It might seem like automation in the workplace is a relatively new phenomenon. The reality is that it’s been on a steady march through business for more than 200 years. Nevertheless, the past three years or so have seen a huge step-change in the variety of technologies being promoted to help businesses automate aspects of their work.
Automation tools are becoming less expensive, they can be deployed more quickly and are easier to use. Moreover, data to train and improve “smart” systems powered by artificial intelligence (AI) is more readily available.
A new wave of machine-learning-based AI services from the likes of Amazon, Google and Microsoft — particularly focused on automated sense-making and interaction using voice and vision — has started to make an impact in some quarters. A big part of this impact comes from the fact that many of the technologies are offered either free of charge or via low-cost subscriptions. Additionally, we see renewed interest in using machine learning to create tailored, predictive analytics models that operate on structured data and can deliver tailored, context-sensitive recommendations to individual business people in the “flow of work.” Big brands like IBM’s Watson, Adobe’s Sensei and Salesforce’s Einstein operate across these categories.
Determine Your Automation Goals
The new wave of interest in AI and AI-driven innovation looks very seductive against the backdrop of a renewed interest in business automation, but it pays to be thoughtful about finding potential applications.
Given the wide variety of automation-related technologies being pitched, and the speedy pace of technology evolution, it’s vital that you develop an automation strategy that clearly and deliberately sets out how and where you will look to apply different kinds of automation technology capabilities — including AI-based capabilities — and why you are looking to apply different technologies in different places. Despite what some vendors will tell you, different kinds of automation technology are specialized to deal with different aspects of work automation.
Related Article: Automating the Digital Workplace
Establish Your Automation Opportunities
A good way to start mapping out an automation strategy is by looking at the different ways in which work gets done. By doing this, particular opportunities for automation become clearer.
The following three types of work are commonly found “in the wild” in organizations:
- Programmatic work. With programmatic work, almost all features of the work in question — from tasks and decisions to the definition of workflows, work inputs and triggers (that signal work needs to be done), as well as the resources required to complete the work — can be prescribed and designed in detail; expert discretion is only very rarely needed. This doesn’t mean that the work is already automated, but it does mean that in theory it’s possible for the work to be almost completely automated (with people only handling any exceptions that occur). Many types of clerical back-office administration work fall into this category.
- Transactional work. With transactional work, many of the tasks and decisions involved, and most of the flow of work, can be prescribed and designed in advance. The goal of the work is often also prescribed in quite a structured way. However, human discretion and expertise are required to carry out some tasks, and to decide when, how and where progress needs to be made in a particular endeavor. A good example of a process revolving around transactional work is mortgage application processing — where the steps and decisions required are almost always definable ahead of time, but human discretion is often required to review decisions and minimize risks of fraud.
- Exploratory work. With exploratory work, the goal of the work will be defined in advance, but it is probably not possible to make decisions ahead of time about the set and sequence of tasks that need to be done and decisions that must be made, and the people or roles who will perform those tasks or make those decisions. There may be some high-level milestones that are common to a particular type of exploratory work (perhaps to ensure, say, regulatory compliance or management approval) but they provide only a very loose structure. In exploratory work, as the label suggests, the overall experience for both the work participants and the “customer” of the work is that of a set of possibilities being explored, rather than a recipe being followed. Common examples of processes revolving around exploratory work include complaint resolution and fraud investigation.
This framework helps to identify opportunities to apply many kinds of automation technology, but let’s look here at two particular areas where AI can play a role.
Related Article: Robotic Process Automation and Roombas: What Could Go Wrong?
AI-Powered Intelligent Capture
In situations where programmatic work is being done to fulfill requests from customers, partners or suppliers (such as a request for a change in a shipment delivery address), it’s often the case that such requests arrive in a variety of formats over a variety of channels. For example, requests might arrive via email, fax, PDF or a form submitted on a website. Here, you should consider using AI-powered intelligent capture technology to digitize the data in those documents and identify and extract key data fields. For example, intelligent capture technology can automate the digitization of contracts and invoices, and automatically identify extract key terms and items (like contract signatories, subjects and key clauses, or purchase order numbers, invoicing addresses, totals and products, payment terms and so on).
Third-party commercial or open-source AI libraries can enhance the flexibility of this technology by improving the accuracy of image or document classification, and by improving the accuracy of detecting important features in the data that will be used to drive subsequent decisions (they can also improve the later discoverability of assets by enriching document metadata).
AI-Powered Predictive Models
In the context of transactional and exploratory work, where work features can’t necessarily be prescribed through upfront design, there are opportunities to use AI-powered predictive models to make recommendations that guide the flow and results of work. I’ve seen examples of predictive models being used to do the following:
- Recommend the best next actions for workers to take as they work on projects — matching features of the project against those of historical cases to recommend courses of action that are likely to be the most effective.
- Recommend colleagues who can help with a task or a decision, based on the features of the project and the skills of all available colleagues provisioned on the system.
- Recommend guidance documents and procedures that should be referenced by people working on a particular task or tackling a particular phase of the project, based on features of the project and historical patterns in the use of reference documents in previous projects.
In the past, computer systems were so expensive and complex that predictive model generation and maintenance were typically handled by specialized teams using dedicated infrastructure, and companies could only apply such models to business data once in a while. However, in recent years, the technology has improved drastically and the cost of technology has declined to the point where it’s now feasible to apply predictive models in real time, in the flow of work, as decisions need to be made. This means that where predictive models were historically used primarily to do things like shape email marketing campaigns, they can now be used interactively as part of a conversation with customers.
Learn how you can join our contributor community.