The last year, Digital Transformation became one of the top priorities of the CEO agenda (KPMG research). As I have experienced, most data analytics projects to not reach their estimated deadlines. Moreover, according to Gartner analyst Nick Heudecker, 85% of all big data projects do not reach their goals.
Companies have to deal with multiple types of generations (Y,Z, Millennials) and the corresponding knowledge gap between the tech team and the management. Therefore, the teams are difficult to manage, resulting in disappointing results and higher costs.
Obstacles that teams have to deal with can verify from poor data quality, missing skills or an excising solution that was founded in the market with the same business value that could be licensed for a fraction of the project price.
However, CEO’s can overcome unforeseen problems by using a simple structure to start and guideline the project.
In this blog post I will show you how to overcome problems and reduce the knowledge gap.
Containments of this article:
- Define a value proposition
- Validate a data science/analytics business case
- Structure the team
- Set up feedback loops
- Execute the plan
Firstly, what problems could occur after starting a data science/analytics project?
- There was something wrong with the data (low quality/incomplete/missing data)
- The enduser of the project didn’t feel the expected business value
- The (technical) model didn’t work, meaning that there was no sufficient technical knowledge to foresee or solve the problem with the model
- There was a lack of in-house project management (connecting departments to each other)
To foresee these possible problems, you have to do research.
Therefor, when discussing a potential business case, choose the right starting point: from a data point of view, or a business value point of view.
You want to take both steps, but can choose to start from one point.
When doing research in data quality assurance, try to answer the following questions:
- Which data is available?
- What is the ratio of data errors?
- In what way is the data delivered?
- What is the completeness of the data? Is data missing?
- What is the uniqueness of the data? Are there duplicated rows?
- How recent is the data?
- Is the data validated? For example: a form that asks your number could be ‘asdkasdk’.
For more information, here is a whitepaper about data quality assurance.
When analyzing the possible added value in a business perspective. Try to answer the following questions.
- What is the current process?
- What is the flow of the business value between the stakeholders?
- Describe each stakeholder and their roles in this process
- What is the new solution and how will it effect the current way of working?
- Which stakeholders will be positively affected and which negative?
The overview below is a variant of the business model canvas, which is still a useful mapping tool. For instance, in here you can describe each stakeholder and their value gains.
This design toolkit consists of 8 steps to create a MVP. Here you find more information.
Secondly: how to define a value proposition.
Keep in mind, you are still doing research. During this phase the value proposition should argue the feasibility of the project. In other words: it should result in a GO/NOGO decision by the management. The previous steps in this handbook should give you, in collaboration with the management team and the data analytics/science lead, the answers to the following questions.
This method refers to the interview case guidance by a former McKinsey Consultant and the handbook of ICC.
With this format, the management should to decide whether to go forward with this proposition or not. After doing so, the next step is to dive deeper into the proposition, validate the business case and manage expectations (time, money, effort).
Next: validate the value proposition and project planning.
Like any investment, double check the management expectations, economic and technical feasibility.
Make sure everyone is one the same page.
After that: structure the team.
After deciding to invest in the proposition, the management should address a clear description of responsibilities.
- The management is responsible for connecting the team lead with other departments of the company
- The management allocates the resources (time, money, effort, information)
Choose the right Product-Owner. He or she is the most important team member.
The Product Owner (PO) is a Principal or experienced technical leader. The main responsibility of the PO is to set out the vision and force the development team to stick to it. You don’t want a PO who changes direction all the time. The PO clearly aligns end-user expectations with the scrum team.
He or she is also the person who gives the management feedback and is the translator of difficult technical challenges that are faced. The product owner should also be capable of setting up his or her own data analytics/science team and responsible for hiring team members.
For a clear overview of skills and roles: use this overview created by altexsoft.
Execute the plan.
You want to choose a lean or agile way of working as you want to find out as quickly as possible if the project will deliver the expected business value.
This method will help you to focus on feedback and validation of the product before you scale up.
- Analyze the quality of the data.
- Max. 2 weeks
- Needed skills: understanding of how to interpret data, f.e. (My)SQL/Oracle/MongoDB/SAP
Proof of concept
- A simple dashboard that should the value of the project. This dashboard can be build with fake data. It should show the vision of project and where it wants to go.
- Max. 6 weeks
- Needed skills: Visualize the concept, f.e. Tableau, Qlikview, Shiny.
- A running model with real data that proves the value of the project. This product should be able to run independently.
- Max. 3-6 months (depending of the complexity and available FTE)
- Needed skills: product owner, data engineering, data architecture, data analysis, data science
- The model is integrated in the existing architecture of the company
- Max. 3-6 months (depending of the complexity and available FTE)
- Needed skills: mostly (senior) data engineering and data architecture
Set up feedback loops with the management.
After the data exploration, make clear agreements about feedback loops.
Preferably you want to have a daily stand-up meetings with your data team and a brief update to your management once per 2-4 weeks, depending on the phase in the product funnel you are in.
More information about (daily) stand-up meetings: https://www.atlassian.com/team-playbook/plays/standups
Besides ‘internal feedback’ you will need end-user feedback. Make sure to give demos of the product to the end-user and gather feedback once per update.
Finally, create ownership and responsibilities.
Besides the PO, you need a bridge to your business: the Business Owner (BO), mostly the COO. The BO makes sure that the PO is facilitated in everything that he needs, which is mostly input and end-user feedback. The PO is responsible for the deliverables and the BO ensures that the stakeholders have a representative who is available for quick questions. This way, you maintain a high pace of delivery.
About the author
My name is Derk Disselhoff, the founder of Worksuite. Worksuite is a freelance platform focused on Data Science/Engineering projects in the Dutch sector. I welcome your feedback to this article on firstname.lastname@example.org. If you are interested, visit our website www.getworksuite.com for more information.