Chapter 5

Augmented Intelligence in a Business Process

Introduction

Humans and machines working in collaboration can have a powerful impact on the effectiveness of business processes. Augmented intelligence overcomes the limitations of isolating human understanding from the massive amounts of both structured and unstructured data available to analyze complexity in record time. So, how does augmented intelligence change business processes and the way that work is getting done?

Defining the Business Process in Context with Augmented Intelligence

A business process represents a series of tasks that are performed in a prescribed end-to-end sequence. There are two techniques for applying augmented intelligence to improve an end-to-end business process: weak augmentation or strong augmentation. In addition, there are situations in which organizations will need a combination of weak and strong augmentation. Let’s look at what each of these business process approaches consists of.

Weak Augmentation

There are many situations in which automation is undertaken to streamline a set of repetitive business processes. For example, a worker may be required to insert the same information multiple times in a series of forms that are part of a single business process. This could be a newly insured motorist who is buying a policy and needs to have the same data in a series of related forms. Often, there are business processes that require that the worker understand precisely which forms are needed in which sequence to execute in a business process. One of the benefits of weak augmentation is that it is fast to implement because the processes themselves are not changing. Therefore, this is a pragmatic approach to saving time and money through straightforward process automation. The automation does require some level of intelligence, since the machine needs to be able to understand the patterns of the text fields being automated. Automating the ability to place the correct information in the right place solves a time problem but does not address the efficiency of the process itself.

Understanding Weak Augmentation

In weak augmentation, the flow of tasks from beginning to end is unchanged. However, machine labor is substituted for human labor for the more routine tasks within the process flow, where machines have an advantage in terms of speed and accuracy. For example, a process might require a series of forms to be manually filled in by clerical staff, even though the same data is filled in over and over again as you move from form to form. This task, manually performed, is inefficient, prone to error, and costly to apply. Augmented intelligence can be deployed to map differently named fields across separate forms—each field capturing the same data. Once this mapping has been established, the fields can be manually filled in once and then the data can be automatically applied (following the map) to fill in the same fields on related forms.

Weak augmentation is the low-hanging fruit of business process improvement. The automation of the form-filling task by an application of augmented intelligence results in increased productivity and efficiency through a process now executed by a mix of human and machine labor. This approach of automating steps of an existing business process is sometimes termed “lift and shift” in that a process is lifted from its former environment (human labor only) to a new environment (a mix of human and machine labor). The term “lift and shift” is borrowed from the world of outsourcing, where a services company is hired by a company to perform a specific process, following the same steps as before, but shifting it to a geography where labor is cheaper. The shift occurs without any changes to the sequence of tasks in the work process.

Not surprisingly, outsourcing services companies are now revising their business model to incorporate augmented intelligence for cost-saving advantages. Instead of relying on finding the cheapest labor markets to get tasks done manually for less cost, they are investing in AI for greater and more sustainable cost advantages. The new “lift and shift” occurs when the services company substitutes machine performance of tasks for human performance of tasks. This automation approach is proving to be a more efficient way to deliver outsourcing services on behalf of their clients.

RPA as Weak Augmentation

Robotic process automation is a technique that has gained considerable traction over the past few years because of its ability to quickly automate repetitive office tasks. This is in contrast to traditional manufacturing automation that focuses on taking one portion of a workflow or even just one task and creating a robot to specialize in it. Office work often requires the same types of repetition, but it is data being manipulated across platforms and applications, so a physical robot is not necessary. Instead, a software robot is deployed with the ability to launch and operate other software.

Robotic process automation and artificial intelligence (AI) each represent a means to provide a level of automation of a task, or even the entire sequence of tasks within a business process, by following a set of rules. For certain steps within a well-defined process such as invoice handling, the rule-based approach works well. You can even specify rules to flag an exception when certain conditions are met, triggering a message to a human agent requesting a resolution of an issue before payment is approved.

