I’m teaching a new course this semester on cognitive technologies (AKA artificial intelligence) to Babson MBAs. Many of them are new to this set of technologies, and seeing the topic through my students’ eyes has made me realize how overwhelming it can be. There are so many different types of AI, each requiring some technical knowledge to fully grasp, that newcomers to the field often have difficulty figuring out how to jump in.
In the simplest case, cognitive technologies can be just more autonomous extensions of traditional analytics — automatically running every possible combination of predictive variables in a regression analysis, for example. More complex types of cognitive technology — neural or deep learning networks, natural language processing, and algorithms — can seem like black boxes even to the data scientists who create them.
Though these technologies can seem daunting, the good news is that getting started with cognitive technologies is getting easier all the time. Many vendors have jumped into the field, and their offerings provide options for any company wanting to make their processes or products smarter. I can think of at least seven ways to begin using cognitive tools, although some are clearly easier (and cheaper) than others. Because implementing these technologies is a key factor in deciding how to move forward, I’ve combined the cognitive entry points into three categories: “Mostly Buy,” “Some Buy, Some Build,” and “Mostly Build.”
- Use an existing vendor’s software with cognitive capabilities. For example, Salesforce.com and Oracle recently announced that they are adding cognitive capabilities to their products. Salesforce is adding Einstein features to its customer-facing software clouds, including the ability to automatically score sales leads, read emails from customers, and classify images used in social media. If you’re already using Salesforce CRM offerings and want to ease into smarter processes for sales, marketing, and service, this seems like one of the easiest ways to do it. Some other CRM companies like Customer Matrix were founded with the idea of combining cognitive tools with customer transactional capabilities. Microsoft has also recently announced that it will add cognitive capabilities to many of its existing software products. If you’re using any of these vendors’ offerings, before long it will probably be harder to avoid cognitive features than to use them.
- Pick a small project and a “low hanging fruit” vendor. Rather than going all in, some companies begin by picking a small project that could benefit from cognitive technology, and use a smaller, less transformative toolset to attack it. For example, Cognitive Scale — several of whose leaders were IBM Watson executives — tries to pick the low-hanging cognitive fruit. It has a “10-10-10” development approach, in which the goal is to build a rough cognitive application in 10 hours, customize it in 10 days, and go live within 10 weeks. Some of Cognitive Scale’s customers, like M.D. Anderson Cancer Center (which is also pursuing an ambitious Watson project for cancer treatment), have many different projects underway with the company’s software. These projects don’t attempt to cure cancer, but rather address narrower objectives like providing patient families with lodging and dining recommendations, or determining which patient bills are most in need of extra collections efforts.Some of the “robotic process automation” offerings from companies like Blue Prism and Automation Anywhere also qualify as low-hanging fruit, although as of yet their software doesn’t learn. Some call center automation offerings like Ipsoft’s Amelia also fall into this category. While these projects will require some consulting to train or configure the software, there are usually services available from the software companies or their consulting partners to perform such work.
Some Build, Some Buy
- Build on your analytics strength to emphasize machine learning. Some forms of machine learning — particularly those based on regression analysis — are straightforward extensions of a company’s analytical capabilities. If you’ve largely mastered “artisanal analytics” based on human hypotheses, it may be time to explore the automated generation of analytical models through machine learning. Cisco Systems, for example, transitioned from creating tens of “propensity models” that predict what customers are likely to buy, to creating tens of thousands (currently about 60,000) of models per quarter through machine learning. Cisco found a way to accelerate their creation so that it only took a few days per quarter. And the increased granularity of the models (specific ones for products, geographies, and customer business types) improved their effectiveness. Depending on the type of models a company wants to generate and the software it uses to do it (e.g., proprietary vendors like SAS and IBM vs. open source), this can be either technically straightforward or somewhat complex, requiring sophistication in data science.
- Go big with Watson. IBM’s Watson occupies the high price, high reward quadrant in the cognitive race. Yes, you can cheaply access one of the Watson APIs on Bluemix, IBM’s developer cloud—my students did it in our last class for free—but the full Watson offering is mostly aimed at large-scale, transformative applications. IBM likes to start customers out with a “Cognitive Value Assessment” that indicates the best place to hit a cognitive home run. Then it’s happy to supply consultants and even Ph.D. researchers to help you hit it. This doesn’t yield a cheap or quick outcome, however. Particularly if you’re the first in your industry to use Watson, there will be a lot of training and integration necessary (as I wrote about last year when I surveyed health care applications of Watson). But companies that are comfortable working with IBM on a large scale and believe it’s important to make a big dent in their business with cognitive technology will find this approach appropriate.
- Start with chatbots. Chatbots are a medium-level cognitive technology that use natural language conversation to interact with apps. Google (with its recent API.AI acquisition), Apple (Siri), Microsoft (Cortana) and Facebook (Messenger Platform) all have platforms for developers to deploy their chatbots. Particularly if your company is focused on mobile, where consumers seem particularly inclined toward chatbots, you might dive into cognitive tech by picking one of these companies’ APIs and connecting your apps to it. These user interfaces will certainly evolve over time, but it’s a good idea to start working with them now if you want to take advantage of easy voice interaction.
- Make an existing application smarter or more autonomous. Using modular, component-based architectures, it is possible to add cognitive approaches to applications. For example, Vanguard, the investments firm, created the semi-autonomous Personal Advisor Services (PAS) capability for its asset management customers. Vanguard already employed several of the capabilities behind the scenes, including investor questionnaires, model portfolios based on analytics, account rebalancing, tax loss harvesting, and goal-based simulations. In the PAS project, it strung those capabilities together, made them autonomous (with review by human advisors), and made them available in investment plan documents and on the web. But this sort of work requires expertise in cognitive tools as well as systems integration capabilities.
- Build from open source software. There is no shortage of open source cognitive software. Google, Microsoft, Facebook, Amazon and Yahoo have all released open source machine learning or deep learning algorithm libraries. The tradeoff here is perhaps obvious. Since the software is free, this approach offers the lowest software costs. But it probably also will yield the highest people costs, since the data scientists who can use such libraries are rare and expensive. Building your cognitive solution from scratch with open source tools could also take longer than some of the other options. So this entry point probably only makes sense if your company has really specialized needs and is willing to make a long-term commitment to building cognitive capabilities. It’s also probably a good approach if you plan to embed cognitive features into your product or service.
I’m sure there are other angles that a company could take to adopting cognitive technology, but to date these seem to be the most common ones. Each has different implications for the kinds of skills an organization needs and how it manages the technology once it comes in the door. Some ambitious organizations may want to pursue multiple entry points at once. It’s great that there are so many options, but when management teams decide to integrate cognitive technology into their strategies, they should think hard about which one they plan to pursue.
Thomas H. Davenport is the president’s distinguished professor in management and information technology at Babson College, and cofounder of the International Institute for Analytics. He also contributes to the MIT Initiative on the Digital Economy as a fellow, and as a senior advisor to Deloitte Analytics. Author of over a dozen management books, his latest is Only Humans Need Apply: Winners and Losers in the Age of Smart Machines.