A CFO’s Machine Learning Journey to Reduce Customer Churn

Thy Enemy Is Churn

As a software CFO, the mandate is to efficiently grow and increase the organization’s net dollar retention rate.   You know the metric; it’s what investors prize and drives a company’s overall valuation.  Boosting revenue cost-effectively with the proper marketing channels is excellent and is often displayed prominently in a leadership group’s executive dashboard.  When net dollar retention grows, it means your customer base is growing, they are buying more licenses or features, and everyone is happy.  And when a company is big enough, you can even stop spending to acquire new customers since the existing base is content to buy more.  The flipside of all this is what keeps CFOs, including me, up at night, the worry and anxiety of customer churn.  When customers don’t renew, it is painful, and it can really hurt your business.

What can be done, and why do companies still get surprised when customers churn?  Is there a systematic approach to predicting costly customer churn?  Similar to how companies try to predict future events or analyze patterns in a given dataset, can this methodology be applied within a finance organization?  The initial output would be which customers will leave. Then, from a more granular perspective, there would be some classification of customers who may churn, who will stay, and the propensity of that customer to leave.

Prepping the Data to find the Propensity to Churn

The labeled dataset to ready in this study is the list of customers while also preparing the additional data such as communications between your sales, marketing, and customer service teams with each specific customer.  What is the response to new features, incentives, and enhanced product offerings?  How quick is the response to a query, and what are some of the words used in the response (natural language processing can help when the volume of data is large).  Once the training data is prepared, a model can be created to understand the relationship between the response and the exploratory variables. For example, the model can be trained and know that folks are likely to churn when customers don’t respond for a specific time or after a particular amount of time.

Customers churn primarily because of a lack of value in the products or services they have subscribed to.  For example, someone got very excited to sign up but was met with disappointment or a better, more cost-effective option.  There was something better out there that caught a company by surprise.  The other item to model is the time frame.  For customers on large enterprise contracts, renewals may come once a year, but churn can come suddenly for smaller monthly subscription packages. Hence, the duration in the model must be mapped accordingly, as the closer the time, the more accurate the model may be.

Lots of Data to Consider

The dataset can be widened to see if external factors contribute to customer churn.  For example, the size of a company’s revenue can be an indicator, and scrapping data from LinkedIn to see if a company has slowed hiring drastically can be used as input.  In addition, data on how often a company has updated or refreshed its website can be fed into the machine learning platform and combined with other data to gauge if there is a propensity for a customer to churn.

Picking a Model

There are various predictive models you can use. Your resident data scientist will offer generalized linear models, logistic regression, random forests, decision trees, and neural networks.  So which model do you choose, and how can one get started using historical data to predict customer churn?  There are some good recommendations you can check out.

Since it’s about pattern recognition, I started my journey with neural networks.  Like our brain’s almost infinite connections, neural networks consist of node layers containing an input layer, one or more hidden layers, and an output layer.  Each node is connected to another node and has an associated weight and threshold.  So if the output of any individual node is above a threshold value that one specifies, that node is activated, sending data to the next layer of the network and so on till the output is achieved.

If you don’t have the time to collect all the data and prep it for the model there are options.  If you want to dive right in, there is a Kaggle project on Telco Customer Churn with the dataset ready to go.

Now the Easy Part…So I thought

Now that the dataset is ready and a model has been prepared, do we just feed it into a platform and be done with it?  Not exactly.  For most people, the project will fail since there are so many steps to get right in order to have the model up and running.  From what my engineers tell me, I need to be a container expert (often Kubernetes); I have to refactor the data, develop the model registry, and do a host of other things to get going. There are many new tools to learn I am told.

As a CFO, learning to use a Jupyter Notebook wasn’t that hard.  A Jupyter notebook is used to create and share documents that have live code, visualization, and text and is very familiar to data scientists.  The Jupyter Notebook is an open-source web application that you can use to create and share documents that contain live code, equations, visualizations, and text.  The people maintain the Jupyter Notebook at Project Jupyter.  Here is a good beginner’s guide: https://jupyter-notebook-beginner-guide.readthedocs.  /en/late   /what_is  jupyter.html.  Now what?  Feed it to the software and get some churn data, right?

