Attainable AI
Attainable AI
Attainable AI
Oct 29, 2024
Oct 29, 2024
Oct 29, 2024
Hype versus Reality
Countless books, articles, blogs, and conference sessions have been devoted to Artificial Intelligence (AI) during the past several years. Business leaders often hear that they must invest in AI to remain competitive and relevant. Consequently, many companies have invested considerable resources to chase aspirational AI capabilities that are unattainable or that fail to provide a sufficient return on the investment.
Homegrown bespoke AI capabilities are expensive to build and present long-term scalability and supportability risks. On the other hand, many SaaS platforms such as Salesforce now offer integral Generative AI (GenAI) capabilities for particular modules and functions—and these built-in GenAI capabilities can offer value for no or relatively low additional cost. And of course, the GenAI capabilities of ChatGPT and Microsoft CoPilot have helped to democratize AI for the masses—but present new kinds of challenges regarding data privacy and intellectual property protection. As with any technology investment, success with AI depends on advanced planning, focused objectives, and assessment of risks. Most importantly, AI should be viewed as a means to an end; it is ultimately another kind of technology and tool that can help solve business problems. The "hammer looking for a nail" mentality—in which AI becomes the objective rather than the means—usually leads to failure.
Key Considerations
To avoid common AI initiative pitfalls, consider the following before embarking on an AI journey:
Process Focus: This is the first and most important consideration. Start by creating current state process maps and identifying target areas for efficiency and automation—and then evaluate whether the target areas might benefit from AI.
Model Training: Custom AI capabilities usually require a combination of underlying data, connectivity to the data, and training of the AI model. For example, an IT services concierge (chatbot) capability will need access to knowledge base articles and other ITSM (ticketing) platform data; once data connectivity is in place, the AI model requires training and testing. Even SaaS platforms that offer integral GenAI capabilities will require customer-specific data along with training and testing.
Third Party Data Restrictions: Assess what type of data you wish to consume in your AI models and whether they include third party data sources. If so, first review the data provider's agreement language for data usage restrictions—and consult with the data provider to clarify any points that are not spelled out clearly in the agreement. In many cases, data providers disallow usage of their data in third party platforms such as ChatGPT.
Data Privacy: Review the AI platform's data privacy language to understand whether they will consume your data in their Large Language Models (LLM). This in turn will influence what data you will need to exclude or redact—or whether the data restrictions undermine the overall AI value proposition.
Change Management: When an organization embarks on an AI initiative, some employees will assume that labor reduction is the primary objective and will therefore feel threatened that their job will be replaced by automation. If employees are likely to carry this type of negative sentiment, consider a communication strategy to address the concerns.
Policy: The popularity of ChatGPT has led to widespread adoption in some organizations. Therefore, it's important to educate employees on acceptable usage—including restrictions on the use of sensitive data. Communications, training, and policies that cover AI acceptable use scenarios will help prevent data privacy and intellectual property breaches.
Practical and Attainable AI
There are countless types of processes that might benefit from AI, but which common use cases are readily attainable and cost effective? During the past year I have encountered two AI use cases that often fly under the radar:
Meeting Notes: I have been really impressed with the notetaking abilities of Otter.ai and Microsoft CoPilot for Teams solutions—and they essentially replace the need for a dedicated human notetaker. In fact, they often do a better job than a human: the notes are contextualized and topically organized (including action items and assignments), and the AI engines provide key metrics such as who was invited but did not attend as well as the participation (speaking) rate amongst attendees. CoPilot also offers an impressive party trick: if attendees join a meeting that is already in progress then they can ask CoPilot to provide a summary of the meeting up to that point. CoPilot does a fabulous job with generating these types of meeting summaries.
Service Management / Ticket Systems: For many years, leading platforms in this space have offered chatbots that can interact with users and attempt to answer their questions via "learning" from attended training or by crawling through knowledge base documentation. More recently, leading platforms have begun to offer more sophisticated capabilities that include request orchestration and CMDB automation. These newer AI capabilities can accelerate system setup efforts and—more importantly—reduce user friction. For example, a ticket system portal is traditionally organized by many types of request categories; the user must review the available categories and select the one that most appropriately aligns with their request. Selecting the wrong category can result in misrouting and request response delays. With AI-based orchestration, a single form can be used for all requests: the user inputs key information about their request and the AI capabilities automatically categorize and route the request.
Final Thoughts
AI capabilities have made amazing strides during the past few years, but as with any leading technology the hype versus reality gap is large. Therefore, organizations can make the mistake of chasing AI as the latest magic wand or shiny object, rather than properly assessing it for what it is: a tool that can help solve problems and yield efficiencies. But this perspective does not diminish the importance and potential of AI; when there are reasonable expectations and forethought, organizations can reap significant benefits from their AI investment.
