The next-gen Moveworks Copilot: An inside look

When ChatGPT burst onto the scene about a year ago, it sent shockwaves through the tech industry. Its human-like conversational abilities went far beyond anything seen before. 

At Moveworks, we recognized the immense potential to transform the employee experience. Rather than taking an incremental approach and bolting generative AI to existing systems, we opted for a clean-slate redesign. We leveraged cutting-edge generative models to build a copilot capable of not only holding fluent dialogues but also solving complex enterprise challenges end-to-end.

After intensive development over the past ten months, we are proud to reveal our new Copilot. The Moveworks Copilot stands at the forefront of the next generation of enterprise AI. Powered by groundbreaking advances in generative AI, this solution fundamentally transforms how employees get work done. At its core sit powerful generative AI models capable of reasoning, planning, and natural language. Powered by specialized tools and integrations, the result is a beautifully simple solution to previously messy problems. 

In this post, I'll dive into how we brought this next-gen copilot to life. From rethinking our foundational architecture to optimizing workflows with plugins, Moveworks has set the standard for the future of work. The possibilities are truly game-changing. Read on to learn: 

• The key enterprise support challenges employees face
• The three principles of the Moveworks Copilot
• The re-architected building blocks of the Moveworks Platform

Overcoming key enterprise support challenges

We know many love our copilots. Over three hundred unique enterprises representing over five million knowledge workers use Moveworks worldwide. But, after more than seven years of learning from our customers, we can now solve once-intractable support challenges thanks to our generative AI-powered copilot. Let's look at a few key challenges:

34% of issues are described with three words or fewer

Today, 34% of employee requests contain just three words or fewer, like "can't login" or "wifi down." It's tough to take action on such sparse information. Our next-gen copilot is designed to have a natural, engaging dialogue to gather the context it needs to resolve issues quickly.

43% of answers return multiple knowledge articles

We see 43% of questions generate multiple relevant support articles from knowledge bases and systems. While we have done a good job compiling information across systems in the past, employees must still read through options. Now, our copilot can synthesize details and provide focused answers.

18% of requests require multiple actions

18% of requests require completing multiple sequential actions across systems. For instance, onboarding new hires involves many cross-functional tasks. With generative AI, our copilot can break these complex workflows down and execute each step automatically.

By leveraging best-in-class AI, our next-gen copilot can overcome previous limitations by having natural conversations, synthesizing disparate data, and completing multi-step processes — all to enable exceptionally fast, frictionless support for employees.

The three principles of our next-gen copilot

With these core challenges in mind, we identified three key principles to transform the employee experience:

1. Engaging experience: We set out to make every interaction more engaging and intuitive than ever before. With conversational AI advances, we can deliver an experience that employees enjoy using daily. This high engagement leads to more questions asked, more tasks completed, and more problems solved.
2. Faster outcomes: Employees turn to our copilot when they need to quickly get unstuck and back to work. By streamlining dialogues, reducing steps, and leveraging powerful AI capabilities, we can dramatically accelerate task completion, meaning every interaction should be as frictionless as possible.
3. Limitless possibilities: Our goal is to remove any artificial constraints on what our copilot can do so that we can support virtually unlimited use cases across systems and teams. No matter the workflow challenge or business system, our copilot will rise to the occasion. And with generative AI at the core, it will continuously expand its knowledge and skills.

To realize this vision, we completely reimagined our architecture. With dynamic follow-up questions, our copilot efficiently gathers context for quick issue resolution. Summaries with citations to sources provide ease-to-consume, verifiable answers. And robust search combined with infinite cross-system actions allows handling even the most complex requests seamlessly.

By ground-up redesign focused on employee experience, we've built a next-gen copilot that delivers tremendous value instantly. We're thrilled to finally realize the full promise of frictionless support.

The building blocks of the Moveworks Copilot

Placing these principles as our North Star, I'd like to outline the crucial building blocks that enable our copilot to deliver this seamless, frictionless experience.

Reasoning

Reasoning is the brain powering our copilot's advanced capabilities. It mimics human cognition to tackle complex challenges that stump traditional, transactional bots. This way, our copilot can strategize solutions and plot the best course of action, just like a support agent might.

