed Apache Superset can power your embedded analytics, self-serve BI, and AI-driven insights. If you’re an engineer building analytics infrastructure, our API-first architecture might be exactly what you’re looking for. And if you’re a consultant or implementation partner, we’re actively looking for partners to help us scale.

The first 100 days have been validation. The next 100 days are about building the future of open-source BI at scale.

---

Building Your BI Stack: Key Takeaways

As we reflect on the first 100 days of D23, here are the core insights for teams evaluating managed analytics platforms:

Platform selection is downstream of data maturity. Before you evaluate Superset, Looker, or Tableau, make sure your data warehouse is clean, your data model is sound, and your metrics are well-defined. The best BI platform can’t fix bad data.

Embedded analytics is a different product category. If you’re building a product that needs analytics, you’re not buying Looker—you’re building with Superset or a similar open-source foundation. The licensing, architecture, and go-to-market are completely different.

AI in BI is still early, but the opportunity is real. Text-to-SQL is the current frontier, but the future is AI-assisted analytics—where the system proactively surfaces insights, suggests questions, and helps teams avoid analytical mistakes.

Total cost of ownership matters more than licensing cost. A $50K/year Looker implementation might be cheaper than a $0 Superset self-hosted solution if you factor in engineering time. Do the math carefully.

Consulting adds more value than the platform itself. This is the insight we’ve been most surprised by. The platform is table stakes, but the expertise—in metrics definition, data modeling, and organizational change management—is where we actually move the needle for customers.

We’re excited about the next phase of D23 and grateful for the customers who’ve trusted us with their analytics infrastructure. The first 100 days have validated our thesis. The next 100 days are about scaling it.

For more information about how D23 can power your analytics, visit our homepage or reach out directly. We’re here to help you build the BI infrastructure that scales with your business.

Our Terms of Service and Privacy Policy outline the details of how we operate and protect your data. But the core principle is simple: we’re here to make Superset work at enterprise scale, and we’re committed to your success.