Skip to Main Content

Generative AI: Using artificial intelligence to make human impact.Learn how

Illustrated computer chip


Deploying Smart Machines

Evan Rowe
Evan Rowe
Max Kirby
Max Kirby

Deliver business performance breakthroughs with machine learning

As emerging technologies in AI, machine learning and data science mature, leading business executives find ways to operationalize it.

What's hard for people is easy for machines

If you’ve ever watched one of the many popular television shows about law enforcement, business or medicine you’ve seen the scenario where an army of young associates pore through boxes of files to find that piece of insight that might pave the way toward resolving a life-threatening situation.1

Maybe they’re attorneys searching for a legal argument for a high stakes trial or young doctors trying to predict how a patient will react to a treatment or a bunch of traders eager to know which supplier stocks will go up when Apple introduces its next iPad.

Smart machines offer an attractive business case

The people searching for the story in the data are highly educated, well-paid and often work all night, days and weeks to find evidence or patterns of insight that can be used to solve a problem. Even when the scenarios are conducted online, these smart associates often need to digest each file as they search for the story.2 And, as humans become exhausted, they make mistakes or overlook a detail. Enter smart machines, who will work 24/7, never tire, and will never demand overtime.

Smart machines are an effective substitute for what humans find unfathomable: digesting enormous volumes of data, taking into consideration each and every fact the data represents. Business people excel where smart machines do not: making decisions from the facts, context and background. Elbert Hubbard sums it up: “One machine can do the work of fifty ordinary men, yet no machine can do the work of one extraordinary man.” Put another way, “What’s hard for people is easy for machines; what’s hard for machines is easy for people.”3 For example, equipped with enough learning data, a smart machine can perform sales analysis on millions of products, but it cannot judge whether a decision to offer a higher price for loyal customers is ethical. Hence, these initiatives require the organization’s combined experience and skills with new smart machine capabilities.

One machine can do the work of fifty ordinary men, yet no machine can do the work of one extraordinary man.
Elbert Hubbard

Traders pose questions to a smart machine named for Warren Buffet

Turning labor-intensive tasks over to smart machines is the “low hanging fruit” of artificial intelligence and machine learning. While still considered an emerging technology, smart machines garnering insight from huge volumes of information (structured or unstructured) is an activity that is quickly maturing—and is proving to be a viable technique for staying ahead of competitors.

For example, tune into CNBC4 and analysts might pose the Apple supplier scenario to Warren (named for Warren Buffet), a smart machine developed by startup Kensho, which ingests massive amounts of stock data from sources such as S&P Global. When asked how Brexit would impact the market, Warren combed through an intelligence-grade database of information and in seconds suggested that events such as Brexit historically lead to an extended drop in the local currency, washing out any short-term recovery (which is exactly what happened in the days and weeks after Brexit).

Lawyers consult a “moneyball” for judges

Founded in 2012, Ravel Law (now part of LexisNexis)5 was spun out of Stanford University’s Law School to develop a type of “Moneyball for Judges,” where millions of mined court documents predict the likelihood of a lawsuit’s dismissal—or forecast the wait time of a trial.

Lawyers report the tools help them set client expectations, influence courtroom decision-making and even save money by avoiding strategies that score a low probability of success (and, when you sign up for Ravel, you get free access to thousands of cases and opinions going back to the 19th Century).

Eric Olson, a Denver litigator, said a smart machine helped him see that a certain judge does not like sports analogies, which are rife among trial lawyers. Says Olson, “I would never use ‘moving the goal post’ in front of that judge.”6

" machines garnering insight from huge volumes of information (structured or unstructured) is an activity that is quickly maturing—and is proving to be a viable technique for staying ahead of competitors."

Implementing machine learning: advice & recommendations

Make sure you have sound data governance

The data ingested into smart machines must accurately represent prior instances of a situation or condition (which makes it suitable for making predictions).  

When conditions change, or new instances uncover fresh insight (e.g., a court decision overturns an old law), smart machines dynamically retrain themselves (meaning, IT professionals don’t have to rewrite the instructions as they would for systems that are explicitly programmed).  

Hence, to deliver ROI in smart machines, data and analytics teams must have good data management practices in place, along with the vertical business expertise to interpret the outputs from machine learning.

Use smart machines to advance other analytics initiatives

Avoid tackling problems that are too big and require time to get experts onboard or that require big funding requests. Rather, start with a common business problem that many stakeholders would understand (and one where money and resources are already available).

For example, start a machine learning effort that will leverage the data you use in common, recurring reports such as sales orders by region. In this example, you can apply machine learning to make forward-looking forecasts for the next several months.  

If your team has a good understanding of how the data is used, along with domain expertise and the ability to interpret outputs, you have the attributes for a successful initiative.

If you’re new to AI and smart machines, begin with a project suited for supervised learning.

Select a business challenge that everyone understands

Use the scenarios we provided for inspiration, assuring you have available, clean information represented through well-understood data sets

  1. Focus on data represented in some of your key business reports, such as revenue by region, which can be used to forecast orders by region for the next quarter. 
  2. Select a problem in which your stakeholders are in solid agreement. For example, the conversion represented by a series of customer actions, or the legal precedents that led to a judge’s decision, or the events that led to an increase or decrease in a stock price.
  3. Focus on the type of predictions or forecasts you struggle to make by attempting to consume volumes of trend data. These are the tasks smart machines do well. Plan on deriving actionable insight from a pilot project within three months.

