/
March 13, 2025
In the current age of Artificial Intelligence (AI) and Machine Learning (ML), most people know that successfully automating a business process within an enterprise can be hard. However, few people realize that even more difficult than the automation itself, is adjusting enterprise-wide policies and procedures to get the most out of your automation process.
Without a proper infrastructure in place, aligning your people, processes, and technology to the world of artificial intelligence and machine learning software solutions can feel chaotic and stressful, making it hard to remember what the benefits of making these changes are in the first place.
So, should you be hesitant to make these changes? Is the juice worth the squeeze? At Blackmarker, we’ve seen that with the right approach and change management strategy, the answer is a resounding yes. In this blog, we’ll lead a crash course on what machine learning is, how it can help you reach your goals, and what steps to take to ensure a successful integration with your organization.
Going back to the early days of data science, the origins of process automation largely stem from something called Robotic Process Automation (RPA). RPA is defined as an application of technology, governed by business logic and structured inputs, aimed at automating business processes.
Put more simply, RPA is the ability to automate keystrokes or clicks within tools on a desktop to replicate the requirements of a human to complete a task. And while over the years RPA proved valuable in many spaces, it certainly comes with its limitations.
Like most other software applications of its time, RPA follows a basic logic pattern of starting with input data, and configuring it to yield an output. This means that RPA is a great solution for predictable, static input data. However, in a world where this rarely occurs, RPA poses a glaring challenge for businesses looking to automate their ever-changing data and analytics processes. With this need, we see machine learning enter the picture.
Unlike traditional software programming where a developer writes a program based on assumptions of inputted information to provide a static output, machine learning focuses on the input information (training data) and expected output information. By focusing on the relationship between the input and output data, machine learning models are able to fill in the gaps with algorithms that provide probability-based answers that adapt with their ever-changing input data.
Rather than a traditional input yields output approach, machine learning observes the relationship between input data and output, and reverse engineers the algorithm to yield said outputs.
Typically thought of as a new technological development, the algorithms for these models actually date back to the 1940s, where software was used to imitate the role of a human in games like checkers and chess. However, it wasn’t until the last 10-20 years that these algorithms were augmented into what we now think of when we hear “machine learning.” With the efforts of companies like NVIDIA, Intel, and Microsoft (inclusive with a ton of open source development), machine learning arrived at the forefront of the data and analytics community.
What about now?
In today’s world, machine learning takes the form of many different use cases, spanning nearly every industry. To look specifically at the use cases of the software to meet the needs of businesses, it helps to look towards the financial services industry.
For banks, credit card companies, and even the federal government, machine learning continues to prove itself incredibly useful in the realm of fraud detection. With the ability to leverage large quantities of training data, machine learning models can analyze thousands of individual records to swiftly scan for fraud indicators above a certain risk threshold, and pass along any suspicious activity to an analyst for further assessment.
Outside of just the financial services industry, we see machine learning impacting enterprises everywhere. Whether through disease diagnosis in healthcare, demand forecasting in retail, or predictive maintenance in manufacturing, it’s clear that the benefits of machine learning touch every single industry, and are here to stay.
Now, with a basic understanding of machine learning and how it can help your process automation, there’s only one question left - how do I integrate machine learning into my enterprise? And you’d be right to ask this question, as our experience shows that this is where most people get tripped up. Even more than the effort to automate a business process, is the effort that comes with aligning the people, processes, and technology accordingly to adopt this change in a manner that allows for the realization of the true value of machine learning. So with that, we’d love to share with you our four key considerations to unlock maximum value in your process automation journey.
To achieve successful automation, we first need to define success. And before we define success, we need a clear understanding of where the issue exists, and what specifically fails to meet the standard. These observations can then be evidenced as metrics, with the goal of baselining the area so that when automation implementation occurs, those same metrics can serve as key performance indicators (KPIs) to measure performance. KPIs can look like time spent on a targeted task, number of keystrokes by a user, percent accuracy, amongst many others.
With a clear understanding of the problem and the desired outcome of the target state, the next step is to think about the specific capabilities that the target state should possess, and develop them into a hierarchy. As people, we want an “easy button” that takes away all of our problems with the first attempt, however with automation (and most things in life) this is sadly not the case.
As a result, we recommend taking a prioritized, incremental approach. Which capabilities are must-have? Which ones are more like a nice-to-have? What constraints do you need to consider? All of these are important things to consider when thinking about technology implementation.
By identifying the most important capabilities, it allows for an iterative, user-centric implementation focused around maximizing value for the organization.
With a clear understanding of your needs and their importance relative to each other, you are ready to choose a software solution. And in this process, we want to drill home one critical piece of advice: end user, end user, end user.
