Big data has already penetrated with every single aspect of modern business, having become a front-and-center topic in a global business environment. Big data analytics is key to unveiling user behavior habits and psychology, which allows for deriving truly remarkable insights to be used as guidelines along any company’s digital transformation journey.
However, very often top managers (including Big Data proponents and evangelists) do not understand what is required to turn Big Data into an innovative tool of precise business forecasting, informed decision making, and actionable results. Many C-level execs still believe that hiring one data scientist will help find a bottleneck in business-generated data and bridge their most critical data-related gaps. As the practice shows, alas, this is not the case.
In this article, I’m going to explore what it takes to build a capable data science team (both in-house and outsourced) and how to turn this team into a well-oiled machine processing huge data volumes and generating commercial value out of insights.
Managing expectations from data science
One of the crucial things about creating a data science team or department is to set the right expectations and KPIs. With data science, as with any other innovation, you need to go through the entire cycle, starting with operational losses. At best, your investments in data analytics architecture and tech hiring can give you returns after six months, or more often after a year or two (or even three), depending on your company size and revenue. You’ve got to be ready for screw-ups and disappointment down the road, so approaching your data science and analytics strategy the right way from day 1 is an essential step.
Here are our five steps to building a stellar data science team, each having been tested and verified on real-world client-tailored custom software dev teams at 8allocate.
Five Steps To Fostering a Data-Driven Culture
1. Start small
The best way to get started is to initiate a proof of concept (PoC) project that shouldn’t be very complicated, cumbersome, or lengthy. For instance, if you have an online marketplace, you can set a goal of increasing monthly revenue by 2% as a result of creating and implementing an AI-based recommendation system. Don’t set rocket science goals at this PoC stage like assembling five custom neural networks, as they’re laborious, time-consuming and expensive.
Even if you have in mind a project on the classification of texts, you can still start with creating simple algorithms such as a bag of words and see a revenue increase at the end of the month. So, you need to start small for quick wins and then scale as your expectations from big data increase.
As a result, this pilot project will be a starting point for further development of data analytics architecture and will help get your top management’s buy-in to carry on with the initiative and increase your investment in data analytics.
When you lack the required skills and competencies in-house, it makes sense to hire an external team of data science consultants for the pilot project onshore (i.e., within your home country) or offshore/nearshore (i.e., overseas). They can help bring your ideas to life in a high-quality way as well as show the direction for further AI and data analytics strategy development.
Please keep in mind that when you outsource your pilot data analytics project to a 3rd party provider, most of them will insist that you hire a full-time resource and pay for it even though they’ll sit on their hands for 60% of their work time. At 8allocate, we offer an extremely lean payment model and fractional resource allocation, which means that you’ll only pay for what you actually use. As such, if you need to hire just one data scientist to work two days a week on your pilot project, you won’t have to hire a full-time specialist, and you’ll only pay for two (out of five) days of using an external consultant with us.
2. Collect data
It’s both simple and complicated: ideally, a company should use all of the data it has at its disposal. For example, if you’re an online retailer, you have at least some data on specific product sales, customer behavior on the site, and email campaigns. Many automated models such as personalized mailing lists can already be built with all this data at hand.
In fact, collecting all the company’s data in one database is often a big problem due to differences in sources, lack of clear interaction between departments or even lack of Business Intelligence (BI) specialists within the company. Organizations that have all their data stored in Excel should first start collecting it in a database (SQL).
All available data should be collected in a way that is convenient for data analysts and data scientists (SQL has proved to be a convenient method on many data science teams). You need to agree in advance with your BI specialist/team on the formats in which you want to receive data, process, and use it in production.
“You don’t need data scientists when you don’t have data to analyze!”
If you can only collect a very small amount of business data, feel free to buy them from credible 3rd parties. For instance, you can try person-based marketing tools like Marketo or influ2, which let you leverage their proprietary data for your benefits.
3. Find and attract data science talent
In the best-case scenario, you already have some kind of an internal analytics department or team before you start building your data science team. Analysts will help funnel your business data to data scientists, explain their sources and origin, and show how to collect the right variables, etc. If you’re a startup or a small company, your webmaster or Google Analytics admin can be regarded as an analyst to help ramp-up the launch of your data science team. Even having Google Analytics put in place and configured adequately can deem as the first step to building a data-driven culture within your organization.
When you’re done with your pilot project, you need to scale and actively develop your data science team.
The Internet is flooded with articles about how to structure a data science team the right way, so I’ll skip long descriptions and will only wrap up what’s already well-known.
For your data science team to be full-fledged and effective, you need to hire the following key roles:
- Project Manager to oversee the project and be responsible for the entire business intelligence (BI) development and maintenance;
- Data Scientist to build different data models;
- Data Engineer to collects data and prepare production pipelines; and
- Software developer to implement a custom Data Science solution, do API integrations and create machine learning algorithms, etc.