Leverage your existing expertise and skills

Many organizations analyze and compare the skills they already have with those needed for a machine learning project. This exercise will reveal the skills and capabilities you have, versus those you need to deploy smart machines.  

Use the results of this exercise to inform the build, buy or outsource decisions. Don’t fall into the trap of hiring people for their current knowledge with no appreciation for their ability to learn. Many universities that offer data science-related degrees also require students to do hands-on work.  

Partnering with academia has also been productive for many organizations, for example, Fijitsu’s partnership with École Polytechnique provides students with opportunities to work with its Center of Excellence for AI and data science.8

"Don’t fall into the trap of hiring people for their current knowledge with no appreciation for their ability to learn."

Score a quick win

In an environment where technology changes and advances so quickly, few organizations have the luxury of learning everything they need to know before they launch an emerging technology initiative.

  • It’s better to achieve a quick win where real business value can be delivered than to consume time studying or monitoring the wins of others (a scenario that is sure to erode your business advantage). 
  • You don’t necessarily need to know how to build a proprietary model for extracting data insights. Rather, explore using a packaged application to address a specific business case (e.g., how to reduce customer churn or acquire more high-lifetime-value customers). Established software vendors are introducing AI into their product strategy.
  • Your project could also use a cloud API that returns output from your data or a model developed by a consulting company (or even a freelancer).

Learn as you go

Barclays South Africa learned how to drive value from smart machines by simply giving them a try. In 2016, it launched its full-chat banking services on Twitter as well as on Facebook Messenger, letting customers do their banking within these respective social media platforms. Says Brett St Clair, head of digital products: “We are greatly excited by the level of activity displayed by early adopters, and this channel is still growing in terms of customer understanding and utilization.”9

When new technologies come along, they can often compromise the user experience if customers are redirected into unfamiliar, adjacent environments. The bank was sensitive to this, hence it took great care to sustain the existing experience in the face of an emerging capability. St Clair continues, “We are the first bank in the world to fully authenticate customers via Facebook Messenger to allow them to do their banking such as making a beneficiary payment directly from Facebook Messenger. Our customers are never re-directed to our web site.”

Garner stakeholder agreement

Start with a machine learning initiative where you have clean data that is properly governed, along with resources commensurate with the project’s vision. But, scope the problem carefully. If you combine a problem with too large of a scope, with one that is potentially controversial (because stakeholders disagree on how to solve it), your initiative will get stuck.  

If you observe, for example, differences of opinion amongst business leaders that have different expectations from the data, find another problem where such differences do not exist.

Start with models that can exploit well-understood datasets

AI projects will fail if the data is insufficient, inaccurate, inconsistent, incomplete or biased. Be sure your historical data represents both desirable and undesirable outcomes. Otherwise, your training phase will not be equipped to properly learn.  

This is why it’s often a good idea to start with a business challenge where data exists in popular, recurring sales and marketing reports. For example, if you have systems in place that track win-loss, where data is available to support the characteristics of deals that were both won and lost, you have a project with high potential.

Do you need data scientists? Not always.

Many organizations get stuck in a classic chicken-and-egg scenario: without data scientists, the organization doesn’t have the skills to fully exploit machine learning, but without any quick wins, business leaders won’t let you acquire them.  

You may be surprised to learn that you already have mathematically skilled people, who have been math geeks all their lives or are using their quantitative skills in other roles. You can also research consultancies to help apply definition to your ideas, and then help you pilot them. Many consultants also have knowledge transfer techniques to help you teach your staff.

Start now

Don't delay your ventures into smart machines even if you believe the hurdles are too high. The emerging technologies around AI have far too much business potential to ignore. Start by framing two or three business initiatives you can reasonably fund over the next year. Then, prioritize your ideas, avoiding any that have the potential to over-promise results. Select the technology that is the least complex for an idea that is the lowest risk. All projects will have challenges but focus on those where risks can be mitigated.


  1. Judge Shames Lawyers Over Midnight Filings, by Stacy Zaretsky, Above the Law, 11 June 2015.
  2. Old police ‘street files’ raise question: Did Chicago cops hide evidence?, by Jason Meisner, Chicago Tribune, 13 February 2016.
  3. How to start a machine learning initiative with less anxiety,” by Svetland Sicular, Gartner, 27 October 2017.
  4. This earning season should be a good one for stocks if history is any guide, by Thomas Franck, CNBC, 17 October 2017.
  5. Ravel Law sells to LexisNexis, by Connie Loizos, TechCrunch, 10 June 2017.
  6. Harvard and Ravel want to free the law, by Jeremy Berke, Business Insider, 26 January 2016.
  7. How Successful Companies Make Big Data Operational, CapGemini Consulting, 14 January 2015.
  8. Fijitsu to establish a center of excellence in École Polytechnic, Entrepreneurship and Innovation newsletter, École Polytechnic University, 13 March 2017.
  9. Absa banks on robotics, artificial intelligence, by Admire Moyo, Pansmart, 2 May 2017.
Evan Rowe
Evan Rowe
Vice President, Data Science and Analytics and Group
Max Kirby
Max Kirby
Director of Digital Identity

Related articles