Topics: Coronavirus | AI | Data Science | Deep Learning | Machine Learning | Python | R | Statistics

KDnuggets Home » News » 2020 » Nov » Opinions » How Data Scientists Can Avoid ‘Lost in Translation’ Syndrome When Communicating With Management ( 20:n45 )

How Data Scientists Can Avoid ‘Lost in Translation’ Syndrome When Communicating With Management


When it comes to data science projects, the disconnect between business executives and data teams can lead to major tension. Keeping these challenges from arising in the first place through effective communication will help reduce friction with stakeholders.



By Pini Raviv, Software engineer and front-end team leader at Israeli startup.

Statistics

Data science is a wonderful field to work in, but like every other highly specialized field, you'll have to deal with frustrations within the workplace.

In my experience, the primary source of workplace issues is the disconnect that occurs between business executives and data teams. Only data that can be understood is helpful and valuable. Data science professionals are sometimes guilty of forgetting this fact.

Communication is important, but how should you communicate? Here are five tips you can use to convey the points you want to make to business stakeholders, which will also reduce the friction you face with them.

Create relevant data visualizations

Unless your boss specifically asks for them, avoid number-heavy reports. People are visual creatures. It's far easier for us to understand conclusions through pictures than tables. Data visualization tools can bring your analysis to life, but the challenge doesn't end there. You still need to ensure your data is easily understood.

Data scientists don't have the time to master graphic design, but there are a few hacks you can use. Online tools such as Coolors and Paletton help you create color schemes that are both appealing and account for color blindness in your audience. A simple DIY hack is to pixelate your favorite picture in an online photo editor and extract those colors.

Minimalism is the key to communicating your conclusion through graphs and charts. Eliminate call-outs in your chart that don't add value to your broad conclusions, and consider removing X and Y axis labels if a call-out can deliver information better. Choose fonts wisely, and don't use more than two throughout your presentation. Google complimentary fonts and stick to that formula.

It can be tempting to add bells and whistles to your presentation (animation, interesting sidebars, etc.), but avoid them unless they're relevant to the core of what your stakeholders want to measure. The average business user is intimidated by data, and your job is to simplify it for them. The simpler your conclusions are to understand, the less you'll need to defend your work to management, and you'll find them more willing to trust their data.

Always provide context

There is a tendency in analytics-heavy organizations to worship data and to forget that data isn't the truth. In fact, data isn't all that relevant at all until it has context wrapped around it. Contextualizing data is part of a data science professional's job. The greater the degree of trust your management places on the numbers, the more should you focus on the biases, defects, and integrity of the data.

Start by evaluating whether you've collected data from all the relevant sources. If you neglect important sources of data, you'll be viewing just a small piece of the puzzle. Always consider the possibility that your data might live in sources you haven't touched as yet.

Next, segment the data to break it down into bite-sized pieces. Data segmentation will help you categorize and dive deeper into your data. If your audience is the average business user, always tie your segments back to business goals and not to interesting technical ones.

Remember your audience

Data scientists are often guilty of forgetting their audience and getting lost in the technical details of their data. You might have had to develop creative coding solutions to arrive at your conclusions, but if your audience isn't technical, they're not likely to care.

For example, let’s say your manager asks you for a report that identifies the highest selling product for every date of the previous month. It's easy to group the best selling products by date but what you need to do is display only the top performers by date. Postgres and Redshift have window functions that simplify this.

However, what if your organization uses MySQL? You'll need to use group_concat to roll your data into CSV strings grouped by date and then use substring_index to extract the best performer. Well played! However, your manager doesn't care for your technical wizardry. All she wants is the results.

Focusing on your audience allows you to manage their expectations. A common complaint amongst data scientists is that management tends to impose requirements that aren't realistic. What seems trivial to a business user often requires complex technical solutions. Instead of diving into the technical details of your task, inform them of the consequences in business terms.

For example, you could let them know that their request will take a week to complete, instead of a day. You'll be speaking their language by communicating like this, instead of coming across as a technical showboat. Data is a black box for the average business user. Your job is to translate it for them, not teach them what needs to be done.

Set expectations

Management often adds variable requests at the last minute, and data modelers grin and bear them. These last-minute requests typically need additional days of data gathering and cleaning and push deadlines even further out.

Another common occurrence is dealing with unreasonable requests. Your company might have just one month's data but might require a projection of a year's sales. Management might have heard about the powers of ML and statistical techniques to fill holes in data and might expect you to plug these in to achieve results.

You must set expectations before every task to avoid problems down the road. Incorporating variable submission deadlines and producing data quality reports in business-friendly language are often effective ways of setting expectations.

Stick to the process

Good data analysis requires you to spend time getting to know your data sets and understanding their sources. In fast-paced environments, you might want to rush ahead and get to the analysis portion and produce your reports.

Remember that your function as a data analyst is to serve business goals.

Producing faulty reports is only going to weaken the organization's trust in you. Many business managers are accustomed to relying on their gut and distrust data. Skipping parts of the process to produce quick reports isn't going to get them to trust you more.

Always communicate

Communication is the key to producing value for your organization. Data scientists can get bogged down in the technical details and communicate in ways that aren't business-friendly. These tips will help you avoid falling into that trap, and you'll manage to provide executives with true insight into their businesses.

 

Bio: Pini Raviv is a software engineer and data scientist with experience as a front-end team leader for an Israeli startup company.

Related:


Sign Up

By subscribing you accept KDnuggets Privacy Policy