BusinessTechnical

How To Effectively Implement Machine Learning at Your Company

By February 5, 2019 No Comments

It can sometimes feel like everywhere you look your business competitors are harnessing machine learning and artificial intelligence solutions. From improving fraud detection to better targeting online shoppers, the benefits of a machine learning solution are practically limitless.  We get it, machine learning seems daunting and impossible, but with the right solution you can be making smarter business decisions in no time.

With a growing number of data scientists in the workforce and machine learning tools being released, leveraging a solution at your company doesn’t have to be a long distant  or unattainable dream.

If you’re interested in exploring machine learning or artificial intelligence solutions for your organization, here are four options for how to get started and the pros and cons of each.

Option 1: Hire a Consultant.

Pros:

Hiring consultants brings expertise, efficiency, and a business-minded perspective to your AI/ML solution. The consultants come with previous enterprise data science experience, and can share their knowledge of what went well or what you should stay away from. Not only have they designed and built similar solutions, but they have also seen it in various business environments. That background allows them to approach your solution with context of what other companies in your specific industry are doing. It also allows them to keep a a high-level, business-oriented focus while designing a very technical solution. That perspective keeps your company at the center of focus.

Cons:

As discussed, consultants bring a level of experience and expertise that helps your project move quickly and efficiently. This expertise, however, can also come with a corresponding price tag. Although you are likely to keep a cost-effective pace and timeline with consultants, you are still paying an hourly rate for a team of people. In a general project, that could mean having a solution architect, a project manager, and one or more developers – all with a billable rate.

Furthermore, outsourcing to consultants means that you are not building the skills in house to maintain the solution that the consultants build. Once the project goes into production, who will maintain it? Who will make sure that the models are still learning or are still predicting accurately? Who will debug if something goes wrong? Does that mean more support hours from the pricey consultants?

Option 2: Hire a Data Scientist (FTE).

Pros:

When you hire a full time data scientist you are investing in AI/ML  for long term projects in your business. They will be able to maintain a machine learning production environment and have the ability to build out additional data science projects in the future. A data scientist is highly educated and will typically have a masters degree or a PHD in mathematics, computer science or a similar field. They come in with the knowledge and skills needed to execute complex algorithms and statistical models.

Cons:

Even though data science is becoming a popular career path for many people, they can still be hard to find. Sourcing the right data scientist for your company can be a challenge. Additionally, their extensive education comes with a high price tag. The median salary for a data-scientist is $128K. Most data scientists require a team to work under them as they don’t typically manage the data architecture and data preparation.  Lastly, when you invest in a data scientist you are putting all your technical investment in a single person.

Option 3: Invest in an automated machine learning tool like DataRobot.

Pros:

Automated machine learning tools like DataRobot are a great way to expedite the modeling process and create a user-friendly, visual representation of what really goes into a machine learning solution. It’s the perfect way to jump start a machine learning solution and get rid of the preconceived notion that AI/ML is a “black box”. By using a tool like DataRobot, you can more clearly and simply show business users what is going on behind the scenes, centralize your solution to one area, and (to an extent) lower the technical/coding involvement of the data scientists.

Cons:

Using a point and click tool for modeling and implementing a data science solution is an excellent launching point, but it’s not an all-inclusive process. Those tools tend to handle the brunt of the modeling work at the center of the solution, but there is still a need for what comes before and after the modeling.

Firstly, the data  needs to be engineered, cleansed, and most importantly, understood in the business context. Next, it is fed into a tool like DataRobot. After that, you still need someone who can interpret, tune, and productionalize the results. This requires the expertise of people with business knowledge, statistical/programming knowledge, and data architecture experience.

Option 4: The Hybrid Approach.

Just like Goldilocks, sometimes we desire something in the middle.

Consultants bring expertise and perspective, but they’re temporary. Data Scientists know what they are doing and are long-term, but they are expensive. Tools like Datarobot are digestible to business people and they’re effective, but they don’t do it all . How do we meet in the middle?

The best way to combine all of these options is by using all three of the options in their strongest areas. More specifically:

Hire a lean team of consultants to start you off.Use the consultants for their expertise and advisory. You want what they bring to the table, but you want them to build up YOUR team long term. Let them create a solid foundation for you, but suggest an agile, lean time to kickstart your ML initiative. Ultimately, you want to create a team for the project that is part consultant, part in-house developers.

Use DataRobot for the part it’s great at. Use a tool like DataRobot to get business buy-in and to demystify the perception of AI/ML. It will also save your developers an immense amount of development time.

Work with the consulting team to hire a Data Scientist or to identify someone on your own team that can grow into the role. Use the consultants’ network to help you find someone who can maintain what they build. They know best what they built and what skills are necessary for maintaining the solution. On the other hand, if the consultants are working with your team, they might even be able to identify some promising member(s) of your existing team that you can leverage. Sure, they may need some training, but the investment may be safer/easier than a new FTE.

Aptitive’s analytics practice understands your data science needs and can help your organization find the perfect solution. Between our data science consultants and our technology partnership with DataRobot, leveraging a machine learning model at your organization doesn’t seem so impossible anymore. If you’re looking to bring data science to your organization, contact us for more information.

This piece was co-written by Aptitive Consultants, Ashley Pradhan and Erin Cullen.

Ashley Pradhan is a Senior Data & Technology Consultant at Aptitive. In her role, Ashley helps clients to find valuable insights through the entire data management lifecycle including, strategy, design, architecture, ETL, custom development, business intelligence, reporting, analytics, and visualizations.