The Difficulties Compound with Existing Products

Having a data set that is cleaned and validated and a notebook is the very beginning of bringing a model to production.  From what my MLOPs people inform me, there is feature engineering involved, model training, pipeline creation, model deployment, monitoring, governance, and security which all need to be processed sequentially.  How long before I can get my churn data?  When someone responds, “it depends,” it usually means it will take too long.  Throw in hyperparameter tuning, algorithm choices, and different model architectures, and the whole project becomes even more involved.

There Must Be an Easier Way

Things get more complicated still and time-consuming when I have to work on the repository, pipelines, and reproduction in case  I have an error in the model.  I’m a CFO, and I have to work on the company’s KPI, grow our customer base, speak with investors, board prep, and work on our budget; I’m not sure I have the time to figure all these steps out, test it and then spin up instances in the cloud to run my churn model.

Enter Kubeflow

A colleague introduced me to Kubeflow, the machine learning platform built on top of Kubernetes.  Integrating many steps of pipeline creation, model training, and complicated workflows is an end-to-end platform for data scientists to get models up and running.  And the good news is that you can stay in a comfortable Jupyter environment.  Like most open-source projects, there is a lot of configuration and manual tooling to get Kubeflow working right. For example, getting pipelines to work, and debugging, and that’s if you can get started properly in the first place.

Quick Start with Arrikto

Arrikto announced its new Kubeflow as a Service, a fully managed version of Kubeflow, so there is no need to worry about infrastructure and all the other things necessary to do Kubeflow work.  I gave it a try, and it was easy to get started and input the data. Data scientists can now get instant access to an end-to-end MLOPS platform.  You don’t even need to know Kubeflow or Kubernetes since all the heavy lifting is done in the background.  I can focus on the data inputs and model iteration and refine the predictability of my customer churn model.  You can give it a try here: www.kubeflow.arrikto.com

Use ML to Reduce Churn

Customer churn is painful and reacting with discounts and concessions when a customer doesn’t want to renew doesn’t work.  By modelling and using the right platform, one can get a sense of which customers have a higher propensity to leave before the action is taken so that remedial work can start in earnest.  There are not enough hours in the day for the modern CFO, so the last thing we need to deal with is to stitch together disparate systems to get the data we need to do our jobs.  Kubeflow when done right can help get your models up and running.

Budgeting Like A Hero with Zero Based Budgeting

It’s budget time everyone and today we look at Zero-Based Budgeting (ZBB).  ZBB is a method of budgeting that requires all expenses to be justified for each new period, starting from a “zero base.” Unlike traditional budgeting (let’s just increase it from last year a little bit),  ZBB builds budgets from scratch, ensuring every expense is necessary and aligned with the organization’s goals. This approach can lead to more efficient resource allocation and strategic alignment.

What Works in Zero-Based Budgeting

1. Cost Efficiency: By forcing a justification from everyone for every expense, ZBB helps identify and eliminate unnecessary costs, leading to significant savings.
2. Strategic Alignment: ZBB ensures that spending is directly tied to current business goals, rather than historical patterns.
3. Accountability and Transparency: Managers must justify their budget requests, promoting a culture of accountability and transparency.
4. Agility: Organizations can quickly adapt to changes in the business environment by reallocating resources as needed.

Why Sometimes Zero-Based Budgeting Doesn’t Work

1. Resource Intensive: Justifying every expense can be time-consuming and labor-intensive, especially for large organizations.
2. Complexity: The detailed analysis and documentation required can be complex and demanding.
3. Resistance to Change: Employees and managers may resist the transition due to increased scrutiny and workload.
4. Short-Term Focus: ZBB may sometimes encourage short-term cost-cutting at the expense of long-term investments.

Best Practices for Zero-Based Budgeting

1. Strong Leadership Commitment: Successful implementation requires unwavering support from top management.
2. Comprehensive Training: Ensure all employees understand the principles and benefits of ZBB through thorough training programs.
3. Clear Communication: Maintain transparent communication to address concerns and foster buy-in.
4. Accurate Data Collection: Collect and analyze accurate data to create reliable decision packages.
5. Ongoing Review and Adjustment: Treat ZBB as an ongoing process, with regular reviews and adjustments to ensure continuous improvement.

Companies in Technology Using Zero-Based Budgeting

Several technology companies have adopted ZBB to improve cost management and resource allocation, including:

Google (Alphabet): Google implemented ZBB to manage its vast and diverse range of projects and operations more effectively. By justifying every expense, Google was able to streamline its budget, focusing on key areas like artificial intelligence and cloud computing, while reducing expenditures on less critical projects. This helped Google maintain its innovative edge and financial efficiency.

Dell Technologies Dell faced significant competitive pressure and needed to enhance operational efficiency. Through ZBB, Dell was able to identify redundant processes and unnecessary administrative costs. By reallocating resources to high-priority areas like product development and customer service, Dell improved its overall cost structure and operational performance.

Coca-Cola Coca-Cola adopted ZBB to revamp its financial management practices and improve profitability. The company used ZBB to scrutinize all marketing and operational expenses. This approach led to a more disciplined spending culture and allowed Coca-Cola to redirect funds to high-growth areas, enhancing its market position.

Unilever Unilever used ZBB to tackle inefficiencies across its global operations. By justifying every cost from scratch, Unilever was able to identify significant savings in its supply chain and marketing budgets. The disciplined approach helped Unilever to reinvest in innovation and sustainability initiatives, driving long-term growth.

Comparison with Other Budgeting Methods

• Traditional Budgeting: This relies on historical data and incremental adjustments, which can perpetuate inefficiencies. ZBB starts from zero and requires justification for every expense.
• Activity-Based Budgeting (ABB): Focuses on activities and their costs, similar to ZBB, but ZBB is more rigorous in justifying every line item.
• Rolling Budgeting: Continuously updates the budget to reflect changing conditions, providing flexibility. ZBB can complement rolling budgeting by ensuring all expenses are justified.

Suitability of Zero-Based Budgeting for Different Companies

• Good Fit:
• Not a Good Fit:

What are the best tools to use for zero-based budgeting?

Spreadsheets: Traditional spreadsheet applications like Microsoft Excel or Google Sheets can be used to create and manage zero-based budgets. They provide flexibility in organizing budget data, performing calculations and generating reports. Spreadsheets allow for customization and can be a cost-effective option for smaller organizations.

Financial planning and analysis (FP&A) software: They offer dedicated features for budgeting, forecasting and financial analysis. These tools provide a centralized platform for top-down and bottom-up budgeting creation, collaboration, scenario modelling, data integration and reporting. They often come with advanced analytics capabilities, enabling organizations to make data-driven budgeting decisions.

Enterprise resource planning (ERP) Systems: ERP systems integrate various financial reporting processes, including budgeting. These systems offer modules specifically designed for budget creation, tracking and reporting. They provide a comprehensive view of financial data, facilitate data integration and support collaboration among different departments.

Budgeting and planning software: Dedicated budgeting and planning software are designed to streamline the budgeting process. These tools provide features like budget templates, workflow automation, data consolidation, scenario modeling and reporting. They often offer user-friendly interfaces and enable collaboration among budget stakeholders.

Data visualization tools: Data visualization tools enable organizations to visualize budget data and financial insights. These tools create interactive charts, graphs and dashboards that enhance the understanding and communication of budget information. Data visualization tools can help identify trends, patterns and anomalies in the budgeting process.

Project management software: Project management tools like Asana, Trello or Jira can be utilized to track budgeting tasks, deadlines and milestones. These tools help manage the workflow, assign responsibilities and ensure accountability during the budgeting process. They enhance collaboration and provide transparency into the progress of budget-related activities.

Why Private Equity Companies Favor Zero-Based Budgeting

Private equity (PE) firms often favor ZBB for several compelling reasons:

