Cloud-based service provisioning has entered the machine learning ecosystem, and the outcome is MLaaS. The long and the short of it is that machine learning-as-a-service is exactly what it sounds like — it's the delivery of accessible, hosted and subscription-based machine learning tools that are ‘ready-to-go’. But is MLaaS right for you? And is there more to understanding the nuance of MLaaS?
MLaaS is bridging the gap between theory and implementation of ML technology for many businesses. But machine learning isn’t the right tool for every job, and even if your data and processes are perfectly suited to an ML solution, MLaaS has some downsides. Building an in-house tool, purchasing bespoke APIs, working with consultants, or drafting a hybrid approach are all options you need to weigh.
Here, we will walk you through some of these considerations and explain the best use cases for MLaaS in businesses of all kinds.
What is MLaaS? What can MLaaS deliver?
Machine Learning-as-a-service delivers ready-made or bespoke software solutions to bring to bear on a range of business operations. It leverages sophisticated algorithms and neural networks to help businesses harness their customer data and process it to achieve deliverable outcomes. These might include:
- Increased operational efficiency
- Reduced cost of running and managing the ML infrastructure
- Greater customer engagement
- Building more accurate and detailed customer profiles
- Targeting customers who are at risk of becoming disengaged with the brand
- A combination of the above
What makes MLaaS so appealing is that none of the computation needs to be done on site. It is all managed remotely by the service provider’s data centres. So there doesn't necessarily need to be any expensive and disruptive hardware upgrades — much less a need to invest in developing the ML algorithms in-house.
How does MLaaS work?
Different MLaaS providers operate differently, and provide different outcomes. The four main providers are the same big names you’ll recognise from the cloud-provider space: IBM, AWS, Microsoft and Google.
Although the above table shows some of the basic options, each platform has a wide range of additional capabilities — all of which is worth investigating in detail. Among the algorithmic models used by MLaaS providers are pattern recognition models, deep neural networks (DNN) and convolutional neural networks (CNN), probabilistic graphical models, and Restricted Boltzmann Machine (RBM) models.
Pricing for these services can also be slightly complex. Normally, you will be charged based on a combination of an hourly rate for compute time, and then set rates for batch predictions once your ML solution has been developed. For AWS ML, for example, that would be $0.42 per hour and $0.10 per 1000 predictions. Other solutions like Microsoft’s Machine Learning Studio operate on a subscription, but with limits to experiment duration.
Picking the right MLaaS tool requires looking at capabilities, pricing and your own internal goals and data flows. Before investing in any IT solution, you need to think about what you want delivered and how you are going to execute that plan. Actually making sure that your data and goal align with what MLaaS can deliver (or ML more generally) is a critical first step and isn’t something MLaaS can directly help you achieve.
How to support an MLaaS solution
MLaaS promises to make machine learning and AI simple. But the reality is slightly different. To have success with MLaaS, you still need to enter the project with a plan and expertise. Increasingly, this is where consultants come into play.
The value of partnering with a consultancy comes from accessing ML/AI expertise from the start. The upfront demands of developing and deploying an ML tool are often higher than the maintenance requirements. For small businesses particularly, it makes far greater economic sense to bring on board the expertise when it's needed, and avoid the long-term costs of staff retention. Consultancies also offer access to leading talent that you may not be able to afford if hiring full time.
Regardless, if you want an effective MLaaS solution, you can’t just go and sign up for a subscription. You have to identify your goals, assess your data and build a solution that is fit for purpose.
Potential applications for MLaaS
The potential applications for MLaaS are many and varied. Service providers tend to create solutions in line with specific use cases. Understanding your options is a critical first step. To that end, here are some of the main outcomes you could look to achieve using MLaaS:
Recommendation engines — Perhaps the most common form of Machine Learning, MLaaS solutions can make it quick and easy to provide customer recommendations based on their browsing and purchase histories and previous interactions with your brand. You can see this at work whenever you log into a video streaming platform or access your eCommerce accounts.
Natural language processing — This is used increasingly commonly by search engines, helping them understand the nuances of human speech, as well as sentence structure, context and simile to deliver higher quality and more relevant search results. They are used to help search engines understand their users better rather than the other way around.
Customer Lifetime Value Modelling — How do you know which customers are worth your time and which you can let go? Customer Lifetime Value Models analyse large volumes of customers’ behavioural data to help determine which customers are the loyal big spenders and which are the casual chancers. These are commonly used by eCommerce businesses who want to give their customers the maximum value in an increasingly competitive sector.
Building customer personas — MLaaS providers can also provide easy-to-implement solutions that allow businesses to create better-segmented customer personas. They can leverage great volumes of historic data to help preempt customer behaviours and help to make groups of customers easier to market to. This can make for much more effective and cost-efficient advertising.
Image recognition — It’s not just stats and facts that MLaaS providers can help businesses to marshall. They can also help to implement a range of image recognition and facial recognition solutions for a wide range of practical applications.
Who can benefit most from MLaaS?
The beauty of MLaaS is that it’s industry agnostic. It can be leveraged by businesses in a variety of sectors and use cases. Let’s take a look at how some specific sectors can benefit from MLaaS.
As retailers scramble to engage their customers, provide value and stand apart from their competitors, MLaaS can help them to better leverage their customer data and create tailored and secure customer experiences, thereby building value in the brand.
As well as making recommendations and delivering a more tailored approach to marketing and promotions, MLaaS can also help to reduce the risk of fraud and optimise inventory management for a smoother and more efficient workflow.
Media production houses
MLaaS can leverage sophisticated image and facial recognition algorithms at the point of ingest to help busy production houses to get quicker and easier access to their archives and improve their workflow to keep up with the rising demand for video content.
MLaaS can help healthcare providers in a number of ways. They can use data lake technologies to streamline the research process as well as leveraging patient data for effective clinical analytics and risk management. Not to mention using Machine Learning to help with predictive maintenance to ensure that complex medical equipment remains in peak working order.
Financial and insurance
Finally, the financial services and insurance industries can also utilise MLaaS to predict market fluctuations based on historic and consumer data or help to guide strategies for forex investment as well as assisting risk analysis for the insurance industry.
MLaaS is here to stay, but it’s not the only option
MLaaS is really just one growing element of a broader ML ecosystem. Although it’s dramatically reducing the barrier to entry, the biggest MLaaS pitfall you can fall into is thinking that “as-a-service” means “no planning required”.
A lot of MLaaS tools are just that — tools. They can provide ready-made functionality within a broader ML strategy. Linking together MLaaS functions with in-house development teams can shortcut your way to building a bespoke solution able to scale. But no matter what strategy you take, you need to think carefully about strategic and tactical planning. Set goals, assess your data and take deliberative steps towards success. Consultants can help you do this right, first time. But no matter who you get there, those simple steps need to form the bedrock of your plan.
As we move further into a data-driven economy, the importance of getting a handle on your data flows increases in value. The MLaaS market is predicted to grow by almost 50% by 2023, so you can be sure that if you’re not looking into this service, your competitors certainly will be. However, PwC predicts that AI will contribute up to $15.7 trillion to the global economy by 2030. The move towards intelligent systems is bigger than ML, and it’s bigger the MLaaS.
Don’t put all of your eggs in one basket — investigate your options and plan. By pulling together the right combination of tools, you can make the most out of this moment of change.
Is Machine Learning right for you? Get in touch with our team now letting us know a bit about your current situation and we'd be happy to explore this with you.