Artificial intelligence capabilities can be used to extend RPA’s capabilities.

  • Natural language processing can be applied to understand text inside of the invoice. This capability enables detecting similarities in field names across documents.

  • Machine learning can be applied to examine the data captured in each field, looking for similar patterns in data values. When two fields in separate documents have the same pattern of data values collected, this provides strong evidence that these fields can be mapped together, as they are capturing the same information.

Applying natural language processing and machine learning on top of RPA is a bridge to the enhanced form of augmented intelligence—strong augmentation.

Strong Augmentation

Strong augmentation is needed when the existing business processes are no longer enough to differentiate how a company conducts business. In strong augmentation, a business wants to be able to change business processes based on the outcomes of sophisticated predictive analytics. The more complex and sophisticated the data is, the more important it is to provide human knowledge workers with the context they need to make well-informed decisions. Some tasks that are straightforward can be done with machine automation. However, other situations will provide the knowledge worker with a variety of choices. At this time, the knowledge worker looks at the options, which may include the percentage certainty that the option is well suited to solve the problem. In other situations, the selection is not as clear-cut. This requires that the expert examine the underlying analysis of recommended next steps. Strong augmentation is intended to provide guidance—not to simply automate a process. Why? Simply put, a professional with deep knowledge will lack the ability to ingest all the complex data in a field of study. Therefore, to have access to data to support decision making can make the difference between success and failure. Strong augmentation helps a business to redesign an entire process via a new mix of tasks that gets the work done in a new way via machine–human collaboration. For example, rather than simply automating manual processes, strong augmentation applies machine learning techniques to help humans reimage a complex business process in new and innovative ways. This would not be possible to achieve with machine learning alone or human experts in isolation.

Strong Augmentation: Business Process Redesign

In strong augmentation, the process flow is changed to better leverage the synergies of human–machine collaboration. By taking advantage of machine learning and natural language processing, the techniques that can be used to get better results can improve process execution.

To redesign a business process, you need to understand the current business process and have insight on how machine intelligence could be applied to improve the process. Hybrid professionals have the benefit of being able to combine a business process and intelligent technology expertise. These professionals are in the best position to recommend business process changes that can exploit the potential of augmented intelligence.

To understand strong augmentation within a redesigned business process, let’s examine two processes in the operations of the MoneyMaker Mining Company that have been changed through the use of augmented intelligence. The first process is predictive maintenance, and the second is predictive sourcing.

Though the company is fictitious, the examples reflect the actual experience of companies that are deploying augmented intelligence to change the way work gets done and to achieve better outcomes than were possible before. In each case, the process has been rethought so as to enable human–machine collaboration.

Strong Augmentation for Predictive Maintenance

You work in the maintenance department of MoneyMaker Mining Company, responsible for the maintenance of the heavy equipment required to drill for valuable minerals in the mines. The company is headquartered in Santiago, the capital of Chile, while the drilling operations are over 1000 miles to the north near the Bolivian border. When a critical piece of heavy equipment fails, output is reduced until the equipment is repaired. If parts are not available locally, it can take days to get the needed parts, and the missed output is costly for the company. It is a key priority for MoneyMaker to improve the equipment-maintenance process. They commit to invest in predictive-maintenance applications so that they can get advance warning of impending equipment failures. This example illustrates the intersection of process, technology, and people that is a hallmark of augmented intelligence.

A new business process for equipment maintenance was established to provide assistance to the maintenance people in the field and the planners at company headquarters. The new process, built on machine intelligence, is composed of the following steps:

  1. Data Acquisition: An array of sensors collects data from a range of parts in the equipment over a period of time.

  2. Data Exploration: Analytical methods, including machine learning, mine the sensor data looking for patterns, searching for signals that indicate a likely future equipment failure.

  3. Prediction: The pattern is expressed as a prediction, such as, “When sensor reading x is observed, failure of equipment due to issue y is likely within the next 2 weeks.”

