GitHub’s Copilot for Customer Success has surpassed two million conversations with over a 68% Self Help Success rate. That’s millions of moments where customers adopted best practices faster, unlocked more value from GitHub, and kept moving forward without friction. That’s hundreds of thousands of hours saved, response times improved, tickets avoided, and users empowered immediately with the pinnacle of GitHub expertise.

This milestone started with a question: What if GitHub could share our cumulative human knowledge and Hubber expertise with everyone? A Copilot that could answer questions, surface knowledge, and guide customers in real time? We envisioned thoughtful automation, seamless user experiences, and advanced AI that could solve real customer problems and scale success far beyond what human teams could handle alone. At the time, AI often looked like rigid support tooling or deterministic chatbots. But GitHub’s rapidly growing ecosystem spanning onboarding, architecture, enablement, and support demanded something smarter. As DevSecOps adoption accelerated and customer interactions grew in scale, the demand for GitHub expertise expanded faster than human capacity alone could meet, creating an opportunity to reimagine how we deliver knowledge at scale.

That challenge became our “why.” We set out to design a Copilot that could deliver trusted, contextual answers instantly across surfaces, while enabling our human experts to focus on the moments that require deep expertise. By combining human judgment with AI-driven context, we’ve expanded the reach of our Support teams, Customer Success Managers, Architects, and Services capabilities — proactively and reactively helping more customers, including those who might never have had access to human assistance.

Today, GitHub Copilot for Customer Success is transforming how we drive adoption, growth, and support at scale: surfacing architectural guidance, clarifying billing and licensing, and accelerating resolutions, all while empowering humans to focus where they matter most!

The challenge: Unlocking the value of self help

We’ve built an incredible ecosystem of trusted knowledge spanning GitHub Docs, Support articles, architectural guidance in GitHub Well Architected, adoption playbooks, certification materials, and community discussions. Yet, customers were still struggling to find the right guidance when they needed it most.

At the same time, our support and success teams routinely answered questions that already had great existing solutions — solutions that just weren’t visible at the right time or place.

We knew we could do better.

The solution: GitHub Copilot for Customer Success

We introduced Copilot for Customer Success: a generative AI-powered assistant made up of distinct, purpose-built skills, each tailored to serve a specific customer interaction moment across the GitHub ecosystem. This is a culmination of our best GitHub knowledge and experience brought to all users, with no scheduling or wait times, and in dozens of languages.

Rather than starting with internal workflows, we made the deliberate choice to launch an external, generative AI experience first by embedding Copilot directly into customer-facing surfaces. This allowed us to immediately eliminate categories of work, freeing up our teams to focus on high-value interactions. While internal AI improves productivity, the work still needs to be done, by contrast, external-facing AI helps us remove that work entirely. This shift enables us to scale, accelerate value realization, and ultimately strengthen both customer outcomes and internal efficiency.

To power these intelligent experiences, we leverage advanced capabilities from Microsoft Foundry, including Azure OpenAI for accurate, context-aware generation, Azure AI Content Safety to proactively evaluate responses for groundedness and factual accuracy, and Azure AI Search to rapidly retrieve relevant, authoritative information from trusted data sources. Together, these components help us deliver seamless, trustworthy, and personalized interactions at scale.

Azure OpenAI on Microsoft Foundry allows us to quickly test and provision various AI model deployments. Private networking features enable secure connections to other Azure services and our internal network. The Chat Playground feature provides an interface to quickly test prompts and model behavior. Our system started with an OpenAI gpt-4 deployment, migrated to gpt-4o, and now uses the latest OpenAI model, gpt-5. Internal evaluation tools and multiple model deployments have allowed us to upgrade models with confidence.

To increase the trustworthiness for designated sensitive topics, such as those relating to billing questions, we use Azure Content Safety. Part of Microsoft Foundry, and with the same private networking capabilities, Azure Content Safety provides groundness detection. When provided with the query, source documentation, and AI-generated response, the Azure Content Safety service alerts us to responses that may not truly represent the sources, such a response may be considered ungrounded. This enables us to avoid responding in a way that could mislead a customer and lets us rely on our human experts for questions that exceed current capabilities.

