Are predictive analytics and machine learning platforms worth the investment?
There is a lot of buzz in the market around AI. However, does it justify the investment?
Here at OpenText™ we regularly speak with leading companies who are interested in partnering with us and deploying OpenText™ Magellan™ to embark on their AI journey. But when it comes to making the investment in the platform and structuring an organization to reap the benefits from it, the decision needs to be financially justified.
We engaged leading independent research firm Forrester’s senior analyst Kjell Carlsson to offer his insights.
In addition to this insightful video, we also had the opportunity to ask Kjell four questions on the value drivers, outcomes expected, where to use open source technology, and what to look for in PAML (predictive analytics and machine learning) solutions like Magellan:
How large an impact should enterprises expect from AI initiatives? Are they worth the investment?
In my surveys, nearly half of enterprises say that their data, analytics, machine learning and AI initiatives will be the most important factor determining their business competitiveness and will affect their bottom lines by 20% or more. Similarly, there is near unanimous agreement that these initiatives have a positive ROI and over a third report expecting an ROI greater than 5X.
Would it be right to attribute 100% of this to predictive analytics and machine learning (PAML) solutions? No. But you cannot get these returns without effective PAML solutions. These solutions profoundly affect the ability of your data scientists to access the data they need, use the tools that they know how to use, leverage the latest innovations in machine learning, get their models into production and embed the results of those models into applications and into the hands of end users.
Further, the tools help your data scientists collaborate and, indirectly, affect your ability to hire, develop and retain data scientists. It’s not just that these solutions help your data scientists be more productive – it’s about enabling them to drive outcomes in your business on a repeatable basis, at scale, instead of re-inventing the wheel on every project.
What business outcomes should enterprises expect from adopting these PAML solutions?
More customers – from more effective sales and marketing, better products – from improved AI and machine learning based functionality and better customer insights, and greater operational efficiency – from better risk assessment, predictive maintenance and workflow automation – are just some of the outcomes that enterprises have dramatically improved using their PAML solutions. In a recent survey I conducted, 50% of firms reported process efficiencies, 40% reported better employee productivity and customer experiences while roughly 30% reported improved innovation, customer acquisition and retention from using these kinds of solutions.
PAML solutions have allowed these firms to develop the machine learning models and analyses that have led to these outcomes faster, to build more accurate models, to get more of them into production, and to ensure that they stay accurate over time. In essence, they have allowed their enterprises to innovate with machine learning and AI faster, and capitalize on that innovation more effectively.
Be aware though, that just adopting a PAML solution won’t drive business outcomes by itself any more than buying a racing car will let you win the next Grand Prix. You will need data scientists, data engineers, and data-savvy managers to use these PAML solutions to develop the analyses, models and applications that will allow you to achieve these outcomes. However, not adopting an effective PAML solution is like expecting your teams to win Le Mans by giving them a box of car parts.
Why should enterprises turn to these solutions to do open source machine learning? Can’t I do this for free?
Companies should not be turning to open source machine learning in the hopes that it will be cheaper. Instead, the key reasons why enterprises should be looking to leverage open source for predictive analytics and machine learning are:
- To take advantage of scalable open source big data infrastructure (e.g., Hadoop and Spark), and open source application management frameworks (e.g., Kubernetes)
- To leverage the latest source innovations in machine learning methods (e.g. through deep learning frameworks like TensorFlow, Caffe, Keras and Pytorch)
- To access the large, rapidly growing sources of data science talent who are being trained on open-source machine learning programming languages like R and Python
- And reduce the risk of lock-in.
While there are instances where you can enable more users to get started with analytics and machine learning with free open source tools, in general relying on entirely on free open source tools is far more costly to any organization in terms of effort, lost productivity and risk, because open source components have not been designed to work well together and have not been developed and tested to the standards that enterprises need.
A far better solution is to leverage PAML solutions that leverage open source components, but have improved on them, made them work together and developed additional value-added features. These offerings will continue to incorporate new improvements from the open source community in a way that internal teams will struggle to do with in-house solutions based on free open source components.
How should I evaluate which PAML solution is right for my enterprise?
There is no one PAML solution that makes sense for all enterprises. The PAML solution that is right for your enterprise will depend on your PAML use cases, the skills of your data scientists, your existing code base, and the data and operationalization challenges you are looking to overcome. Indeed, most large enterprises will need more than one PAML solution as the unique benefits each solution can bring to bear, in terms of business value, will far outweigh the headache of managing multiple solutions. That being said, generally the best PAML solutions drive productivity in each stage of the analytics and machine learning project lifecycle, streamline the lifecycle as a whole and promote collaboration and re-use across individuals, roles and teams.
Pay particular attention to PAML solutions that have a broad set of features that enable data access, discovery, preparation and governance as well as features that help you bridge the last mile to the end user by making it easy to put models into production and create end-user apps and visualizations – as these stages of the project lifecycle are usually the largest barriers to driving business outcomes.