What do you think? Please drop us a line if you would like to share your thoughts, experiences, or questions:
Hype versus Reality
Countless books, articles, blogs, and conference sessions have been devoted to Artificial Intelligence (AI) during the past several years. Business leaders often hear that they must invest in AI to remain competitive and relevant. Consequently, many companies have invested considerable resources to chase aspirational AI capabilities that are unattainable or that fail to provide a sufficient return on the investment.
Homegrown bespoke AI capabilities are expensive to build and present long-term scalability and supportability risks. On the other hand, many SaaS platforms such as Salesforce now offer integral Generative AI (GenAI) capabilities for particular modules and functions—and these built-in GenAI capabilities can offer value for no or relatively low additional cost. And of course, the GenAI capabilities of ChatGPT and Microsoft CoPilot have helped to democratize AI for the masses—but present new kinds of challenges regarding data privacy and intellectual property protection. As with any technology investment, success with AI depends on advanced planning, focused objectives, and assessment of risks. Most importantly, AI should be viewed as a means to an end; it is ultimately another kind of technology and tool that can help solve business problems. The "hammer looking for a nail" mentality—in which AI becomes the objective rather than the means—usually leads to failure.
Key Considerations
To avoid common AI initiative pitfalls, consider the following before embarking on an AI journey:
Process Focus: This is the first and most important consideration. Start by creating current state process maps and identifying target areas for efficiency and automation—and then evaluate whether the target areas might benefit from AI.
Model Training: Custom AI capabilities usually require a combination of underlying data, connectivity to the data, and training of the AI model. For example, an IT services concierge (chatbot) capability will need access to knowledge base articles and other ITSM (ticketing) platform data; once data connectivity is in place, the AI model requires training and testing. Even SaaS platforms that offer integral GenAI capabilities will require customer-specific data along with training and testing.
Third Party Data Restrictions: Assess what type of data you wish to consume in your AI models and whether they include third party data sources. If so, first review the data provider's agreement language for data usage restrictions—and consult with the data provider to clarify any points that are not spelled out clearly in the agreement. In many cases, data providers disallow usage of their data in third party platforms such as ChatGPT.
Data Privacy: Review the AI platform's data privacy language to understand whether they will consume your data in their Large Language Models (LLM). This in turn will influence what data you will need to exclude or redact—or whether the data restrictions undermine the overall AI value proposition.
Change Management: When an organization embarks on an AI initiative, some employees will assume that labor reduction is the primary objective and will therefore feel threatened that their job will be replaced by automation. If employees are likely to carry this type of negative sentiment, consider a communication strategy to address the concerns.
Policy: The popularity of ChatGPT has led to widespread adoption in some organizations. Therefore, it's important to educate employees on acceptable usage—including restrictions on the use of sensitive data. Communications, training, and policies that cover AI acceptable use scenarios will help prevent data privacy and intellectual property breaches.
Practical and Attainable AI
There are countless types of processes that might benefit from AI, but which common use cases are readily attainable and cost effective? During the past year I have encountered two AI use cases that often fly under the radar:
Meeting Notes: I have been really impressed with the notetaking abilities of Otter.ai and Microsoft CoPilot for Teams solutions—and they essentially replace the need for a dedicated human notetaker. In fact, they often do a better job than a human: the notes are contextualized and topically organized (including action items and assignments), and the AI engines provide key metrics such as who was invited but did not attend as well as the participation (speaking) rate amongst attendees. CoPilot also offers an impressive party trick: if attendees join a meeting that is already in progress then they can ask CoPilot to provide a summary of the meeting up to that point. CoPilot does a fabulous job with generating these types of meeting summaries.
Service Management / Ticket Systems: For many years, leading platforms in this space have offered chatbots that can interact with users and attempt to answer their questions via "learning" from attended training or by crawling through knowledge base documentation. More recently, leading platforms have begun to offer more sophisticated capabilities that include request orchestration and CMDB automation. These newer AI capabilities can accelerate system setup efforts and—more importantly—reduce user friction. For example, a ticket system portal is traditionally organized by many types of request categories; the user must review the available categories and select the one that most appropriately aligns with their request. Selecting the wrong category can result in misrouting and request response delays. With AI-based orchestration, a single form can be used for all requests: the user inputs key information about their request and the AI capabilities automatically categorize and route the request.
Final Thoughts
AI capabilities have made amazing strides during the past few years, but as with any leading technology the hype versus reality gap is large. Therefore, organizations can make the mistake of chasing AI as the latest magic wand or shiny object, rather than properly assessing it for what it is: a tool that can help solve problems and yield efficiencies. But this perspective does not diminish the importance and potential of AI; when there are reasonable expectations and forethought, organizations can reap significant benefits from their AI investment.
What do you think? Please drop us a line if you would like to share your thoughts, experiences, or questions: