Data Science Communication

• Effective communication for data science

Why is communication difficult?

• Stakeholder expertise can vary widely
• Conflicting priorities or goals among stakeholders can cause a misalignment of interests
• Sales teams may want to highlight successes, while operations teams focus on areas needing improvement
• Time constraints may limit stakeholders' ability to absorb detailed data analysis
• Audience members may struggle to interpret complex datasets or visualizations due to low data literacy

Identifying Stakeholders

• Primary stakeholders are directly affected by your analysis, e.g., customers, internal teams, decision-makers
• Secondary stakeholders have a less direct, but significant, interest in the project, e.g., regulators and investors
• Non-obvious stakeholders may be less visible, but are still impacted - e.g., the customer service team when analyzing customer data
• Evolving stakeholders require reassessment regularly, as the stakeholder landscape can shift during a project

Managing Stakeholders

• Build relations through proactive communication and trust
• Understand stakeholder needs
• Tailor your communication to suit each stakeholder's message and level of technical detail
• Involve stakeholders to ensure alignment and address any concerns
• Manage expectations around timelines, data limitations, and the scope of insights

Why involve stakeholders early?

• Prevents misalignment and misunderstandings
• Builds trust and fosters buy-in by valuing stakeholder input
• Enables iterative improvement through feedback, which improves the final output

Adapting to Stakeholders

• Tailor language, vocabulary to suit the audience
• Provide engineers detailed length, executives want conciseness
• Take into consideration the audiences risk and certainty
• Consider what the audience wants from the data, how much explanation is needed, and whether to include opinions or just facts

Forms of Communication

• Data science spans academia and industry, making communication important
• Different forms of communication include:
• Academic research papers
• Academic posters
• Executive reports/summaries
• Dashboards

Academic Papers

• The most common paper structure in science is IMRaD (Introduction, Methods, Results, and Discussion)
• Introduction section sets up the problem, explains its importance, reviews literature, details the solution, and outlines the paper structure
• Methods section explains the solution and evaluation quality
• Results shows evaluation results
• Discussion includes discussing the results, potential research/real-life impact, limitations, and future work
• Predictable structure is important for efficiently reading hundreds of papers
• Related works are placed after the introduction, but at the end of the paper in some fields
• Include a conclusion and an abstract
• Final structure resembles AILMRaDC (Abstract, Introduction, Literature, Methods, Results and Discussion, Conclusion).
• IMRaD is successful because it is both boring and predictable
• Level 1 headings contain big sections (IMRaD)
• Level 2 headings contain themes (e.g., Methods > Evaluation)
• Level 3 headings contain topics (e.g., Methods > Evaluation > Datasets)
• Paragraphs contain ideas

Academic Posters

• Academic posters involve graphical text and inform readers/advance knowledge
• Designed to allow for replication/reproduction and can be displayed while the author answers questions
• Posters meant to be read autonomously contain text
• Posters meant to be explained by a speaker use plots

Executive Reports/Summaries

• Executive reports/summaries are typically one-page summaries linked to longer technical reports, for executives/decision makers
• Intended to inform readers and assist in decision making; key elements include:
• Problem statement
• Methodology
• Key findings
• Recommendations
• Data visualizations

Dashboards

• Dashboards are interactive graphical applications that allow users to explore data, and are fed from live data
• Intended for both technical and less technical people in order to inform and aid decision-making
• Designed around key performance indicators (KPIs) and metrics
• Follow the intuitive layout, information hierarchy, appropriate charts, should be interactive, clean, and uncluttered
• Common features include KPIs, trends over time, comparisons, filtering & drill-down, and alerts

Data Storytelling

• Data storytelling connects data, transforms data into a narrative, make findings relatable/memorable, and inspires action
• Start with a compelling question, build tension, resolve insights, use visuals to guide, and include a call to action

Presenting Your Analysis

• Begin with a hook (relevant question, fact, or anecdote) in order to get the audiences attention.
• Highlight 2-3 key insights, and avoid overwhelming the audience with detail
• Make recommendations actionable, and end with a clear takeaway
• Effective presentations should explain why findings matter, use impactful visuals, avoid data dumps, practice for confidence, and anticipate questions

Best Practices in Data Science Communication

• BLUF (Bottom-Line Up Front): State key findings/recommendations at the beginning, instead of "I analyzed customer data and found several patterns..." lead with "Our marketing campaign is underperforming with demographics X and Y..."
• Be clear and concise by cutting jargon, and focus on impact/decisions
• Use active voice and avoid subjective words
• "Show, don't tell" using appropriate charts and clear labeling
• An abstract should be short, and advertise your piece with context and contribution

Ethical Aspects of Writing/Communication

• Avoid plagiarism by copying text, asking a friend, or using ChatGPT
• Properly cite your sources
• Avoid manipulating data, cherry picking, or random seed optimization.
• Write for your audience
• Scientific writing is rarely aimed at a general audience
• Find out who your audience is
• Discover their conventions
• Use their vocabulary
• Audience of a paper = academics who attend your conference
• Audience of a dissertation = your supervisor, and a second marker who might not be expert of the field
• Audience of an executive summary = your boss, who may or may not decide to get rid of you
• Audience of your report = the module delivery team
• Proofread, make sure each sentence has a purpose, and is not redundant
• Figures have a number and caption, use right figure (box plots preferred), refer to figures!
• Format to avoid chart junk