  4. Recommendation: The prediction is associated with a recommendation for action to be taken by the human maintenance specialist close to field operations. For example, the recommendation could state: “Change part within the next three weeks to avoid equipment failure due to an electrical short circuit.” The recommendation could also trigger a request to ship the needed part to the remote field operation.

  5. Monitoring: Data coming from the equipment’s sensors is monitored on an ongoing basis, looking for signals of imminent failures. When this condition is encountered, the recommended action is communicated to the maintenance specialist.

  6. Action: The field maintenance specialist (working at the drilling operations site) approves the order to ship out the part (per the recommendation) and replaces the part when it is received within the timeframe indicated.

  7. Monitoring: Through a combination of machine collection and analysis of sensor data with human oversight, the status of the equipment is monitored to ensure that the fix was successful.

A related process (equipment-maintenance scheduling and planning) is the responsibility of the maintenance planner at company headquarters. Aggregating data from the operation of the equipment provides feedback on the current maintenance plan and schedule. Based on an analysis of this information, the maintenance schedule can be adjusted within the limits contained in the associated financial plan. The goal is to maximize the impact of scheduled repairs to address the most likely causes of equipment failure.

The entire maintenance cycle, from pattern discovery to planning coordinates activities, is performed by a combination of humans (maintenance specialists and maintenance planners) and intelligent machines. From the perspective of the maintenance specialists and maintenance planners, the revised business process changes the mix of tasks they are assigned. To realize the benefits in the maintenance operation, field maintenance workers and headquarters-based planners must acquire new skills to be able to interact effectively with machine intelligence and implement the machine-generated recommendations.

Strong Augmentation for Predictive Sourcing

You work in the procurement department for MoneyMaker Mining Company at headquarters in Santiago, Chile. Your company participates in a business-to-business trading network (SAP Ariba is one such business-to-business network), which brings together suppliers of goods (i.e., sellers) with consumers of goods (i.e., buyers). You are a specialist with expertise in certain types of heavy equipment required for the mining operations.

The current process works in the following way. When a request comes into procurement from the field-operations group, it is routed to the appropriate specialist. The specialist then researches the request to find a supplier, using one of the following methods:

  • Search on the network and select the supplier with the equipment in hand, the lowest price, and the earliest delivery date.

  • Send an email to other specialists to see if any of them have experience with the supplier you found on the network. Also, check out reports from the field, looking for any reported issues on the quality of the equipment produced by the supplier and the timeliness of their delivery.

  • Check out reviews of the suppliers you are considering and select the supplier with the highest overall score.

Each of these research strategies can yield data that is relevant to your decision. But such an elaborate process can be excessive, especially when sourcing items that are reliably available from a large number of suppliers. And when sourcing routine, easily available items, this level of effort may not be necessary.

Procurement specialists can redesign this process to reflect differential treatment for a more critical versus a less critical purchasing request. The new process uses machine intelligence to assess the incoming request based on the supply risk (many or small number of qualified suppliers) and financial risk (how critical is this purchase request to a company’s profitability). Sourcing paper clips is routine (with low cost and many suppliers) and, therefore, a good candidate for machine-led automation, whereas sourcing a critical piece of equipment (with high cost and few suppliers) is not. These two dimensions of purchasing risk— financial risk and supplier risk—were first developed by Peter Kraljic in 1983.1

Combining these principles with the capabilities of machine intelligence, the purchasing process could be redesigned as follows:

  • Low supply risk, low financial risk: When the purchasing request, as evaluated by machine intelligence, is not critical (both supply and financial risk are low), a purchasing request is considered routine. This request can then be processed autonomously by a machine without the intervention of skilled procurement professionals.

  • High supply risk and/or high financial risk: When the purchasing request, as evaluated by machine intelligence, is more critical (high supplier and/or financial risk), an approach that combines human and machine intelligence is needed. Machine intelligence builds a prediction by scoring capable suppliers, then delivers a recommendation to the procurement specialist who decides how to source the needed goods or services.