When an employee poses a question or request, reasoning kicks into gear. This engine reviews the inquiry and begins an internal dialogue to determine the optimal way forward, much like a person working through a problem. It taps knowledge resources to understand context and history. It engages in conversation with the user to clarify ambiguities in the original request. And it interfaces with various systems to orchestrate actions across enterprise infrastructure.

Planning

Planning is the set of steps that the brain devises to complete the task at hand. It's the project plan laid out to take the conversation through to resolution.

Once reasoning has worked out the optimal approach to the employee's request, planning maps out each specific step required to get there. This could involve clarifying questions to ask, systems to search, actions to take, and responses to provide. 

Planning carefully sequences each of these steps, ensuring they build off each other effectively and defining any logic or conditions needed to handle different scenarios along the way. Without planning, you're just hoping for the best. This building block grounds our copilot by providing a clear roadmap to the solution, removing guesswork.

By leveraging reasoning to strategize and planning to map out details, our copilot can tackle multifaceted issues smoothly and systematically. The combination of advanced cognition and meticulous planning enables truly intelligent conversations and automation.

Plugins

The next step is execution — and that’s where plugins come in. Plugins are specialized components that actually carry out the planned actions.

You can think of plugins as the hands that do the work. Just as hands have unique fingerprints, each Plugin is custom-built to handle specific use cases and integrate with certain systems. For example, plugins may reset passwords in Okta, provision users in Google Workspace, onboard staff in BambooHR, manage access in AccessHub, deploy servers, update Salesforce, pull reports from analytics platforms, and sync product data between ERP systems. The list could go on.

The magic happens when plugins work together. Like a skilled chef using multiple utensils, our copilot combines plugins to unlock tremendous possibilities. They let us automate intricate processes across the systems you already use with Moveworks.

Plugins are the powerhouse that turns thoughtful plans into tangible actions. With these specialized components at our disposal, we can resolve nearly any request via seamless integrations. And the best part — no reconfiguration needed. Plugins leverage your existing setup.

Clarifications

Employees rarely provide every detail upfront. Clarifications are critical when the initial request lacks enough context to determine a solution.

Our copilot proactively seeks clarification to enrich vague requests. It can have natural language conversations with employees to ask thoughtful follow-up questions, or it can search integrated systems to uncover additional helpful information.

Clarifications enable our copilot to dynamically gather the complete picture so it can craft an accurate response. Rather than make assumptions or ignore gaps, it leverages two-way dialogue and system data to fill in the blanks.

Clarifying vague, brief requests allows our copilot to better offer customized solutions. The process of adding more context gives the copilot more information to develop strategies and more specifics to take focused action. Asking for clarification turns unclear questions into clear answers.

Summaries

Summaries are the final responses our copilot delivers after executing the action plan. 

Without generative AI, you end up with static, rigid responses — just copying and pasting text. Summaries empower our copilot to synthesize dynamic answers in an easy-to-digest way.

The copilot analyzes all the information gathered, including clarifications, system data, and actions taken. It then uses its generative capabilities to summarize the most important details in simple language. 

Then, the copilot can tailor the response to that specific employee's original question and context. They get a summarized explanation that is personalized, focused, and human-friendly.

Summaries unlock the true potential of AI to have natural conversations. Employees receive crafted responses vs. chunky copied text. 

Citations

Citations are a critical complement to summaries. They list where the summarized information is coming from —be it knowledge base articles, system data, or other sources.

Citations serve the vital purpose of establishing trust and credibility. Employees can instantly see the responses aren't fabricated — they're grounded in real corporate information and facts.

This verification gives employees confidence in the accuracy of the summaries. If our copilot users ever want to dig deeper or learn more, they can trace back to the original sources.

Together, summaries and citations provide the best of both worlds: synthesized responses that are easy to consume yet fully transparent. Citations unlock the ability for generative AI to provide constructive, trustworthy responses. Employees get AI's advantages without losing human oversight and validation. This balancing act is essential to user adoption and continued reliance on virtual support.

Now, let's examine how we architected a copilot capable of such sophisticated reasoning and execution in the first place.

Re-architecting the Moveworks Platform to power the future of work

In the past, many solutions’ conversational AI relied on rigid dialogue flows and pre-mapped scenario planning. But this approach doesn't scale — you can't predict the infinite number of ways someone might ask a question.

