30 Jun

11 Ways to Get Your Machine Learning Model Into Production

Sean Robertson

A staggering 90% of machine learning models initiated by businesses never make it into production.

Not only does this leave companies massively out of pocket, but it’s also bad for morale as teams fail to see their work coming to fruition.

With the global business value of AI estimated to hit $3.9 trillion this year, it’s critical that companies fix this and get into that 10% of models that make it over the line.

Here’s how:

Identify a problem that’s really important for your business to solve

When starting a data project, people tend to jump straight to the technology but it’s important to first take a step back and think about what you are trying to achieve.

Think of a business problem you’ve been trying to solve that has clear KPIs associated with it. Make sure it’s a big enough problem for the business to want to invest money into it.

Business leaders want to make sure they’re getting value from data projects so think about the benefits both internally and to your end users.

Once you’ve worked out what the problem is you’re in a much better position to start thinking about the data you will need to access in order to solve it.

Ask the right questions at the start of your project

Now you’ve identified your problem, you need to ask the right people the right questions to make sure you are all headed in the same direction.

So, start asking: ‘What do we want to achieve?’, ‘Who’s going to be involved in the project?’, ‘What are the enablers that will drive it forward?’ and ‘What are the potential blockers?’

You should also think about the bigger picture in terms of your business: ‘What are you trying to do as an organisation?’, ‘What's your purpose?’, ‘What's the big thing you want to try and work on and solve?’

It’s also crucial that you are able to measure and prove the success of your project so ask what the metrics are and how you are going to access the data.

Make sure you know what the business value is

Right from the beginning, it’s important to know how your ML project is going to impact the business.

Perhaps it will boost revenue growth or help to make cost savings? It could be that it will make a huge difference to safety for employees or transform day-to-day tasks for your customers.

Think about what constitutes value to your business leaders and define your headline KPIs so that everyone is aware what you are working towards.

Data literacy and understanding of AI is quite low out of the data science community so you need to talk in terms of tangible goals that everyone can understand.

Remember that the vast majority of problems aren't technical

While you will come across some technical problems that you need to solve, remember that there are a huge number of solutions out there for these and they are fixable.

What you’ll quickly find is that a lot of the problems aren’t technical at all – they generally involve people and processes.

People are generally the hardest part of the puzzle to move forward with.

Think about how your organisation is set up, break down any silos and make sure the communication is there so that people can be fully invested in your project.

Start small and scale up

It’s fine to have a big vision and a big picture but it’s a good idea to start off by getting the basics right and then scale up.

Choose a key problem for the organisation, making sure it is one that is solvable and will demonstrate true value to the business.

Try not to focus too much on the technology discussion at the beginning of your project. You’re not just talking about a data lake or a data warehouse, you're talking about something that is a product.

Always think about your people – whether that’s your team, suppliers or end users – and how this product is going to benefit them.

Make sure there is clear ownership of the project

If you start thinking about your data as a product, it’s much easier to identify data product owners.

There has previously been a history of a general lack of clearly defined ownership of data in companies.

Traditionally the owners have been IT, who will have a different objective for a project than someone who is in Marketing or Risk.

Ideally data product owners should be domain owners who are close to the business.

By working in this way, you have people taking ownership who are much more engaged in what the product is, how useful it is and how it will improve the business.

Decide what you want your business to be famous for

Pick one area of the business that you want to be really successful and known for and make this the focus of your project.

Think of businesses such as Tesco who wanted to be known for loyalty or Netflix who want to be known for their recommendation engine.

Once you have chosen your core USP, everything else falls behind that.

Make sure your data science team isn’t isolated in the organisation

It’s common for organisations to hire in data science teams and then fail to integrate them into the rest of the company.

Silos need to be broken down so that the data scientists aren’t locked away in a room working on their own but are integrated into other teams such as IT and product.

Squad-based teams that work by domain can help to achieve true integration and make all team members feel fully bought into your projects.

Hire the right roles into the business

We’re starting to see new roles emerging as companies get to grips with the fast evolution of technology.

While product owners have always had a presence, we are now seeing organisations recruiting for the more specialised data product owners as well as MLOps engineers.

These roles will help to bridge the gaps in the organisation as well as increasing understanding of data throughout the business.

Adopt more agile ways of working

We’re going to see an exponential growth in companies adopting new ways of working over the next few years.

There has traditionally been a reluctance to make sweeping organisational changes but companies are now starting to see the benefits of new ways of working and how it can lead to true transformation.

Learn from other organisations who are doing it well

Some sectors such as the financial industry are much further ahead in terms of how they are using machine learning. Don’t be afraid to research what other businesses have been doing and to learn from them.

Manufacturing and utilities, as well as many oil and gas companies are closer to the start of their journey so will be able to take all the learnings that exist for other industries.

If your business is currently starting out, you will be able to find out about some of the barriers that slowed things down in other sectors and then plough straight through them.

Final Thoughts

Starting out on a ML project can feel daunting for many organisations who may feel overwhelmed due to their lack of technical understanding.

But as this list demonstrates, the majority of problems aren’t technical at all but are about getting the basics right with your people and your processes.

No one wants to spend their time and energy working on a project that doesn’t make it into production.

Follow these fixes and you will be able to get your data project over the line and into production, transforming fortunes for your people, your business and your customers.

Interested in seeing our latest blogs as soon as they get released? Sign up for our newsletter using the form below, and also follow us on LinkedIn.

Latest Stories

See More
This website uses cookies to maximize your experience and help us to understand how we can improve it. By clicking 'Accept', you consent to the use of these cookies. If you would like to manage your cookie settings, you can control this in your internet browser. Find out more in our Privacy Policy