Azure AI Marks a New Era of Enterprise Intelligence
Unlocking New Horizons in Business Analytics and Decision-Making
C. J. Kozarski, Senior Solutions Director of Application Services
6 Min Read
Prompt-based AI language models have burst to the forefront of conversations across all industries. The capacity of this generative intelligence to synthesize output from vast indexed open/close datasets based on natural language requests has never been seen before. Both personally and professionally, users are just starting to scratch the surface of its capabilities.
The launch of OpenAI’s ChatGPT, followed by Microsoft’s deployment of Azure AI featuring OpenAI’s models, marks the beginning of the era of genuine Enterprise Intelligence.
While generative intelligence is in its nascent stages as a consumer-grade cloud service, it has never been easier for organizations to create, deploy, and maintain their own ChatGPT-like experience. Organizations can easily and cost-effectively train these Large Language Models (LLMs) to integrate their specific institutional or company knowledge and capabilities.
Bring Prompt-Driven AI to Your Business
In the coming months, the concept of Copilots will become more prevalent in the Microsoft ecosystem.
Microsoft has been actively involved in the development of various AI and machine learning tools and integrations across its product suite, particularly within the realms of Azure AI, Microsoft 365, and others. Copilot is a front-end interface that presents the prompt-driven user experience to and from Azure AI. Whether you elect to use the built-in Power Platform widgets and Teams modules or create a completely custom experience, the underlying cloud infrastructure and architecture remains the same.
These tools and services are now consumer-grade and can be enabled with the click of a button. Before, you would need a multi-million-dollar data science and machine learning team that would develop and maintain a custom-coded solution. This shift brings about a new paradigm: your knowledge base is now your source code that drives your AI.
Whether you’re merging the basic collection of disjointed company wikis or developing extensive domain intelligence across various segments, the functionality of generative intelligence now depends entirely on the knowledge you train it with; Azure AI will handle everything else.
Consider this scenario: Instead of an expert spending exuberant amounts of time answering countless questions that flood their mailbox or chat client, imagine if they could allocate that time working with your enterprise intelligence team to fine tune and train a segment Copilot that can interface with client instead.
For example, you have a complex system(s) that have years of development and updates captured in documentation scattered around emails, Agile boards, SharePoint sites, etc. When an employee has a question on how a business process works, someone needs to exhume that information and pass it along to that person/employee to read with limited contextual understanding. That will likely result in follow-up questions until the employee feels they have sufficient information to move forward.
Instead, an Enterprise Intelligence strategy could be deployed to centralize and curate all the documentation surrounding these systems to train and implement a Copilot. Employees now ask questions, gain context, and reduce dependence on expert intermediaries.
Where to Begin?
Azure AI has tremendous capabilities right out of the box and is also cost effective for companies of any size. To get a feel for how a deployment would look inside your organization, create a basic Copilot in Microsoft’s Copilot Studio, train it on a document set, then have a particular company segment test it. Make adjustments based on their feedback, iterate, rinse, and repeat.
The ability to create a Copilot and provide it a SharePoint location with a knowledge set to train on and answer prompts to is very approachable. The proof of concept is quickly achievable, and you can immediately see the surface area of benefits this could create.
What’s the Catch?
Much of the maintenance in this technology is simplified, but realizing ROI demands a well-thought-out strategy. Streamlining delivery and scaling effectively for consumers (inside and outside your organization), is the key aspect of an implementation strategy. If companies hastily adopt this technology without a strategic approach across the organization, they may face unexpectedly high costs at month’s end. This mirrors early cloud adoption, where companies moved servers and services to Azure primarily for convivence.
So How Do You Scale It?
Initially, with collaboration.
A few key strategic questions for any organization that is arriving at the inflection point where they know they need to adopt enterprise intelligence on a large scale.
The following questions only scratch the surface, but they will generate a useful thought experiment on managing change in your organization. This list will grow and evolve over time as the industry continues to develop best practices around consumer grade machine learning, MLOps (Machine Learning Operations) tools, and Artificial Intelligence services.
Like any good knowledge base, quality curation and effective security controls across your knowledge sets and now the resulting intelligence will need to be a top priority.
- What users have access to certain levels of intelligence within the same trained model?
- How will those controls be defined?
- Will intelligence be deployed as numerous Copilots throughout the organization or only a few monolithic ones?
Growing and training your Copilot intelligence is an ongoing task that would require continual delivery of new information. This will be a familiar concept for those who have experience in the Data Warehousing and ETL spaces.
- How will you configure the access of structured data (e.g databases, flat files, etc.) for your models to query?
- How will you automate delivery of unstructured data (e.g. PDFs, Word Documents, etc.) for your models to train on?
- What will be the single source of truth?
- Pipeline cadence? In other words, how often will you need to provide your model updated information.
This refers to the concept of your knowledge base is your source code. Anyone in software engineering and ITSM knows that Change Management is critical in mitigating risk and outages.
- How will change managements be assessed?
- What will knowledge and model promotion look like?
- What does User Acceptance Testing look like?
No Need for Copilots?
Whether your immediate organizational needs would not be solved with a prompt-based AI solution, or you would just see this all as a more long-term objective, elevating enterprise intelligence can be done in other ways. The rails of solution composition with Azure AI have been available in the Microsoft ecosystem for quite some time using their Power Platform stack of services. There is a deep list of AI models that can be trained and seamlessly weaved into your existing systems and processes within a no/low code set of tools and can have a massive impact on your organization.
Most of these services are free to try in order for users to develop basic solutions and learn the capabilities that are available to them without ever having to write a line of code.
About the Author
Chester “CJ” Kozarski is our Senior Solutions Director of Application Services. He is a Senior Technologist and Enterprise Solution Architect who drives measurable business value through leadership, innovation, and delivery of IT services across software, data, business intelligence, analytics and architecture domains. He has a bachelor’s degree in computer science from Shippensburg University.
GDC IT Solutions (GDC) empowers businesses to increase employee productivity, maximize investments and improve operational efficiencies. With experienced and certified professionals, GDC delivers services in the areas of application development, data center, 24/7 multilingual service desk, managed IT services, desktop lifecycle management, project management, and business process consulting. Learn more.