Why Consulting is the ideal way to implement data science
As more companies and executives are recognizing the benefits of data science, the number of options on how to implement data science is increasing. For me, there are four core types of data science implementation: hiring data scientists, investing in Software as a Service solutions, hiring freelancers, and bringing in a consulting team. While all four options have their merits, I'll go into why consulting is the most effective way to start data science in your company after weighing the pros and cons.
Hire Data Scientists
The easiest and most common way to implement data science in your company is to go out and hire data scientists who are full time employees in your company. Relative to other positions in a company, the demand for data scientists is exponentially higher than the supply. Using Glassdoor, we find that postings per million that include “Data Scientist” are 3 times higher than the searches per million over a 4 – week moving average. This may not be a perfect indicator of the shortage, but the job listings have a 37% growth YOY from 2019-2020 and 67% of companies are in the process of expanding their data science teams, leading to average vacancies of 51 days. This is driving the salaries of data scientists to an all-time high of $113,309 per year. Even if you are lucky enough to hire a data scientist, that does not mean the data scientist search is over because the average tenure of a data scientist is less than 2 years, with 44% of data scientists at any given point looking for another position. This turnover is unsustainable for any company to implement data science in their company. Along with the turnover, data scientists tend to be congregated in big cities with a lot of demand and specialize in industries that service big cities. If you are a company that is not centrally located, it is unlikely you will be able to even convince a data scientist to come to work at your location.
Software as a Service
The next option to implement data science in your company is to purchase a Software as a Service package. S.A.A.S (Software as a Service) packages claim to deliver accurate and actionable results through predefined models using data from other companies leading to a “one-size-fits-all” approach to data science projects. This S.A.A.S approach is not efficient for implementing data science, especially in industries such as manufacturing, where the needs and requirements are unique to each company as every company’s processes and products are different. Implementing these S.A.A.S solutions end up taking up just as much time as building models from scratch because of the complexities of most businesses. Additionally, S.A.A.S solutions create a scenario where you as the client are left feeling as if the end products/recommendations are a black box and you are unable to follow the train of thought behind the creation of the model and recommendation that is given. Even after paying the initial cost, S.A.A.S solutions often time require a recurring fee to be paid every month or year just to use the software. This blanket approach to software does not translate well to data science solutions.
The third option to bring data science to your company is to hire freelancers who specialize in data science through platforms such as Upwork. While these freelancers may be providing their service as at cheaper cost for a set period, the freelancer model does not translate well to data science projects. Freelancers tend to be specialized in one aspect of data science (such as building models in R, deployment, or ETL), but to implement data science, you need the ability to do end-to-end projects beginning with data discovery and ending with model maintenance, otherwise you will not be able to begin or you will be stuck in the middle of a project. You could hire freelancers for every step of the data science journey, but why hire people who haven't worked with each other when you could hire our fourth option: Bring in a consulting team.
Bring in a consulting team
This brings us to the last core option to implement data science: bring in a consulting team. Data Science consulting teams help mitigate the risk of the options above. After factoring in the high wages, benefits, and turnover costs, consulting teams may be cheaper overall than hire in-house data scientists. Consulting teams are also more flexible in the moving outside of big cities as they will most likely be at your location temporarily and/or remote.
Consulting teams, while they use previously trained models and use cases in the development of their models, every solution is tailor-made to the client’s specific data and unique requirements. This allows for a high amount of transparency between the team and the end user and after the project is completed, it is rare for consulting firms to ask for a recurring fee to use the model built for them. Consulting firms are known to limit to a defined market, which allows them to have substantial knowledge about the industry.
Finally, as stated above, data science projects need to be implemented end-to-end, and consulting teams make that happen by having multiple people as part of the project that are dedicated to finding the solution. With a team, there are be people who are specialized in different parts of the data science project, so there is expertise throughout the entire journey, from ETL to deployment/maintenance.
Hiring a team can also allow the team to evaluate the current state of data science in you company and help you develop a plan after they complete their project.
Overall, data science is a new service offering that many companies are still learning to navigate. Bringing in an expert team who have experience working together is the best way to know you are doing data science right.