Best Practices for Applying Data Science Techniques in Consulting Engagements (Part 2): Scoping and Expectations

By Jonathan Balaban • December 12, 2017

This is part 2 of a 3-part series written by Metis Sr. Data Scientist Jonathan Balaban. In it, he distills best practices learned over a decade of consulting with dozens of organizations in the private, public, and philanthropic sectors. You can find part 1 here.

Credit: Lánluas Consulting


In my first post of this series, I shared four key data strategies that have positioned my engagements for success. Concurrent with collecting data and understanding project specifics is the process of educating our clients on what data science is, and what it can and cannot do. In addition — with some preliminary analysis — we can confidently speak to level of effort, timing, and expected results.

As with so much of data science, separating fact from fiction must be done early and often. Contrary to certain marketing messages, our work is not a magic elixir that can simply be poured on current operations. At the same time, there may be domains where clients erroneously assume data science cannot be applied.

Below are four key strategies I’ve seen that unify stakeholders across the effort, whether my team is working with a Fortune 50 firm or a small business of 50 staff.

1. Share Previous Work

You may have already provided your client with white papers, qualifications, or shared results of previous engagements during the ‘business development’ phase. Yet, once the sale is complete, this information is still valuable to review in more detail. Now is the time to highlight how previous clients and key individuals contributed to achieve collective success.

Unless you’re speaking to a technical audience, the details I’m referring to are not which kernel or solver you chose, how you optimized key arguments, or your runtime logs. Instead, focus on how long changes took to implement, how much revenue or profit was generated, what the tradeoffs were, what was automated, etc.

2. Visualize the Process

Because each client is unique, I need to take a look through the data and have key discussions about business rules and market conditions before I share an estimated process map and timeline. This is where Gantt charts (shown below) shine. My clients can visualize pathways and dependencies along a timeline, giving them a deep understanding of how level-of-effort for key people changes during the engagemenCaCption

Credit: OnePager
3. Track Key Metrics

It’s never too early to define and start tracking key metrics. As data scientists, we do this for model evaluation. Yet, my larger engagements require multiple models — sometimes working independently on diverse datasets or departments — so my client and I must agree on both a top-level KPI and a way to roll up changes for regular tracking.

Often, implementations can take months or years to truly impact a business. Then our discussion goes to proxy metrics: how can we track a dynamic, quickly updating number that correlates highly with top-level but slowly updating metrics? There’s no “one size fits all” here; the client may have a tried and true proxy for their industry, or you may need to statistically analyze options for historical correlation.

For my current client, we settled on a key revenue number, and two proxies tied to marketing and project support.

Finally, there should be a causal link between your work/recommendations and the definition of success. Otherwise, you’re binding your reputation to market forces outside of your control. This is tricky, yet should be carefully agreed upon (by all stakeholders) and quantified as a set of standards over a period of time. These standards must be tied to the specific department or scale where changes can be enforced. Otherwise, the same engagement — with the same results — can be viewed unpredictably.

4. Phase Out Efforts

It can be tempting to sign up for a lengthy, well-funded engagement off the bat. After all, zero-utilization business development isn’t actual consulting. Yet, biting off more than we can chew often backfires. I’ve found it better to table detailed discussions of long-term efforts with a new client, and instead, go for a quick-win engagement.

This first phase will help my team and the client team properly understand if there’s a good cultural and technological fit. This is important! We can also gauge the willingness to fully adhere to a ‘data science’ approach, as well as the growth prospect of a business. Engaging with a non-viable business model or locking down a sub-optimal long-term path may pay out immediately, but atrophies both parties’ enduring success.

5. Share the Internal Process 

One easy trick to work more efficiently and share progress is to build a scaffold around your internal tasks. Again, this changes by client, and the platforms and tools we use are dictated by the scale of work, technology needs, and investments our clients have made. Yet, taking the time to build a framework is the consulting equivalent of building a progress bar in our application. The scaffold:

  • - Structures the work
  • - Consolidates code
  • - Sets clients and stakeholders at ease
  • - Prevents smaller tasks from getting lost in the weeds

Below is an example template I use when I have the freedom (or requirement) to work in Python. Jupyter Notebooks are fantastic for combining code, outputs, markdown, media, and links into a standalone document.

My project template

The template is too long to view inline, but here’s the section breakdown:

  1. Executive Summary
  2. Exploratory Data Analysis
  3. Scaling Data and Model Prep
  4. Modeling
  5. Visualizations
  6. Conclusion and Recommendations: 
    • - Coefficient importance: statistically significant, plus or minus, size, etc.
    • - Examples/Story
    • - KPI Visualizations
    • - Next Steps
    • - Risks/Assumptions

This template almost always changes, but it’s there to give my team a ‘quick start’. And yes, coder’s block (writer’s block for programmers) is a real malady; using templates to break down tasks into manageable bits is one of strongest cures I’ve found.


So, there we have five key strategies for scoping engagements and setting expectations. I hope you find these tips transformative in how you approach clients!

What techniques have you found useful in your engagements? Let me know on Twitter at @ultimetis.

Similar Posts

data science
Learn Machine Learning in 6 Months

By Zachariah Miller • May 24, 2021

I came across a question on Quora that boiled down to: "How can I learn machine learning in six months?" I started to write up a short answer, but it quickly snowballed into a huge discussion of the pedagogical approach I used and how I made the transition from physics nerd to physics-nerd-with-machine-learning-in-his-toolbelt to data scientist. Here's a roadmap highlighting major points along the way.

data science
Misleading Graphs: Manipulating the Y-Axis

By Roberto Reif • April 06, 2020

One of the most commonly used charts for data visualization is the bar chart. But too often, the starting value of the y-axis is intentionally modified to skew our interpretation of the chart and the data. In this post, see examples and learn how to readily identify this issue.

data science
Python Guide: Tutorial For Beginners

By Adam Wearne • July 28, 2021

Welcome to a brief introduction to Python. In this article, we'll provide an overview of the Python language, some of its many use cases, how to install Python on your computer, and how to use Python.