Four Considerations When Pursuing Intelligent Automation

Whether starting off with automation or maturing your existing automation practice to pursue more high value projects, there are four key factors that will be critical to your success.

  1. Complexity
  2. The Level of Automation Delivered
  3. Effort Needed to Execute
  4. Value – Is it Worth It? 

1. Complexity

There are plenty of legacy AI and OCR vendors and open-source machine learning models that can be used to brute force the processing of a fairly static, one-page document when you only need to pull three to five data points. It may not be pretty, reusable, or easy to maintain, but with a little patience, it will be automated.

There are VERY few automation platforms that give you the ability to reliably automate complex use cases. Example: Extracting twenty fields (name, account number, etc.) from ten widely varied document types with structured, semi, and unstructured data, all of which are over twelve pages in length.

Here’s what really matters when gauging the complexity of an intelligent automation project:

  • Number of different types of documents
  • Document variability – does the format change significantly from doc to doc
  • Document quality – crystal clear to illegible
  • Average document length
  • Number of fields that need to be captured within the documents

Serious vendors with solutions capable of driving meaningful automation within complex use cases rely on a comprehensive suite of AI technologies within their platforms and are powered by GPUs. Successful intelligent automation practices require:

2. The Level of Automation Delivered

One of the obstacles standing between meaningful value and lackluster results is the true level of automation delivered. In a perfect world, a set of documents or images would be flawlessly processed by an automation every time, which is known as 100% straight through processing (STP). 100% STP is almost always impossible within intelligent automation. Poor document quality, errors or missing info within the document, and human error are all stumbling blocks, but there is an even larger challenge.

Imagine you need to accurately identify and extract five pieces of information (fields) from a document. Your “ACME AI Solution” manages to confidently capture each of those fields correctly 80% of the time. At first glance, it’s easy to assume that you will have 80% straight through processing, with a person only needing to be involved 20% of the time to help process any exceptions. In reality, that automation is likely to have only a 33% STP rate, with a person helping out with 66% of the documents. This is because the probability that the AI captures ALL 5 fields correctly would be 80% x 80% x 80% x 80% x 80%, which is 32.7%.

This raises two important points about intelligent automation:

  • “Close” isn’t good enough. In the example above, if you were using an AI solution that provided 95% accuracy instead of 80% for each field, the STP would 77%. That’s over twice as much automation value from the same project. This is one of the reasons why it’s worth pursuing a true intelligent automation leader & innovator when selecting a solution.
  • Make sure you are factoring in people. Even if the AI can only confidently capture some of the required fields and a person needs to be involved to help identify or confirm the remaining fields (aka an exception), it’s still much faster to manually find and capture half the fields in a document vs all the fields. Straight through processing is the gold standard within automation, but the efficiency gain offered by intelligent automation to existing workers is also a major part of the value of the solution. This is why having good “human-in-the-loop” functionality within your intelligent automation solution is critical. The more seamlessly existing employees can work with intelligent automation, the more value the solution offers.

3. Effort Needed to Execute

The difficulty of delivering and managing intelligent automation efforts come down to a few factors, all of which determine whether your projects are joyful triumphs or drawn-out headaches.

Build, deploy, & maintain: How much work will it take to customize an intelligent automation solution to meet your needs, how hard is it to get the solution in place and how much work is it to keep the solution running.

  • Build: With intelligent automation, the primary factor in the build phase is training the AI model(s) to excel at handling your specific use case. Most intelligent automation vendors have pre-built models trained for different types of projects, but even then, they usually need some additional training to deliver optimal results. The faster the AI can learn; the less time and effort is required to build. Here are the two factors that drive AI model training/learning:
  • Data Intensity: How much labelled (or gold) data you need to run through the model before it can reliably get the job done. The more data that needs to be curated and fed through the model, the more time and effort is required. Generally, the less data an intelligent automation platform needs to train a model, the more advanced the platform is. Less data required upfront also means less data will be required down the road when your processes change or evolve.
  • How the models are trained: There are two different ways automation models are trained: supervised and unsupervised learning. Any intelligent automation solution you select should use both of these training methods in tandem to optimize the time-to-delivery and the resiliency of your automation solution.
  • Deploy: Thankfully, between open APIs and RPA, it’s easy to inject an intelligent automation model into most processes regardless of the surrounding applications. As long as your intelligent automation solution is using modern protocols and keeping up with the times, this is unlikely to be a stumbling block for your projects.
    This leaves only one major consideration when thinking about deploying an intelligent automation solution, which is where to install it. Each organization has its own preferences when it comes to on-premises vs the cloud (aka: Software as a Service – SaaS). Generally, installing technology on your company’s own hardware or virtual private cloud (VPC) provides more direct control over security but is also slower and more expensive to scale and maintain. Conversely, deploying in the cloud is faster and easier but provides your organization with less direct control over security. Regardless of which approach you choose, be sure to address this topic early and find a vendor that best aligns with your preferences.
  • Maintain: Maintenance is a particularly important consideration when thinking about intelligent automation. Business processes and documents are dynamic and continually evolve. Any change could break the automations you have in place. Fortunately, a well-engineered intelligent automation platform can turn major re-work projects into minor tuning exercises.
  • Minor changes: When using a poorly designed solution, any change to the document breaks the entire process. The automation has to be taken offline while the models are retrained. Conversely, a well-engineered solution will provide intelligent, flexible models capable of processing minor variations without any human intervention or disruption.
  • Moderate changes: Advanced intelligent automation platforms thrive on continuous learning. As exceptions arise, humans are alerted to correct the model, so it gets smarter; consequentially, exception rates diminish, freeing the human to do more valuable tasks. With less sophisticated platforms, even with a human-in-the-loop, models are incapable of real-time learning and exception rates remain static.
  • Major Changes: When documents, processes, and requirements inevitably change, data intensity and the combination of supervised and unsupervised learning become paramount. With less advanced automation tools, these changes will take you back to square one and will require an entirely new automation cycle. With next generation solutions and their sophisticated use of unsupervised & supervised learning, only a small amount of data is needed to turn what would have been a total outage and a new full-scale IA project into a one- or two-week corrective sprint.

