Working with Stakeholders

Your work isn't in data, it's in people.
Four types of stakeholders: business, engineering, leadership, and your manager.

Business Stakeholders

• usually have little technical background
• each individual's background can vary widely
• These individuals rely on data scientists for critical information but lack the background to make sure it's right
• These folks need to be constantly involved in any projects
• You need to build trust with these stakeholders in your recommendations and your overall work. Do this by building their understanding of what you're doing.
• These individuals can open avenues in the business for more data science projects. These people are the reason you have a job.

Engineering Stakeholders

• brothers from another mother, sisters from another mister, siblings from another parent
  • You can talk shop with them - they get it.
• These folks might not understand that what you build may not work - regular development doesn't have that unknown factor associated with it. Communicate early and often that you're the equivalent of "Research and Development"

Corporate Leadership

• very scary people - you really want to impress these people. They can green light big resources and projects.
• super busy and don't care about the details. Communicate very high level stuff in simple and straightforward ways.
• Make your final deliveries as easy as possible for them to access
  • these people think logging in is too time consuming (from personal experience)

Your Manager

• it's a two way street: they want you to succeed so they look good, you want them to help you in your career
• talk to them all the time! tell them when you're struggling and ask for help earlier rather than later.

Working with Stakeholders

Understand Their Goals

People don't make decisions in a vacuum. Ask stakeholders directly what they care about, ask others on your team about your stakeholders, and infer goals from a stakeholder's behavior. How someone might react to different outcomes informs your communication decisions. You may have difficult conversations with someone, and remember: you can ask your manager to help you navigate tough conversations.
If someone has key performance indicators (KPIs) or objective key results (OKRs), you can deliver your results based on those metrics.

Over-Communicate

We usually don't communicate enough. If you feel like you're over-communicating, then you might be communicating enough.

• Tell stakeholders about your progress and any changes in scope or timelines. Some stakeholders have the ability to remove barriers in your work.
• Tell stakeholders why your work matters. Tell them how the work informs the business and what comes next.
  Bake these updates into your work with regular check-in meetings or a scheduled email update. Schedule ad-hoc meetings as necessary.

Be Consistent

Deliver a consistent product
Analyses

• structure it the same - abstract, objective, data, conclusion, next steps
• deliver the analysis in the same file type - PowerPoint vs pdf vs html stored in the same place
• style each analysis the same - colors, font, text size, etc
  APIs
• consistent input format - same parameter names, only JSON objects, etc
• consistent outputs - structure should match the input
• consistent authentication - don't lose track of which credentials belong to each API
  Make yourself templates and don't waste time trying to orient yourself to a plethora of different styles. More standardization means easier sharing.

Create a Relationship

Be professional. Be dependable. Expectations $\equiv$ Reality.

Prioritizing Work

Tasks come in 3 buckets:

1. Quick tasks that come directly from stakeholders
2. Long-term projects for the business
3. Ideas that you think have a long-term benefit
   For each task, ask:

• Will this work have an impact?
• Will this work do something new?
  Tasks can then be categorized into 4 buckets.

Impactful and Innovative

Not many projects fall into this category:

• There needs to be enough data for data science methods to be useful.
• There has to be an interesting signal in the data that the models can pick up.
• The part of the business has to be large or important enough that changes can make a difference to the bottom line (so the project probably isn’t optimizing dry-erase marker inventory for the office).
• The problem must be complex or unique enough that people haven’t tried it before.

Impactful, not Innovative

Mundane work. This is the day-to-day job. This pays the bills.

Not Impactful but Innovative

These are ivory towers. It's usually a waste of resources: huge amount of time and resources spent with little to show for it. If you can’t see a use for the project immediately, stakeholders probably won’t either. Don't lose your business value doing these.

Not Impactful or Innovative

Lots of work falls here too. Do your best to advocate for your time being well used.

Sources: 1