In building a score to evaluate candidate suppliers, the algorithm selects those factors that have been demonstrated to be most impactful in predicting a supplier’s future performance. The data sources for these factors include transactions in the network, text reviews of supplier performance, and reports within your company on the performance of candidate suppliers for the needed equipment. The scoring algorithm uses methods such as machine learning, predictive analytics, and natural language processing.

Automating a subset of the tasks performed by a procurement specialist does not replace all of the skills, tasks, and responsibilities of the job. But the availability of machine intelligence changes the job, enabling specialists to concentrate on the tasks that humans are better suited to perform than machines. In the redesigned procurement process, the procurement specialist will have time to refocus on critical sourcing tasks that benefit from the human touch.

One key task is maintaining relationships with key suppliers who are the sole or nearly sole sources of critical equipment. Another task is reviewing the terms of a written contract. For contract review, there are now intelligent programs that use natural language processing to read each clause of a contract, understanding the terms and conditions expressed by this language. Routine terms and conditions can be automatically approved. But if issues are discovered, they can be flagged. The review of these exceptions then becomes the responsibility of the human procurement specialist.

We’ve considered two examples of augmented intelligence in a business process—predictive maintenance and predictive sourcing. Predictive maintenance is a process within business operations—a process that enables a business to produce or deliver its products or services. Predictive sourcing is another process within business operations. But sourcing is closely linked to accounts payable, one of the group of processes within financials—processes involving monetary transactions and planning. The third major process area for an organization comprises processes related to people. Predictions concerning people have unique characteristics that we will examine next.

Augmented Intelligence in a Business Process about People

What distinguishes the application of augmented intelligence in people-centered processes is that these predictions are psychologically predictive. This type of prediction takes into account data reflecting past human behavior under specific conditions and seeks to predict future human behavior under similar conditions. Where would psychologically predictive insights be useful? Marketing is a critical process area that is being fundamentally changed by machine-based predictions.

Strong Augmentation for Predictive Digital Marketing Campaign Management

Traditionally, retailers used direct mail campaigns seasonally, to either target all of their customers, or major customer segments. There were often fall, winter, spring, and summer sales, occurring on predefined calendar dates each year. The mailings went out to the list of existing customers who had already done business with the company. Sometimes the customer list was expanded by purchasing additional lists from data brokers and then sending the campaign materials to these people as well. The process of delivering the content to a customer list or lists was fairly routine, with many opportunities to automate select steps via robotic process automation.

But the move to digital marketing has dramatically changed the process of marketing campaign management. Retailers now can target individual customers (not just segments or groups of customers). Personalized content can be delivered directly to an individual customer’s online environment in the context of their real-time online activities. But what content should be delivered at what time to which person? When marketing professionals create and deliver personalized content, they need to be supported by psychologically predictive algorithms that are operating at the level of an individual customer.

A predictive digital marketing campaign management process:

  • uses data collected on an individual’s shopping and buying habits, combined with data on where the individual lives and who he or she associates with;

  • anticipates how individuals will respond to specific stimuli such as targeted online messages;

  • guides the development of content and the testing of the effectiveness of that content by creative marketing professionals; and

  • drives the rules for automated delivery of content to customers based on their demographic and psychological profile.

A predictive digital marketing campaign management process requires human–machine collaboration. The scale required to deliver personalized communication utilizes machine-based psychological predictions and automation. But the creative concepts behind the campaign require human involvement. Humans must not only create, but also test, monitor, and review the campaign to ensure that the results meet campaign goals.

Humans also have the responsibility to govern the process. The data privacy rights of individuals must be respected. A general opt-in for any and all uses of personal data will not comply with the European Union’s influential General Data Protection Regulation (GDPR) standard. Marketers must ensure that the proper permissions are granted by the consumer covering specific actions to be taken based on the use of their personal data.