World class technology vs people dressed as robots. It is an unfortunate truth that a significant portion of “artificial intelligence companies” in the market today are actually loose product concepts with a fleet of offshore engineers desperately trying to throw together machine learning models, rules, and scripts to deliver each use case. Buying one of these “solutions” might seem like a very low-effort way to quickly build and deploy automations, but first, consider these three factors: Resiliency, Reusability, and Results.

  • Resiliency: Business processes change, and when they do, automation solutions cobbled together behind the scenes by a team of mystery engineers halfway around the world will break and will be unavailable until that group of engineers has the time to fix it—if they can fix it. One minor update to a form or a tweak to a business process and the value of that automation effort will evaporate. It’s very common to see these types of solutions form a heartbeat pattern of uptime, downtime, uptime, downtime and all the while, you’re paying with time and money.
  • Reusability: Well-engineered platforms enable you to leverage previous work on current projects. This speeds along delivery and increases resiliency. Conversely, poorly engineered solutions offer little you can, or would want to, reuse.
  • Results: As we mentioned above, “close” isn’t good enough. The difference between decent automation results and excellent results has a major impact on the value provided by automation projects. Better technology consistently generates better results, regardless of how many engineers are thrown at the problem.

4. Value – Is it worth it to automate the process

This can be a surprisingly hard question to answer, but before you spend time and money on an automation project, you need to be able to answer it accurately. Here are a few simple guidelines:

  • Discover & Validate: It is important to do process discovery upfront and capture reliable data about both the general process (what is the process, what are the steps, why is it important) and details on the area you wish to automate:
  • Frequency: How often does the process happen?
  • Duration: How long does it take to manually go through the process?
  • Applications: Which systems are involved?
  • Employees: Which roles are involved in doing the process manually? Are they exclusively focused on the target process, or do they split their time?
  • Automation Value: Does automating the process
  • Reduce the average handling time (AHT) or improve the customer’s experience?
  • Improve employee job satisfaction or allow employees to focus on higher-value work?
  • Increase accuracy & reduce mistakes?
  • Improve visibility into the overall process or help maintain compliance?

Always validate your initial findings with the end users, process owner, and stakeholders ahead of taking the time to build a full business case. Getting everyone on the same page early helps avoid 90% of the typical pitfalls that automation projects face and solidifies the project scope.

  • Quantify & Validate Again: The value of some automation projects can’t be easily quantified, but whenever possible, do the math. If the value of the automation can’t be directly measured, list out the expected benefits. Clearly articulating the anticipated results of a project, as well as its scope and timeline within the business case, is the best way to confirm everyone is on the same page. Set expectations ahead of a project and measure your success once the project is underway.
    Once your draft business case is complete, review it with the process owner and end users/subject matter experts to again validate your assumptions and expected results. It’s easy for minor misunderstandings during a discovery session to slip through and lead to issues down the road. Measure twice, cut once.
    Finally, when presenting your business case to project stakeholders for final sign-off, confirm their understanding of the project scope. It is easy for the scope, and therefore the expectations, of a project to expand over time. Happiness = reality/expectations. So, firmly cementing project expectations in place before starting a project is the best way to ensure everyone is happy when you deliver on what was promised.

A great way to narrow in on likely high-value intelligent automation opportunities is to learn from the success of others. If you’re not sure where to begin, take a look at a few proven winners for intelligent automation below:

  • Common back-office use cases:
    • Procure-to-Pay
    • Order-to-Cash
    • Record-to-Report
    • Hire-to-Retire
  • Logistics:
    • Custom Declarations
    • Driver Logs
    • Maintenance Logs
  • Manufacturing:
    • Quality Assurance Records
    • Change Requests
    • Sales Orders
  • Financial Services:
    • Mortgage Applications
    • Know Your Customer
    • Insurance Claims
    • Account Openings
    • Check Processing
    • Transaction Statements
  • Healthcare:
    • Physician Referrals
    • Claims documents
    • Patient Records
    • Patient Onboarding
  • Supply Chain:
    • Order Scheduling and Tracking
    • Bills of Lading
    • Proofs of Delivery
  • Finance:
    • Invoice Processing
    • Purchase Orders
    • Financial Spreading
    • Fraud Detection
    • Contract Analysis
  • Government:
    • Immigration Applications
    • Tax forms
    • Education Applications

Latest Post from AYR

About the Author