The key here is to gain buy-in from those impacted by the change. This requires assessment by subject matter experts within the enterprise to best map out the process or organizational changes needed based on the impact of the automation being contemplated. Once the changes are better understood, stakeholders can review the high level plan that describes the baseline, the automation to be assessed/deployed, and how it will impact their processes and teams.
The most prevalent issue a team may have with this sort of change is the sense that a job/role is going away. In our experience, with proper change management, staff can be upskilled and redeployed as their complex but repetitive tasks (like redaction) or automated. Staff can focus on the more interesting edge cases rather than the mundane.
AI is a tool to help humans complete tasks more efficiently
In a previous article, we discussed process change in more detail as it pertains to automated redaction with Blackmarker, along with a process/personnel example of this type of adaptation portrayed in the movie Hidden Figures. Not a fan of American historical dramas? Not a problem! We have another cinematic analogy for all of the Marvel fans out there - Iron Man’s suit.
Think about the suit. It contains seemingly endless power and potential, but it’s useless on its own. It’s not until Tony Stark takes the helm and aims the powers of the suit to fight evil that he realizes the true value of the suit. We feel that artificial intelligence and machine learning work the same way.
Once goals and solutions are clear, the final step is breaking work into tasks. We recommend an iterative approach - tracking development using Agile methodology.
To achieve this, think about the tasks required for the initial introduction of the implementation - do not worry about follow-on work, as the entire effort will collapse if the first step fails. The task list serves as a backlog, where you start with the most important tasks, and subsequently work through lower-priority subsets in defined intervals. An interval typically spans around two weeks, and is called a “sprint” in the software implementation world. Over each sprint, you can look for ways to prove out progress with stakeholders in the form of demos, allowing for feedback in real time to make adjustments, rather than waiting a prolonged period of time to find out about any potential misses.
A successful automation effort depends equally on the internal planning and on the external solution. When choosing an automation solution and an integration partner, here are a few traps to avoid:
Be aware of overly-simple approaches that make all of your problems look like “nails”. Your business is unique, and the technology adopted should reflect that. In other words, the vendor will need to bring their entire tool kit, not just the hammer!
Another term for this is the “land and expand” approach where a vendor introduces a general tool in order to stay entrenched in your business for what feels like forever, tailoring it to your specific needs.
The benefit of a machine learning approach to automation lies in the minimal “set up” work required of the users within the enterprise. If a technology requires a lot of work to set it up on the front end, one should question if the vendor is really employing machine learning techniques at all. A key sign of the more traditional software programming approach to automation, as we discussed earlier, is the need to put up a lot of configuration or other data up front. If the technology vendor uses terms like “profile” or “rules,” this is usually a sign of a less-than-optimal machine learning-based automation.
While configuration or profile data in large amounts is a warning sign, machine learning-based automations do require labeled sample or training data to be as accurate as possible. For most enterprises, the technology should come with pre-trained models to avoid this or the training happens within the tool to handle this work for the enterprise.
As we suggested previously, the automation journey is an iterative one, and the early steps in that journey should leverage existing models. A characteristic of a machine learning automation technology is the ability for users to label new data when warranted. As noted, this can arise when new data exists from the models previously created by the vendor. Staying with our movie analogies, this feedback loop within the technology allows Tony Stark to drive the suit when new conditions arise.
Lastly, given the amount of computing power required to run sophisticated machine learning models, you should look for technology hosted in a cloud-based manner. Among many benefits, this removes the burden of the enterprise to have to stay up to date with ever-evolving computing infrastructure.
Another pro of this approach is that scaling should not be a concern for the enterprise (nor the vendor), as current cloud technologies handle this automatically. From a business perspective, this typically evidences itself in a usage model versus a per-subscriber or flat fee model.
Given Blackmarker’s origin in the artificial intelligence and data science consulting space, it should come as no surprise that our company serves as an example of a machine learning automation technology. Below we’ve outlined some of the difference-making characteristics of Blackmarker that tie back to the topics discussed in this article:
As businesses continue to evolve, automation is no longer a luxury—it’s a necessity. While RPA has helped streamline repetitive tasks, machine learning takes automation to the next level by enabling systems to learn, adapt, and improve over time.
The key to unlocking this success lies in the strategic implementation. In a world with many nuances and challenges in the implementation of enterprise automation, we have seen that if handled properly, machine learning adoption can operate smoothly with immense benefits to an organization. Here at Blackmarker, we provide all of the necessary resources and experience to position enterprises to not only be prepared for these changes, but use them to drive real business value.
Interested in learning more about Blackmarker and the services we provide? Click here to talk with us or request a demo!
