Science is the process of curiosity. It doesn’t care about profits, margins, timelines, or shareholders. In science, failure (learning) is an acceptable outcome. In business, however, companies invest in data science with an expectation of ROI.
In order to obtain the corporate “buy-in” for more resources and support, new data science programs scramble to find results worth shouting from the rooftops or at least highlighting at the next board meeting.
We crave the hole-in-one even though it’s our first time golfing. While you can’t be guaranteed immediate success, there are three ways to pick an early project that supports the value of your initiatives and helps you gain momentum with important stakeholders.
Access trusted data sources
You may have data coming in from an overwhelming amount of siloed data sources. Choosing the right data to kick off your project is an important step. Start by identifying the sources both IT and the business already have access to and trust.
Here are three areas for additional documentation that are well worth the time:
- Project Description
- Recipe Short Description
- Prepare Recipe Comment
The data used for this example is from a Kaggle project using financial transaction data to detect fraudulent transactions and money laundering.
Project description
When creating a project, there should be a clear project description on the project page. Not only does it make it clear why the project exists, but also opens the opportunity for it to be reused by other projects. For example, if the project identifies target markets, another project could use this data for further analysis instead of reinventing the same data cleansing.
The project description should include:
- The project purpose or goal.
- A project sponsor or department owner.
- A short description of the data sources.
- A description of any data that was filtered before entering this project.
To view and edit your project description go to your project homepage by clicking on the project title near the top left corner of the screen. Add a project short description under the title for quick reference and searching.
If it will take IT a long time to provide access, you will lose momentum early in the project. If the data sources have missing or user-entered data, you may spend too much time cleaning it and less time on complex analysis. Make sure the business has established clear definitions for the data (ex. defining a product, company, or target customer) to avoid controversial topics.
Questions to ask to ensure you have access to trusted data:
- Which sources have database diagrams already created?
- Are there a few business subject matter experts that truly understand the data?
- Can you easily access the data for analysis?
- Can you easily tie the data to other trusted reports?
- Are there standard definitions for the data being used?
Find a project with measurable value
The data request list may be overflowing, but use this flagship project to solve a problem with measurable impact. Will a larger value for a small number of users be more impactful than a little bit of value to many? It depends on how measurable the results are. Saving 100 hours or $10,000 is not only easy to explain but will grab attention!
If your organization is revenue-focused, prioritize projects with measurable revenue increases, even if just for one department. While a flashy new dashboard might be nice for the salespeople, will it actually increase their sales? If so, how will you measure the sales increase in a way that can be communicated clearly through the data?
Questions to ask to ensure the project has measurable value:
- Will this insight increase revenue? How?
- Will this solution create time savings? How much?
- If this information will unearth great conversations, how will you measure the value of those additional insights?
- How often will it be used once complete?
- How many users will be able to use this information?
Use your results to support positive action
Using the data to allow leadership to engage with the business in a way that guides, empowers, and positively propels your business forward will be the ultimate “win”. For example, leadership may be excited about increased transparency into performance metrics previously muddled by the time it reaches them. They may request the ability to drill into which regions, salespeople, and products are excelling and use this information to learn and inform, benefitting the entire organization.
Communicate clearly with leadership in the early stages of your project. Set clear expectations and agreement on the intent of the project and the goals it is meant to achieve. Without this consensus, a project could be used to impact negative consequences such as layoffs or budget cuts or cause people to discredit the data, process, tools, and solution. Aimed toward a positive purpose and fueled with clearer information that supports relevant business outcomes, your organization can accelerate its progress and move the needle confidently.
Questions to ask to ensure the project will result in positivity for your organization:
- What action will be taken with this information?
- Will the action based on these results be celebrated openly?
- What incentive do end users have to provide this information?
- Who will be excited about the results?
- Who will be apprehensive of the results?
Introducing data science into any organization takes time. Identifying an early project that can be proudly showcased to leadership and throughout the organization will expedite this process. You will quickly go from trying to convince others of its value to needing to prioritize a large list of data project requests! If you select a project where you have access to trusted data, measurable value, and results that create positive action, you will be on track to building a successful data science practice.
Recipe short description
Imagine opening a new project and being able to understand the way information is flowing and changing through the project without clicking on anything. Visual recipes make it easy to see the functionality of each recipe, but you can follow the builder’s train of thought even better by adding a recipe short description. This description is displayed when you hover over a recipe, allowing for a simple overview of each step in the flow without clicking into the recipes themselves.
The recipe short description should include:
- The purpose of the recipe.
- Business logic or reasons for removal of rows.
- Key facts that would allow the user to understand the flow.
To edit the short description, click on the recipe and go to the information panel on the right side of the screen. Under the details header, navigate to About and click Edit. Enter the short description and click Save. You can also add a long description which will show in the Details panel to the right. In the long description, you can even use Markdown to add formatting.
Prepare Recipe Comment
Within the prepare recipe, Dataiku allows you to create steps as part of the preparation script. While these steps give a lot of transparency into the actions taken on the data, it is still important to add comments to any step that isn’t straightforward, or steps you want to remember the reasoning behind.
There should be a comment on the step if:
- There was business logic incorporated.
- Any step rows are removed.
- A column has been duplicated or “find and replace” was used.
- It is a formula step.
- A column is renamed.
To add a comment to a preparation recipe step, open the preparation recipe and navigate to a step in the script panel on the left. Click on the “i” symbol on the step to comment and edit. Select Always show comment to keep it in view.
A few minutes of documentation will save you time, help you collaborate with others, and provide you with troubleshooting assistance in the future. Dataiku’s visual recipes already provide valuable documentation out of the box with other great features such as wikis, dashboards, and the timeline on the project page.
These features combined with our list of additional documentation will ensure your project stays clean and tidy and ready to reuse again!