All these roles are optional and can change depending on your current goals. For example, some data science teams involve a business analyst and a QA engineer, while others merge data engineer and developer roles. That being said, there are many options for data science team configuration, and it’s up to a business to decide which structure to follow.
Besides setting up a data science team, you’ll most likely need to hire a big data evangelist who’ll educate all of your data project stakeholders on data analytics and how to unleash its potential to affect business and make an impact on revenue growth and who’ll help your company gain a reputation as a thought leader for data analytics (if you have such an ambitious goal to have a good bite of the global big data pie). This role can be played by your data science team lead or project manager or even your company’s CTO/CIO/ Chief Digital Officer (CDO).
Also keep in mind that if you hire one data scientist and assign to him/her analytics, architecture, or even software development tasks, don’t expect any quick or effective result. You’ll most likely end up having a highly demotivated employee on the edge of the occupational burnout.
“Most companies estimate they’re analyzing a mere 12% of the data they have, according to a recent study by Forrester Research”
If your company is large and has many opportunities for Big Data development, you’ll definitely need to hire a good data architect to help you build data architecture, set up multithreaded data collection and deploy Hadoop or Spark (tools for processing large amounts of data), which your data scientists will be working with down the road.
4. Foster inter-department communication and professional motivation of data scientists
Now when you have your data science team up and running, start developing your team competencies. For this purpose, your company needs to organize two types of Big Data training: one for data team members and the other one for your top management and project stakeholders.
Training for data scientists
This may include knowledge sharing sessions and workshops, weekly meetings, hackathons, and masterclasses. If you have a core data science team and backup teams (e.g., set up offshore), you need to organize knowledge sharing business trips between two or more locations. This will also help keep your distributed teams synched and on the same page and increase overall team morale by fostering collaboration in a friendly, open-minded, and culturally diverse environment.
What also proves to be effective is purchasing an online course (on Coursera or Udemy) for the whole team or individual team members and even putting it as a KPI. Again, this will help keep the team up to date in a rapidly changing area and improve internal communication.
Training for PMs and top managers
These can also be workshops in the form of business case studies, review of AI strategies of different companies, as well as extracurricular courses on the basics of machine learning and, say, deep learning. This is exactly what will help your senior management set the right expectations for data science endeavor.
Also, there’s always a chance that your employee in a different department may have aspirations to develop as a data scientist. So don’t overlook them and make sure to give them opportunities to gain and develop skills (yes, they can and should be involved in your data science team being built). For instance, by training your WordPress developer from HR team in machine learning methods, you can get a competent and motivated specialist who knows the internal structure of the company and is way cheaper than the average data scientist on the market. Besides, such re-skilling of existing employees can help save your data science hiring budget by eliminating skills to be sought from outside.
5. Data science team outsourcing
Data science jobs are among today’s highest paying careers around the world, thanks to the rapid pace of data collection and business value derived from its analytics. While tech giants outbid any small company in terms of developer salaries and drain local data science talent pools by offering attractive perks and social benefits, startups and SMEs have to struggle to find and hire the necessary skills. In some countries, time to hire a right data specialist can be very long, from a couple of months to a year or so, which significantly slows down the implementation of corporate data science strategies and hinders digital transformation and automation of business.
Data Science Job Market 2019: Trends, Challenges and Expectations
According to the August 2018 LinkedIn Workforce Report, over 151,000 data scientist jobs remain unfilled through the United States, with acute talent shortages in NYC, San Francisco, and LA.
Data science team outsourcing can be a healthy alternative for companies struggling to find and attract appropriate data talent onshore.
The principles of building an outsourced data science team are the same as creating an in-house one. However, when choosing a data science outsourcing vendor, we suggest you pay attention to the following criteria:
- Access to a vast data science talent pool and a positive track record in the internal job market (the better employer reputation the company has, the higher the chance it can attract seasoned talent to work on your outsourced data project);
- Flexible engagement and pricing models (avoid getting trap in vendor lock and make sure your prospective vendor has a good contingency plan put in place);
- Ideally, you should be able to delegate/hire from scratch a product owner who’ll work/communicate with your offshore data science team on a regular basis and will be responsible for project milestones and delivery;
- The vendor lets you test drive your prospective data science team members for free or replace them fast if they don’t match your core team processes and corporate culture and values;
- The vendor assigns an account manager to help you evaluate your team performance, bridge skills gap, do due diligence and scale your team depending on your current project needs and economic situation;
- The vendor has access to STEM students and interns as they can easily be nurtured and grown as internal data science workforce;
- Your data science outsourcing vendor has a robust data security policy.
So what has been your experience with building and managing in-house or outsourced data science teams? Please share it in the comments below!