24 Sep 5 Key Intelligent Automation Concepts To Keep In Mind As You Plan for 2019
If you’ve been following our recent blog posts, then you now know what Intelligent Automation (IA) is, you understand how to build a strategy for this type of emerging technology, and you have reviewed relevant use cases to help you conceptualize how IA can work within your organization – and now you’re ready to get started with your own IA initiative. With planning and budgeting for 2019 well underway, this is the appropriate time to prepare to implement these powerful capabilities so that you can reap their benefits in the year to come. As you develop your plan and prepare to allocate spend, there are some key IA concepts that we recommend you keep in mind.
1. Focus on value that is immediately achievable.
It can be easy to get carried away with the possibilities that IA can offer your business, but you should maintain focus on those that will create the most value for your business. We recommend starting with a few simple use cases, gathering data, and proving their impact to stakeholders. By starting with a proof-of-concept (POC) and then expanding to a minimum viable product (MVP) or pilot launch, you can establish value, gain internal buy-in, and then move on to more complex opportunities.
In our experience, the root cause of many failed IA initiatives is starting with a highly complex use case. Solving the most difficult issue facing your business is great, but building up to that challenge will help you gain momentum and credibility so that you can leverage your experience and stay within the defined budget and timeline when your program is ready. The most difficult problems are tough for a reason; no one reasonably expects you to solve them from the start.
2. Data labeling is critical – and manual.
Intelligent systems can’t teach themselves. Think of IA systems as children. They will do exactly what you teach them to do, if you teach them clearly, consistently, and broadly. Data labeling is a necessary part of the teaching process, whereby you identify exactly what data your IA system will interact with so that it can make the decisions you expect.
One of our clients recently implemented an automated customer support bot. We worked with them to examine thousands of existing customer phone records, emails, and chat transcripts. We then enriched those interactions with relevant customer information and labeled each data point with a specific type of customer intent and suggested response. We used that labeled data to train the bot, and it “learned” how to triage and respond to future issues.
Data labeling is a labor-intensive, manual process, but it’s critical to IA success. It’s also important to recognize that this is not a one-time activity; after your launch, you must monitor the system’s actions to see if it provides the expected outcomes – and potentially reexamine your data and retrain your system to make better decisions moving forward.
3. Transfer learning is years away.
Machine learning is when a machine learns from patterns in data and modifies the way it handles future interactions based on those learned patterns. Transfer learning is when a machine takes that knowledge and transfers it to a different problem. Humans do this every day but intelligent systems simply can’t, at least not yet. In our experience, a lot of people expect transfer learning to be standard practice. But because intelligent systems are solving for specific things, it’s difficult to make that data directly applicable to another situation. At least for now, transfer learning is years away, so don’t expect to leverage one system’s training for another.
4. Traceability is complicated.
We train intelligent systems to look for specific pieces of information and then to make decisions based on that information. The conditions for each decision are often unique. When auditing decisions and outcomes, it’s difficult to exactly recreate the historical event. If your IA system deals with decisions that could be controversial (e.g. mortgage lending), you’ll need to take extra time to create fully auditable decision records, so that you can trace the chain of events leading to a decision.
5. Becoming an IA-driven organization requires more than just technology.
Technology is at the root of every IA implementation. But technology alone won’t foster a successful IA implementation. An IA initiative can require a major shift in the way your team works, the processes in place, and the technology that you use. A strategic change management plan will ensure that you have the training, executive sponsorship, and organizational structure to promote immediate and long-term adoption, thus enabling your IA implementation to succeed.
If you keep these concepts in mind as you plan for an IA initiative, you’ll be better prepared for success and to answer key questions from senior executives and colleagues about how your system will function. While the impact of IA can be great, that success is only possible with careful planning and executive sponsorship. The Navigate team can help you find real value from Intelligent Automation. Contact me at email@example.com or 484.383.0606 to discuss your organization’s goals and our recommendations for moving forward.