As with all AI systems, the data behind our Copilot experience is one of the most important parts. Being able to quickly retrieve relevant information to augment the inherent knowledge of base AI models provides the context that is needed to answer our customers’ specific GitHub related questions. Azure AI Search provides this retrieval capability and integrates seamlessly with other parts of Microsoft Foundry, such as OpenAI model deployments, while still allowing for a secure private network environment. Multiple indices enable AI responses to derive facts from a tuned mixture of trusted documentation sources. The hybrid search feature of Azure AI Search, which provides vector based similarity search powered by our Azure OpenAI on top of traditional full text search, improves the relevancy of retrieved information and matches content by semantics.

This approach enabled us to solve customer problems directly at their source before they escalate to human assistance, driving faster impact, deeper insights, and continuous learning.

We embedded personalized, contextual GitHub experiences directly into customer-facing touchpoints and products meeting customers exactly where they meet a number of functional needs for our customers across multiple key domains:

Rather than confining AI to a single product, we’ve woven it across surfaces, drawing on GitHub’s deep knowledge and expertise to tailor solutions to each customer moment.

The impact

As of August 2025, GitHub Copilot for Customer Success has powered 2M+ conversations.

✅ 68% Self-Help Success (SHS) ✅ +0.15% Higher CSAT on tickets touched by Copilot

🌟 What this means for GitHub customers:

• Guidance at scale – complex, multi-step support delivered in high-volume product areas helps developers keep moving without disruption.

• Architectural best practices – contextual recommendations aligned with the GitHub Well-Architected Framework accelerate adoption and reduce downstream rework.

• Proactive engagement – tailored, goal-aligned interactions help customers realize value faster and drive adoption of GitHub.

• Faster self-service resolutions – fewer tickets created through the Support contact form, allowing customers to unblock themselves quickly.

• Higher satisfaction and confidence – tickets touched by Copilot drive stronger CSAT outcomes.

• Simplified operations – clear answers on billing, subscription, and licensing reduce friction in managing customer relationships.

• Staying in flow – just-in-time documentation summaries surface the right knowledge when it’s needed most, keeping customers productive.

What’s notable isn’t just the volume, it’s the shift in how customers are solving problems: faster, with more confidence, and without needing to open a ticket.

Built on trusted knowledge

Copilot for Customer Success is powered by continuously updated, trusted sources. To keep the AI current and contextually accurate, we’ve established feedback loops that empower teams to contribute directly to these sources.

Internally, we’ve built systems that enable GitHub engineers to document unique or repeated technical scenarios transforming real-world expertise into structured, trusted knowledge. These contributions are fed back into the AI system and surfaced in relevant skills, ensuring the model reflects the latest customer-facing insights.

What we’re learning

Building successful AI experiences isn’t just about accuracy — it’s about trust, design, and responsibility. Here are the principles that guided us:

1. Trust first Every Copilot response cites trusted sources like GitHub Docs or internal knowledge. Guardrails like hallucination filters, fallback messages, and data boundaries ensure responses stay reliable.

2. Human and AI, better together Engineers act as curators, reviewers, and prompt authors, teaching the AI with every turn. Continuous refinement of prompts, feedback loops, and real-world usage keeps Copilot learning and improving.

3. Responsible by design Privacy, fairness, and transparency aren’t add-ons — they’re built in from day one. We partner closely with Microsoft’s Responsible AI team to uphold safety, transparency, and ethical use, with rigorous model evaluations and user feedback loops baked into the process.

What’s next

We’re just getting started.

Our focus is now on making Copilot for Customer Success even more adaptive, proactive, and agentic, moving from reactive Q&A to intelligent, personalized guidance. This includes:

• Giving customers more confidence in AI-suggested answers

• Supporting dynamic decision paths when the solution isn’t one-size-fits-all

• Surfacing predictive insights before a problem even arises

At the end of the day, our goal is not to replace human interaction, but rather to amplify its impact and scale customer success with precision, empathy, and intelligence.

Final thoughts

AI isn’t replacing our human intelligence: it’s scaling the impact of it.

Copilot for Customer Success helps customers self-solve in moments of need, empowers our internal teams to focus on complex challenges, and continuously improves with every conversation.

We’re excited about how far we’ve come and are just getting started. We’d love to hear your perspective. Go try Copilot for Customer Success and let us know how we can make it even better for you and our teams across GitHub!