Unlike these many solutions, Moveworks has never relied on rigid dialogue flows. From the beginning, we recognized the limitations of predefined paths for handling dynamic conversations. Instead, we designed our system to understand context and user intent.

With the rise of tools like ChatGPT, we’ve evolved our architecture even further with generative AI. And today, by integrating state-of-the-art AI techniques, the upgraded Moveworks Platform delivers even better performance. 

At the core of the Moveworks Platform are two engines — a generative reasoning engine powered by LLMs and a robust action engine. Together, they dissect requests, communicate with employees and systems, and devise custom game plans.

But crafting a plan is only half the battle — you must act. That's where our plugins come in. They're the power behind turning thought into action. Plugins provide a vital connection between strategic reasoning and real-world execution. They enable natural language conversations and orchestrate interactions across systems via APIs.

Not all plugins are created equal, though. Ours are meticulously crafted for specific use cases and finely tuned for precision. They feature remarkable interoperability to collaborate across dynamic, multi-system environments.

The synergy empowers the reasoning engine to make informed decisions, while plugins enable seamless execution. By re-architecting with generative AI, we've built a copilot that handles the most complex requests even better than before.

Built for the unique demands of the enterprise

Designing a new copilot architecture is one thing, but making it work seamlessly for enterprises brings unique demands. Factors like accuracy, security, scalability, and compliance become essential.

We emphasized four core attributes in our architecture to enable successful enterprise adoption:

• Extensibility: Our modular design allows easy expansion to new use cases by leveraging mix-and-match plugins. This limitless flexibility ensures we can automate any workflow.
• Steerability: Advanced controls give admins visibility into how the copilot operates and the ability to guide its actions, ensuring alignment with policies and governance.
• Interoperability: Seamless integrations with hundreds of HR, IT, and other systems unlock end-to-end automation. We meet enterprises where they are.
• Verifiability: Detailed audit trails and citations provide transparency into the copilot's decision-making process, which is critical for verification and compliance.

By focusing on these pillars, we've built a copilot purpose-made for the complexities of enterprise. Our copilot doesn't just perform — it works wonders in the most demanding enterprise environments.

Extensibility through modular plugins

A key to unlocking our copilot's full potential is designing for limitless extensibility, as enabled through our flexible plugin architecture.

Plugins are mix-and-match modular components that extend functionality. We orchestrate seamless plugin selection and invocation to activate the right capabilities for each request. Out of the box, we offer 35 plug-and-play plugins covering common needs. And we're continually developing more — over 100 are in the works.

Even better, these plugins leverage all the same systems and integrations we've built over the last seven years. For current customers, expanded automation is instant. And the possibilities are endless. Our customers have thousands of unique APIs we can harness. With Creator Studio, you can build custom plugins that elegantly interface with our reasoning and action engines.

Our extensible plugin architecture is key. It allows us to start automating workflows of any type right away. And flexibility to constantly expand ensures we can continue tackling novel use cases as they emerge.

The sky's the limit when it comes to what our copilot can automate. Our modular design future-proofs companies to adapt at the pace of business change.

Steerability for alignment

While our reasoning engine is powerful, it's crucial that companies can steer its decisions to align with business needs. Steerability ensures outcomes map to specific policies and objectives.

For instance, companies may want to guide the copilot's responses to use approved vocabulary and styling. Or strictly adhere to set permissions and access controls. Compliance with governance processes is also key.

That's why we've implemented extensive steering capabilities and a security framework giving full control:

• Fine-tuning shapes how the copilot interprets requests and crafts responses
• Comprehensive permissions and access policies constrain its actions
• Oversight modes, including approvals, give humans final sign-off
• Explainability tools provide visibility into decision-making

With robust steering, companies can leverage AI's advantages while ensuring perfect alignment. Our copilot does the right things in the right way for the right reasons.

Interoperability for unified automation

Our plugins don't operate in isolation — their interoperability enables unified automation across systems. They fluidly collaborate to execute multi-step processes.