1. Rapid Cost Reduction: PE firms typically have a relatively short investment horizon and aim to quickly improve the financial performance of their portfolio companies. ZBB helps identify and eliminate unnecessary costs swiftly, enhancing profitability and making the company more attractive for maximizing sales or public offerings.
2. Enhanced Financial Control: ZBB promotes rigorous financial discipline and transparency, which are crucial for PE firms. By justifying every expense, portfolio companies can ensure that funds are used effectively, minimizing waste and optimizing resource allocation.
3. Alignment with Strategic Goals: PE firms often implement strategic changes to unlock value in their portfolio companies. ZBB ensures that all expenditures are aligned with these new strategic goals, facilitating the execution of the PE firm’s value creation plan.
4. Improved Cash Flow: Efficient resource allocation through ZBB can significantly improve a company’s cash flow. For PE firms, strong cash flow is essential as it enables debt repayment, funds growth initiatives, and supports dividend distributions.
5. Focus on Value Creation: ZBB forces managers to think critically about their spending and its impact on the company’s value. This aligns well with the PE model, which emphasizes creating value through operational improvements and strategic investments.
6. Cultural Shift: Implementing ZBB can foster a culture of cost consciousness and efficiency within the portfolio company. This cultural shift can lead to sustainable operational improvements, even after the PE firm exits the investment.

The Flip Side of ZBB

In the late 1970s, then-U.S. President Jimmy Carter attempted to implement ZBB in federal spending. The idea was to scrutinize and justify every dollar spent, rather than simply building on the previous year’s budget. However, the complexity and scale of federal budgeting proved too unwieldy for this approach, and the initiative was ultimately abandoned.

Despite this setback in the public sector, ZBB experienced a renaissance in the corporate world about a decade ago. The catalyst for this revival was 3G Capital, a Brazilian investment firm known for its aggressive cost-cutting strategies. 3G Capital applied ZBB principles to great effect at Anheuser-Busch InBev, the global brewing giant it helped create through a series of mergers. The firm later replicated this success at Kraft Heinz after orchestrating the merger of Kraft Foods and H.J. Heinz but things did not work out that well this time around.

While both Anheuser-Busch InBev and Kraft Heinz achieved industry-leading profit margins, things went pretty bad at Heinz. Cost-cutting was followed by more cost-cutting until top-line growth was impacted and other problems ensued. Kraft Heinz faced a triple blow: a $15.4 billion write-down of its Kraft and Oscar Mayer brands, a significant dividend cut, and an SEC accounting probe. The news sent its stock plunging 27% in one day.

As noted in a 2017 report by consultants BCG, while the cost reductions can be impressive, they don’t guarantee faster growth. The report cautioned that the clumsy application of ZBB can have a demoralizing effect on the organization, potentially distracting from growth and value creation.

Getting Started with Zero-Based Budgeting for a SaaS Company

Implementing ZBB in a Software as a Service (SaaS) company involves several key steps:

1. Define Objectives and Scope: Clearly define the goals of ZBB implementation, such as cost reduction, efficiency improvement, or resource reallocation. Determine the scope, whether it’s for the entire organization or specific departments.
2. Engage Leadership: Ensure strong commitment from top management to drive the ZBB process. Leadership buy-in is crucial for overcoming resistance and ensuring successful implementation.
3. Form a ZBB Team: Create a dedicated team to oversee the ZBB process. This team should include representatives from finance, operations, and other key departments.
4. Conduct Training: Provide comprehensive training to all employees involved in the budgeting process. Ensure they understand the principles of ZBB and how to develop and justify budget requests.
5. Identify and Categorize Costs: Break down all expenses into categories, such as personnel costs, software development, marketing, and customer support. Create detailed decision packages for each cost category.
6. Justify Each Expense: Require each department to justify their budget requests from zero. This involves explaining the necessity of each expense and how it contributes to the company’s strategic goals.
7. Review and Approve Budgets: Have the ZBB team review and approve the budget requests. Ensure that all expenses are justified and aligned with the company’s objectives.
8. Monitor and Adjust: Treat ZBB as an ongoing process, it’s not one-and-done. Regularly review actual spending against the budget and make adjustments as needed. Use insights gained from the ZBB process to continuously improve.

Zero-based budgeting offers numerous advantages, including cost efficiency, strategic alignment, and enhanced accountability. However, it also presents challenges such as resource intensity and complexity and it can go badly as it relates to the impact on future growth.

Following best practices and understanding their suitability for different types of organizations, you can leverage ZBB to drive financial discipline and achieve strategic objectives. It’s effective when used properly and for bloated companies, it’s great for delivering rapid cost reductions and improved financial performance.

For more details on budgeting, please refer to our handy dandy reference: The Guide to Finance & Accounting.