The requirement for obtaining permissions in advance severely curtails the practice of purchasing lists of additional individuals to target. After all, if you have no pre-existing relationship with a customer, you would not have had the opportunity to obtain permission to direct your content to these individuals. Achieving sufficient levels of governance is essential for determining which traditional marketing practices can be retained or need to be revised in light of the new data-driven approaches to marketing campaign management.

Redefining Fashion Retailer Business Models with Augmented Intelligence

We’ve considered how augmented intelligence can improve the efficiency of business processes. Applying weak augmentation, routine tasks within a process can be automated via robotic process automation. Applying strong augmentation, business process steps are redefined to take advantage of machine intelligence with algorithmic predictions and natural language processing.

But we are only at the early stages of leveraging the full impact of augmented intelligence in supporting business goals. In an era of business disruption, where new competitors are emerging to challenge market leaders, core business models are changing dramatically. Let’s examine two companies in the fashion retail business that approached business-process transformation in different ways. Both The Gap, Inc., and Stitch Fix have incorporated machine-generated predictions of fashion trends in the delivery of goods and services to their customers via human–machine collaboration. These changes reflect two different business models and two different views on what humans do best and what machines do best.

Business Model Changes at The Gap, Inc., Using Algorithmic Fashion Predictions

The creative director in a fashion retailer has the task of deciding on the direction of the product lines for the future. What styles will catch on with consumers? How can forward-looking design get attention in a crowded marketplace? Fashion directors must combine experience in past trends with future insight. Historically, great designers were able to attract a following among buyers by anticipating their changing taste based on their own creativity and ability to set trends. For a fashion retailer, the creative director’s decisions cascade down to multiple teams, who are responsible for bringing the fashion line to market.

With the growing use of machine-generated predictions, it wasn’t long before this technology began to be applied to the realm of fashion. The challenge in applying machine intelligence to anticipate and predict fashion trends can change suddenly. So training a predictive model using last year’s data may not yield a great prediction for next year’s fashions.

The Gap, Inc., is a long-time fashion retailer, founded in 1969, which had seen declining sales in recent years, especially for the marquee Gap brand. The company also owns Banana Republic and, until recently, Old Navy. In response to the sales downturn, Art Peck, the CEO of The Gap, Inc., decided in 2016 to dismiss the creative directors at each of the company’s brands. He told The Wall Street Journal that creative directors were “false messiahs,” implying that their predictions were not a reliable guide to fashion trends and to critical merchandising decisions for the chain’s stores.2 In place of the creative directors, Peck brought in a data science team to develop algorithms to predict fashion trends, setting the direction for designers and merchandisers at the company.3 At the same time, Peck instituted a process to ensure that the supply chain suppliers (moving manufacturing from Asia to the Caribbean) would be able to get products to market faster, given the volatility of fashion tastes by consumers.

The switch from relying on humans (creative directors) to relying on machines (algorithms) to predict fashion trends was part of a major business model change for the company. Moreover, relying on machine prediction rather than human prediction represents a radically different view on what machines do best and what humans do best. This reliance on machine-based prediction led to a rethinking of how humans and machines collaborate at the firm.

Have the business model changes been effective? Since the time of the CEO’s decision, the Gap brand has had declining year over year sales for nearly every quarter. Of the major brands, Old Navy has achieved the best results. Finally, a decision was made in February 2019 to divide the company into two pieces, separating the Old Navy brand from the other brands of the company (the largest being the Gap and Banana Republic).4 It is not yet clear about what further adjustments to the business model for the Gap and for the new company will be made as each seeks to optimize the mix of machine versus human labor in the production and delivery of clothing fashions.

Another Fashion Retailing Business Model Using Algorithmic Predictions: Stitch Fix

Stitch Fix is a fashion retailer that was founded in 2011 and went public in 2017. Like the Gap, it combines algorithms with human judgment in the delivery of clothing to customers on a subscription basis. Stitch Fix had a very different business model from the Gap. There are no retail stores. Instead, customers subscribe to a service in which they receive five items of clothing in a box shipped to their home on a periodic basis. The customer selects which clothes to keep and which to send back.

