Machine Learning Ops: from experimentation to successful implementation

Many organisations are using Machine Learning (ML) to try and gain a competitive advantage. However, a large chunk of them are failing as the transition from experimentation to implementation proves too difficult. Up to 80 percent of all ML projects in 2020 will remain stuck in limbo, according to global research and advisory firm Gartner. VIQTOR DAVIS can help your organization to get its Machine Learning projects off the ground and finally deliver on the promise of ML.

Eliyahu M. Goldratt, author and originator of several management paradigms including the Theory of Constraints, coined the phrase “Automation is good, so long as you know exactly where to put the machine”.

This, to an extent, also applies to Machine Learning. One of the reasons why ML projects fail is because there is an initial focus on proving that these can add business value within the company, while questions on how this value can actually be generated by implementing ML applications are overlooked. Enter Goldratt: we have proven that ML is good for our company, but now we don’t know where to put it or how to implement it.

Why Machine Learning solutions fail

There are a number of reasons why the implementation of ML solutions fails. An obvious one is that the actual implementation requires more of everything, as well as additional considerations, than when the project remained in the experimental phase: more and better infrastructure, more collaboration between different teams within the organisation, more stakeholders to tend to, and so on. What started small now needs to be rolled out throughout the entire company, bringing with it new challenges not considered before.

Complementary to this is the fact that many ML initiatives are isolated. Often, ML projects in an organisation are completely disconnected, leading to inefficiencies and hindering the sharing of best practices across different teams.

Furthermore, the risk of ‘black-box decision making’ – i.e. when decisions are based on opaque ML applications with non-visible inner processes – is often underestimated in the experimental phase. But when such an application goes operational, businesses need to make sure ML-based decisions are auditable and can be explained to demanding clients, auditors or stakeholders.

ML Ops as an enabler

Having invested heavily in a ML initiative within your company, it would be a shame to let all that time and resources go to waste. That is why a package of best practices has been formed to help businesses move their ML applications from experimentation to successful deployment: ML Ops.

ML Ops is paramount in developing ML governance processes (pipelines) and deploying trusted applications, while also focusing on business and regulatory requirements. The key to this set of best practices is automation. When as much as possible of the ML application is automated, the less it will be prone to human errors, the easier it is to actually implement it and the more consistent it becomes.

Apart from automation, other best practices include manageability (enforcing model governance and track changes to models and code throughout the development lifecycle), reusability (assuring continuous delivery and that the same configuration can be re-used) and reproducibility (code, models, libraries, SDK’s etc. are versioned and maintained such that they can be reproduced).


VIQTOR DAVIS helps organisations to accelerate effective and efficient deployment of ML applications, so that these can (finally) generate business value.

Because we have blueprints and ML pipelines readily available, we can get right to work and deploy your ML applications using the technology of your choice, whether that is Microsoft Azure, Google Cloud or any other. We deliver ML pipelines with speed and agility which automate processes across the ML lifecycle. By using these standardised pipelines we ensure that you can rapidly experiment with new ideas, implement value-adding applications and scale them throughout your organisation.

We also have a track record in transparent ML applications; we can make sure your applications are auditable, helping you to avoid black boxes and allowing you to perform in-depth analysis on ML applications and decisions.

VIQTOR DAVIS employs over 60 data scientists and is a Microsoft AI Inner Circle and IBM Business Partner. We are a frontrunner in NLP, AI and cognitive computing. If you would like to know how VIQTOR DAVIS could help you implement ML applications and scale them throughout your organisation, please feel free to contact us.

Read about the 'Machine Learning (ML) Ops Accelerator'.

Recent articles

Learn about our Media & Events
learn more

Thanks for reaching out. We'll be in touch.


Successfully subscribed to the VIQTOR DAVIS newsletter