It’s no longer rare to see machine learning (ML) models used to support a variety of business decisions, from whether a financial transaction should be sent to the fraud investigation team, to what discount a distributor should get.
Still, even in organizations that have embraced ML-based systems, it’s common for business problems that could benefit from machine learning to be solved using a less effective (and often more costly) approach.
As a business analyst, it’s useful to be aware of some signs your business problem might benefit from ML:
1) The answer to a business question is buried in unstructured data.
Imagine that you are working on an initiative to reduce the volume of IT support calls by employees.
Knowing which requests are more common and taking more time from call agents will allow the organization to prioritize new self-service tools with the most impact on reducing support call volume. But the reasons for the calls answered by agents are logged as free text when each agent closes a ticket. Similar password reset requests may have been logged using different wording, such as, “User forgot password, I reset it for him.”, or “User couldn’t log in. I created a new password, and she confirmed it’s working now.”
Trying to manually classify tens of thousands of support call logs would be very time-consuming. And a rules-based solution is likely to generate many mistakes. For example, it might fail to categorize a log that says “User couldn’t log in, I used the admin reset option to fix that” as a password reset request because it doesn’t contain the right matching keywords. In contrast, natural language processing (NLP) tools that are pre-trained on vast collections of documents can dramatically reduce the work required to accurately group tickets based on user intent: password reset, gain access to a new system, troubleshoot email issues, etc. Even a data analyst with basic coding skills should be able to leverage a language model to automate the task.
This situation is only one of many scenarios where unsupervised machine learning methods leveraging NLP can help improve business decision-making by finding answers in unstructured data. Other opportunities include mining email messages ("What percentage of employee emails is about questions whose answers are available on our knowledge base?"), nurse notes ("How many of our patients are smokers?"), employee survey data ("What are the top five complaints mentioned by employees in internal surveys?”).
2) Your organization is struggling to improve the performance of a rules-based decision service.
Rules-based decisions are a component of many business processes. When a rules-based decision model is working, there is no reason to change. But sometimes, institutional knowledge loss, changes in regulations, policies, market conditions, or customer behavior may cause existing business rules to age and lose their effectiveness.
You may have seen situations like these happen in your organization:
- A lead scoring model that relies on a wide range of data points (company size and revenue, videos watched, etc.) to determine propensity to buy starts losing accuracy, causing the sales team to waste time pursuing the wrong prospective customers.
- The rules used to guide manual decisions (which next product to recommend to a customer considering regulatory restrictions, whether to approve or deny a medical claim), become so complicated that it’s impossible to keep up with the demand for trained employees to perform the task.
- Workers in key positions retire, taking with them critical information about the business rules used to correctly process certain types of orders under regulatory restrictions, price new products to ensure they become profitable, etc.
When rules become too difficult for humans to understand, enhance, or enforce, machine learning—which has some capabilities that rules do not—may offer superior results. ML-based models can take vast amounts of data into account to foster a probabilistic view of the world that often is the most accurate way to view frequent, tactical decisions that benefit from modeling, like the examples above. And with the proliferation of open source libraries and user-friendly ML tools, experimenting with machine learning to replace rules-based decision systems may be cheaper and produce a higher ROI than more traditional approaches.
2) You have repetitive work that can be delegated to a generative model like ChatGPT.
A few days ago, I was asked to help an executive document the requirements for a secure data capture tool. As I was starting to write some sample requirements statements to highlight what should be considered, I remembered I could delegate the task to ChatGPT.
ChatGPT is a language model from OpenAI that uses a dialogue format and is designed to follow an instruction in a prompt like “write a set of requirements for X.”
Sure enough, ChatGPT was able to generate a sample requirements document that saved me significant time. Rather than having to write the sample from scratch, I merely took the model output and quickly adapted it for my purposes (for example, changing the verb from “should” to “must” for a more precise communication).
An excerpt of the list of requirements generated by ChatGPT based on the short prompt at the top.
In another project, the same generative model was able to take a list of competencies like “Data analysis: the ability to collect, organize, and interpret data in order to draw insights and make informed decisions," and quickly produce 4 levels of proficiency each, eliminating the need for hours of tedious manual labor.
Here ChatGPT did an impressive job generating four levels of proficiency for various functional competencies.
As I wrote in My Favorite use of generative AI so far, tools like GPT-3 and ChatGPT are not only good at automating repetitive tasks, but also at teaching us how to write code and software requirements, offering useful definitions for technical terms without the need for sifting through online search results, and enabling a fully personalized learning journey. By keeping such models in our “BA toolkit”, we can help our organizations become smarter and more productive.
# # #
In essence, business analysis is about affecting change: helping an organization go from A to B, from an existing situation to a preferred one. And since all business processes run on data, it pays off to cut through the hype of AI/ML to understand its real potential to deliver business value.
Under the right conditions, machine learning can be a key piece of the optimal solution to achieve business objectives. By paying attention to the three signs above, you can broaden the opportunities for your organization to add ML to the options on the table. Whether you’re trying to extract meaning from large volumes of unstructured data, extract business rules from undocumented or complex business processes, or automate time-consuming, repetitive tasks, ML can be the key to drive innovation and elevate the quality of business decisions.
Author: Adriana Beal
Adriana Beal has been working as a data scientist since 2016. Her educational background includes graduate degrees in Electrical Engineering and Strategic Management of Information obtained from top schools in her native country, Brazil and certificates in Big Data and Data Analytics from the University of Texas and Machine Learning Specialty from AWS. Over the past five years, she has developed predictive models to improve outcomes in healthcare, mobility, IoT, customer science, human services, and agriculture. Prior to that she worked for more than a decade in business analysis and product management helping U.S. Fortune 500 companies and high tech startups make better software decisions. Adriana has two IT strategy books published in Brazil and work internationally published by IEEE and IGI Global. You can find more of her useful advice for business analysts at bealprojects.com