How a professional AI solution should look like?
ZF Digital Summit 2024 | 4th of November 2024 | Reflections
At the invitation of Ziarul Financiar and the ZF Business Hi-Tech News Team, I had the honor to having a closing session at the 14th edition of the ZF Digital Summit 2024. I would like to use this opportunity in putting a few still-fresh-into-my-mind aspects discussed around the AI "storm" that is pushing the IT&C industry to its limits nowadays.
Who and What was the ZF Digital Summit 2024 (14th edition)?
The agenda was packed during the whole day with amazing speakers from multiple industries, moderated by Ioana Nita , Adrian S. & Cristian Hostiuc . From top representatives on behalf of the Communications Authority, Telecommunications Operators, Legal Firms, Education, Banking, Retail & E-Commerce, Tech-Startups and many more.
There was no surprise around the intensity of the discussions around AI and the tools that are being used now inside organizations, with or without the knowledge of top or even middle management.
How was it like?
I will continue this article from my own perspective, reflecting on my learnings based on the discussions.
Since I was a guest speaker on the agenda, I went there physically and even though I was there with a few hours before my session even started...my smartwatch sent me a notification (>120 BPM heart rate):
Note: I am always nervous before any presentation. However, in time, I learned to channel emotions towards offering a genuine, valuable and useful message towards the audience I am in front of.
My heartrate moved back to normal while listening to the panels and the one-on-one discussions, because I knew more and more, that what I have prepared for the event, for my closing session, was answering lots of the aspects discussed on the stage already.
So, if you were not able to attend the event, either physically or virtually, therefore this is the reason of writing this. I am doing a short recap with key aspects of the AI "storm" of today.
Do you know that AI is already working inside your company?
This was the title of my session at the event. After a full day event, with 17 different sessions one after the other, we encountered an unfortunate but expected delay. Even the technical equipment played us some tricks so we had to cut it a bit short on the session. However, the importance of the topic and based on the promise made by Adrian S. on a follow up session, we will definitely do more discussions for the general audience around AI implementations in business.
75% employees use AI at Work. Majority of them (78%) they bring their own AI Tools to work.
Shadow IT, now became Shadow AI
This is in line with everything that was discussed. Many presenters acknowledged they are using tools such as ChatGPT in their activity. Imagine that this has a lot of impact, because there is one thing in creating a new social-media post for a public campaign using generative AI, and totally another when using a public AI to extract information from a customer contract or other confidential information about customers, suppliers, partners and even internal staff.
Leadership needs to understand the difference between "consumer AI" and "professional AI"
So, we kind of know that AI became "a must" for a modern organization, but more than half of the leaders lack real implementation plans for AI technology inside their organizations. This is practically leaving the door open for Shadow AI.
How do we identity a professional AI Solution?
As you may know and see in my profile, I work for Microsoft so it was at hand for me discussing about Microsoft 365 Copilot Advantage.
However, there are some key principles that one should take into consideration when implementing any AI into their organization. But, as subjective as I may sound, I genuinely believe that Microsoft 365 is the best solution for a modern organization today.
Here are the top principles I recommend when thinking about "injecting" AI capabilities inside any type of organization:
1. Integration INSIDE the organization, not OUTSIDE of it
Keep in mind that AI requires individual adoption. Therefore, a tool such as Microsoft 365 Copilot becomes part of the tools employees use daily: Outlook, Teams, Word, Excel, PowerPoint, etc. From the browser, from the client app, on PC, Mac, iOS, Android. Therefore, the learning curve for this tool, is very lean and easy to understand.
2. Working inside the company workflows
Microsoft Graph is the key element that "glues" the digital ecosystem of a company who uses Microsoft 365: from emails, meetings, digital identities, files, documents, chats etc. So, an AI which works internally on this specific set of data makes much for sense than any other "external" and generic AI.
3. Extensibility towards other <internal> business data sources
Same Graph can leverage more than 1400+ connectors towards 1PP (first party) and 3PP (third party) systems. Meaning, that we can use Copilot in getting responses from other apps and systems which we use inside the company (let's say a specific CRM or ERP system, or handling LinkedIn posts).
4. Customization capabilities
What if we want to create our own "Copilot"? or simply put, our own "AI"? What if we want to have a mini-AI system that will only handle requests and be trained on a very specific set of data? As an example: "We need an AI Bot, as powerful as GPT-4 Turbo, but trained on a specific set of documents regarding our products documentation". So the sales, technical and support teams can ask questions to this AI system and get relevant answers based on that specific set of data, which we can dynamically control. This is what we call Copilot Studio and Copilot Agents (i will refer to these in more detail a bit later).
5. Security, Privacy and Compliance at an Enterprise-grade level
There is a reason why Microsoft 365 is the standard in the collaboration and productivity spectrum. Actually there are more...but one of the most important ones is the security base that Microsoft 365 is putting in place for any customer. It makes total sense that Microsoft 365 Copilot follows strictly the security, privacy and compliance policies of the Microsoft 365 deployment itself inside the organization.
To put this into a clear example: If I have an email account in a company, and I do not have access to a specific document library, because those files were not being shared with me, then my Copilot (AI) will not have access either. Easy as that.
