November 13, 2018 | Written by: Salil Ahuja
I started my journey in cognitive and artificial intelligence (A.I.) technologies back in 2013. As one of the first product managers for IBM Watson, I had identified customer self-service as a go-to-market opportunity and the Watson team launched one of the first A.I. self-service virtual agent technologies, Watson Engagement Advisor. Since those early days, I have worked with hundreds of customers globally and, have been fortunate to be part of some amazing transformation journeys based on Watson. With a deep understanding of what works and, what does not, the following are some tips to consider when you’re ready to bring A.I. to your enterprise. As a precursor, the following link describes the differences between AI and machine learning – a distinction that I have found to be blurred many times.Know know more click here.
With that said, let’s start:
Tip #1 – The most important leading indicator for a successful implementation of cognitive technologies is effort, and more specifically human effort. For example, let’s take machine learning (ML) and natural language processing (NLP). If a task requires a human to repeatedly read the same kind of documents with very similar formatting to find facts, it is probably a good candidate for AI. A.I. can reduce human effort by reading each document, and mining the facts in those documents that are pertinent to a set of business needs. In fact, in IBM we prefer to call A.I. ‘Augmented Intelligence’ rather than Artificial Intelligence. The neural networks, ML and NLP capabilities today like Watson, augment our intelligence by making data effective to our decision making. They can process for example thousands of documents, social feeds, reports in far greater volumes and speed than humans. Information overload is a key hindrance to effectiveness of employees and esp. knowledge workers. “19.8 per cent of business time – the equivalent of one day per working week – is wasted by employees searching for information to do their job effectively,” according to Interact. Source: A Fifth of Business Time is Wasted Searching for Information, says Interact
A.I. is well suited help with this problem. It can make employees more effective at their jobs by augmenting search, retrieval, relevance and business utility.
Tip #2 – Don’t think about the use of A.I. technology without considering the data that drives it. ML that drives AI needs data to get trained on. When you think of a use case for these technologies, first think of the data that is available to execute that use case. This data can be internal, external, of both. For example, let’s say your use case is understanding customer complaints. Consider all the digital touch points and points of interaction that customers have with you: phone calls, emails, comments on websites, social messages, SMS, etc. These data artifacts will help you understand and identify customer sentiment, potential service improvements, and other business insights that may have previously been invisible. When building an A.I. system, you will need all of this data in order to truly train your ML models. As you think of designing your A.I. system, think about where this data exists, how you will access it, and who owns it. Ensure that you have the legal rights to use the data or can gain rights to use the data.
Tip #3 – A.I or Machine Learning technologies work best with classification, regression and clustering kind of use cases that are supported by large historical data sets – Given the maturity of A.I. technologies at large, these systems today can be broadly applied to classification, regression and clustering kind of use cases. Some examples are finding insights in data, understanding text, classifying unstructured data like text, speech and images and identification of patterns within large data sets. Given unlimited time and resources we could probably make A.I. solve all use cases, but in the real-world businesses do not have that luxury. Businesses need solutions that are quick to implement, deliver high value today and simple to maintain. Therefore, understand the business needs, the data, the business outcomes, enterprise policies, timelines and then consider all available technologies along with A.I. when you architect the best possible solution. For example, a system to make decision based on discrete enterprise policies is still probably better solution using a rules engine or a system to execute a chain of events for a business process is probably better suited to business process management solution. That said, A.I. even today can add value to specific steps in those use cases.
Tip #4 – Find the AI tech that is trained, or can be trained, for your specific business need. Machine learning capabilities and A.I systems, like Watson, are considered to be ‘applied artificial intelligence’. This means that they have been trained on a very specific task. We are still decades away from a truly super intelligent machine or ‘General A.I.’ Hence gain an understanding on what training sets the ML models were trained on and what business problems are the models designed to solve. For example, in visual recognition models that recognize images, one service may recognize people and emotions better than cars and landmarks. It is also very important is to understand the flexibility that exists to train the model on your specific enterprise needs. Not all A.I. allows tools to adapt the models that are specific to your environment. This distinction and a recognition of the nature of this insight is very critical to business.
Test the A.I. technology with your data before selecting the right capabilities. This may sound like a long, drawn-out process, but it is not. Fortunately, most A.I. technology today has ways of rapidly evaluating the effectiveness with your data. For example, in my world with Watson, we provide live instances of the capabilities on our website -http://www.ibm.com/watsondevelopercloud. Most APIs have a live instance that you can run your data through without needing any minimal training or programming. You can quickly begin to see the applicability of the A.I. technology for your particular use case.
Tip #5 – Is your enterprise ready to adopt A.I? Like all leading-edge innovation, it takes a certain mind set, persistence, and vision to bring about true transformation. Sometimes people and processes are the issues, not the product. The good news is with A.I. being delivered on the cloud a lot of the initial teething pains in terms of installations and configurations have been removed. Technology fit assessments now take weeks and not months. So start on focused implementations, prove value, gain support and then expand. That is the only way you can get support through your enterprise and get them ready for this transformation journey.
Hope this helps provide some insight on what to look for when in the market for A.I. And if it is Watson you’re after or want to learn about we are always available for a chat, call or coffee.