Algorithmic predictions drive recommendations of what clothing to send to a customer in their box, initially based on a survey that each customer fills out when they begin a relationship with Stitch Fix. As a customer begins to make selections to purchase from their subscription, the data on sales is captured and used to tune future predictions that drive future selections on merchandise for the customer’s box. There is also an integration with Pinterest, where customers can post their clothing collections. This provides an additional source of personal data for developing future versions of the algorithm, tuning the prediction to reflect the tastes of individual customers.

There are two key roles in the process:

  • Data scientists build and maintain the predictive algorithms that drive recommendations on what clothing to send to each customer.

  • Personal stylists combine the insight gained from the algorithmic prediction with their own knowledge of the customer to guide the merchandise selection for the box. The stylists then continue to engage with customers to build business and aid retention.

As reported in 2018, Stitch Fix employed 75 data scientists and 3,000 personal stylists.5

The redesigned process features strong augmentation for the successful execution of the process. The model for human–machine collaboration at Stitch Fix (algorithm–stylist–customer) is a redesign of the merchandising process from that at a traditional retailer. The unique Stitch Fix method of delivering a personalized collection of merchandise to a customer could not scale without machine-aided predictions and recommendations at the level of the individual customer. But the process also requires a human in the middle (between the machine prediction and the customer) to provide the personal support and customer relationship to make this subscription model work.

Hybrid Augmentation

Of course, not everything is black and white—shades of gray are often the norm. Therefore, it is not surprising that in many situations, an organization will want to use a combination of weak and strong augmentation to support the same set of processes. This hybrid approach is required when implemented business models change. It is possible to automate simple manual tasks in the context of an overall redesign. To understand how these different approaches to intelligent augmentation work, we will delve deeper into what is needed to transform businesses. Let’s consider these approaches as ways to best leverage human and machine capabilities together within a business process for achieving the best results.

Summary

Applying augmented intelligence to a business process is a path to process improvement. The type of improvement depends on the approach taken:

  • Weak Augmentation: In this approach, several steps or tasks of an existing business process are automated without changing the workflow of the process itself. This technique is well suited to repetitive transactional processes, such as invoice handling. These processes typically feature routine but labor-intensive tasks such as filling out a set of forms and documents. Robotic process automation can be applied selectively to several tasks, reducing the cost and improving accuracy for the end-to-end process. Measuring the costs to run the process before the automation is implemented sets a baseline against which the post-implementation process costs can be compared.

  • Strong Augmentation: In this approach, the task flow in the process is redesigned so that the work is done in a new way via machine–human collaboration. This technique is well suited to processes that involve judgment, such as deciding on what to offer an individual customer using knowledge of their preferences. Machine-generated predictions and recommendations enable this process to scale with humans applying judgment to review and adjust the recommendation. This approach can yield higher revenues. Setting up a benchmark for before-and-after comparisons is important.

  • Business Model Change: This hybrid approach goes beyond process improvement but seeks to change the fundamental business model of the organization through the application of augmented intelligence. The goal is to deliver goods or services in a new way, which becomes the firm’s competitive advantage.

These scenarios for augmented intelligence are not mutually exclusive. An organization can take advantage of weak augmentation, which substitutes machines for human labor for simpler, routine tasks. The result can drive down costs. With strong augmentation, tasks involving judgment can be improved via machine-generated predictions. The resulting recommendations can be reviewed by human agents, with the goal of driving top-line increases to revenues. as needed. Finally, the biggest potential is for an organization to revise their core business models, finding new ways to deliver goods and services via human–machine collaboration, which provides a competitive advantage in the marketplace. The most advanced organizations utilize all three of these augmented intelligence strategies.

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
3.144.10.69