6. Measuring tools for AI impact and business value for the organization
A professional AI must have the right tools to drive adoption and impact. For Copilot we are using the Copilot Adoption Center, Copilot Success Kit, Copilot Lab, Learning Academy, Copilot Dashboard (available for Microsoft 365 customers)...and so on. I strongly encourage you to bookmark these links and check them out after reading this lovely article.
All in all...
One thing is for sure, a professional AI is so much more than a fancy LLM, or a nice webapp. Is a tool which requires: Integration, Understanding of work context, Extensibility, Customization, Security, Privacy, Compliance and clear Measurement of impact and adoption tools for a successful deployment.
So, whatever AI you might take into consideration, take at least the principles above as a must-have, before anything else. On the good side, Microsoft 365 Copilot checks them all from start.
Can an AI respect the data security, privacy and compliance inside the organization?
It does! When the right AI is implemented in the right way. At Microsoft we put security above anything else. I strongly recommend the Microsoft Trust Center for more deep-dive information on the topic.
A very old friend of mine, who is now one of the most esteemed GPDR consultants in the country, talks about common sense when handling data. This means, by design that a professional AI solution must handle data in a way that:
Recommended by LinkedIn
Good! We got strong principles and when you think about them, they do have common sense.
How do we apply them?
Many customers and partners asked me and my colleagues, why is Microsoft 365 mandatory for Microsoft 365 Copilot? This is the reason. Without a strong foundation of security, that Microsoft is offering by design to its customers via Microsoft 365, we cannot think about any further implementation of Copilot, or any other AI on that matter.
Think about creating a presentation based on a confidential report on the company results. Many of us are doing presentations based on specific files every day. Copilot in PowerPoint is crushing it! However, that report should have a sensitivity label (let's say Confidential), therefore I cannot (based on the policies) send it outside the organization, to say the least.
Shouldn't the PowerPoint .pptx file inherent the same sensitivity label as the report file it was used to create it? We believe it should!
Inheritance of DLP Policies and sensitivity labels follows the data sources, across application when using AI to create and generate content based on them, inside the organization.
The main question, is:
Do you know which files can be accessed by whom inside your organization?
With or without AI in your pipeline of future implementations, this question should be answered now more than ever.
Data sources:
From a security perspective, I believe is hard to find a more compelling solution than what Microsoft is now offering towards any customer. Especially when it comes to handing security and governance of a professional AI system, such as Copilot for Microsoft 365.
From baseline towards best-in-class solutions for the most complex organizations, we do keep our standards high when it comes to security.
Let me give you an example: Microsoft Security Defaults. This is a policy Microsoft started to enforce across all customers (tenants) last month (Oct'24). This implies MFA (multi-factor authentication for all users, among other things. For sure, many are were not very happy with this "update", but at the end of the day what is more important, commodity or Common Sense security policies of customer's and company data?
How a good prompt should look like?
In general terms, the majority of prompts are between 5-7 words. Prompts like: "Make me a summary about X", or "Write me a presentation regarding Y".
However, more relevant results can be reached when we get a bit more specific in our prompting and get above 15-25 words per prompt.
How can we build our own AI?
Copilot Studio & Copilot Agents
There are actually only 2 terms I need to introduce when it comes to building own AI. There is Copilot Studio and the Copilot Agents.
Basically, we use the Copilot Studio to build Copilot Agents.
We can enrich/extend the Copilot capabilities with Copilot Agents. Many of them are already build and available in the Teams Store, Microsoft 365 Store and the Microsoft AppSource marketplace.
Let's say we use UIPath or Mural, inside our organization. Wouldn't it be awesome if Copilot would be able to offer answers based on the information which resides into those ecosystems? Well...we only need to activate the agent, make the connection and that is it (this is point nr. 3 from the beginning of the article).
Ok. But, we want to really build our own stuff! What then?
Then we discuss about Copilot Studio!
I encourage you to look at this 1 minute video...is that easy.
You can ask what you need, you specify the data sources (may it be public/internal websites, internal SharePoint libraries, etc.), how should the agent behave, even put your own logo.
And yes, if you share this agent with someone who does not have access to that specific SharePoint library (to say the least), then the agent will not answer the questions for that person. Point nr. 3 - Protection at every step, remember? Pretty cool, isn't it?
What's next?
If you reached this line, means you actually find this interesting and valuable, and I am deeply humbled.
I can only recommend you to get in touch with your Microsoft Partner and be part of our local events driven together with partners and customers. Because, beyond the text, images and videos above, we are really looking forward in showcasing this in a live demo to you. And here, I can only send my full gratitude towards the Microsoft Partner Community who is working shoulder to shoulder with us in bringing professional AI solutions towards the market.
In conclusion, I would stick to these principles in any AI discussion: Integration, Customization & Extensibility, Security, Privacy, Compliance, Impact measurement and Adoption tools.
Thank you,
Robert Pufan | Sr. Partner Technical Architect for Modern Work, Microsoft Central Europe
5th of November 2024
Distribution Business Manager at HP
3wThanks for sharing you much appreciated view on AI, Robert! Amazing how you describe it so everyone can understand and take action! Kudos! 👏
I help Romanian and Eastern European Countries Companies in Retail, Manufacturing and Distribution to grow their online B2B & B2C e-Commerce Sales leveraging the most powerful All-In-One composable platform @DynamicWeb.
3wYour presentation delivery is great Robert Pufan! Loved the advice with keeping a steady heart rate. Thank you for sharing your reflections from Ziarul Financiar event!