For example, our search plugin aggregates data from diverse sources while respecting permissions. A lookup plugin retrieves details from the HRIS. An action plugin provisions access from the ITSM. This interplay powers end-to-end workflows so that requests involving multiple questions, issues, and actions can be resolved seamlessly by the copilot instead of by support agents. With bi-directional orchestration between plugins and the reasoning engine, they can dynamically exchange information to handle evolving situations.

The reasoning engine decides which plugins to invoke based on the context. It's like having a toolbox where you can grab the right tools and use them together to build something powerful.

Interoperability is key to automating convoluted business processes. Point solutions only go so far. By enabling unified automation across systems, we deliver the end-to-end resolution employees need.

Verifiability for trust

For employees to trust and rely on AI assistance, verifiability is essential. Our copilot grounds responses in given enterprise data sources — it verifies that its synthetic answers are supported by the provided information rather than hallucinating responses.

Every summary provided cites the underlying sources, whether knowledge articles, CRM records, or IT systems. This approach lets users instantly validate the veracity of information.

We also link structured entities like names, addresses, and dates back to the originating systems so employees can verify these details if desired.

And our advanced search capabilities ensure we're summarizing from comprehensive, relevant data — the bedrock for accurate answers.

With full verifiability, employees know our copilot's responses are grounded in reality. They won't hesitate to adopt AI support, knowing its information is concrete versus hypothesized.

This transparency is vital for trust. Employees will only use an AI assistant confidently if they can understand and validate its reasoning from end to end.

In addition to these four pillars, our architecture provides two more vital advantages for enterprise adoption.

Flexible LLM architecture

Though our platform has undergone a major redesign, it is architected with the same philosophy as always — use the right models for the right jobs. The AI landscape evolves rapidly, so we must build for agility.

We leverage not only large language models (LLMs) from OpenAI, Google, Microsoft, and more but also our own proprietary MoveLM™, purpose-built for enterprise assistance. MoveLM excels at understanding business context, following instructions, and crafting workplace-appropriate responses. And it's optimized through clean, synthetic enterprise data.

This flexible design means we can constantly incorporate the latest, greatest LLMs. As models advance, so will our copilot while maintaining security and stability.

We do not use customer data to train MoveLM

We are pleased to announce we do not use customer data to train MoveLM. While we have customer-specific models that use customer-specific data, MoveLM itself is powered by high-quality synthetic enterprise data.

This was a deliberate decision to make our platform as secure as possible. MoveLM provides the core intelligence about workplace environments, while customer data provides specific personalization.

The combination enables a copilot maximally attuned to each organization without compromising sensitive information. Data security has always been our priority. This architecture reflects that commitment.

Security

We've earned the trust of the world's largest, most reputable companies because we meet and exceed their rigorous security standards. Our commitment to security is unwavering. We adhere to an extensive set of standards and hold numerous certifications, compliance standards, and privacy standards, including but not limited to:

• ISO 27001
• SOC 2 Type 2
• CCPA
• HIPAA
• GDPR

We ensure the complete protection of customer data and operations, implementing controls including but not limited to:

• Encryption of data in transit and at rest
• Access controls and permissions
• Comprehensive auditing

Our architecture and policies enable full customer ownership of data. By prioritizing enterprise-grade security from the start, we've created an AI platform built on an unshakeable foundation of trust. Our customers feel fully confident entrusting us with their most sensitive data.

This is the just beginning.

The Moveworks Copilot is already delivering a remarkable impact across our customer base. Yet, this is merely the beginning of our journey.

Up to this point, we've focused on empowering the use cases you know and love. But the real innovation lies ahead in the coming months as we build upon this foundation.

With our architecture, enhancements will compound, creating a flywheel effect. Each additional capability will unlock even more transformative power across your business.

We will tackle more complex workflows, integrate with more systems, and handle more multifaceted requests. The more we automate, the more value we will bring.

This blog offers just a preview of what's to come. We are dedicated to continuous innovation as we optimize and expand functionality over time. The magic happens when all the pieces come together into one frictionless, unified experience.

This next-gen copilot marks a new era of employee support and workflow automation. We are thrilled to partner with you as we realize its immense promise in the months and years ahead. The future is bright.

See how employees can use one conversational interface to surface information and take action across every enterprise system. Get a live product tour of Moveworks.