Introducing AI in Enterprise: A Data Scientist's Perspective

Knowledge gained through experience

During my work at Safran, I realized that while there exist a lot of articles talking about the math and statistics behind ML/AI, few are comprehensive enough to mention the pitfalls that a lot of people fall into while introducing ML into an enterprise.

In theory there is no difference between theory and practice -  in practice there is.

This series of articles is meant for two kinds of people in mind, a younger me - someone who knows the technical fundamentals of machine learning and statistics, has gained experience from Kaggle challenges, but has not worked in a large organization that hasn't set up an infrastructure for ML yet - or someone who is looking to add value to their business by using ML/Data Science.

When you've made the decision to onboard a new technology, there is a multitude of things to consider.  By the end of this series of articles, you will have an understanding as to how you can do so and this can act as a checklist before starting any ML project.

1. Start Small.

All complex systems that work evolved from simpler systems that worked. If you want to build a complex system that works, build a simpler system first, and then improve it over time. - Gall's Law.

When introducing AI in your company it is important to pick a small project and then go from there. Doing so will help you identify the limitations, growth areas, strengths and other possibilities for exploration. You will get insight into your data lifecycle, deployment strategies, infrastructure requirements etc, and avoid costly mistakes.

2. Identify and communicate with your stakeholders and end-users.

Find out which people will positively/ negatively be impacted by these changes.

Figure out the level and frequency of communication each stakeholder will require, manage and set expectations such that they understand that it is better to start small and get quick wins rather than jumping in the deep end and getting overwhelmed. Communicate continuously through the different phases of the project.

Set clear expectations when it comes to the major influencers/ upper management in your company.

3. Calculate the pros, cons and limitations of introducing AI.

Identifying the benefits and risks helps manage and set expectations and avoid non-performance. Here are some you could consider-


Data difficulties, Technology troubles, Skill shortage, Integration challenges.


Increase in the overall efficiency of a process, reduction of human errors in repetitive tasks, optimization of internal business operations, help in pursuing new markets, automation of repetitive tasks, freeing up workers to focus on other problem areas.


Model interpretability (might be a challenge depending on the algorithms used, this may result in a lack of transparency), Model will have to be configured to adapt to the change data over time.

4. Have a data strategy.

There is a popular adage in the world of ML which goes "Garbage in, garbage out" which means that your model is only as good as the data you put in it.

You have to ensure you meet the requirements for data quality, quantity and security.

In many companies, as policies and regulations change, so does the method of collecting data. Many a time what might have been usable at one point in time may not be usable anymore.

Having a robust data pipeline helps you take care of such problems.

5. Recognize operational necessities.

There are various enablers for an effective project, such as computing hardware, training of your employees to provide them with the required skill-set and tools, setting up a development environment.

6. Implement Change Management.

Unfortunately, when a lot of people hear "AI" they imagine robots coming to take over their jobs.

Management plays a key role in the framing of whether AI is seen as "here to take our jobs" or "here to free us by removing the more monotonous parts of our work".

Get people on board with adopting AI and help them look forward to the tangible benefits it will bring them. It is important for the management to have an understanding of how to incorporate AI in their processes, and to have a level of enthusiasm regarding the same. If this is not achieved it can have a trickle-down effect throughout all levels and results in resistance and mistrust.

Education is the key to adoption and there are steps you can take to ensure that this is done pragmatically and thoughtfully.

7. Integrate domain experts with the data science team.

Any long-running/ legacy system has a lot of quirks in the way it collects and processes data, and this domain knowledge can have an outsized impact on the success of the project.

A prerequisite of deploying an AI system requires a thorough understanding of the domain in which the system will operate. Domain experts provide critical insights that make or break the project.

Domain experts will have to shell out extra time to help your ML team understand the specificities and workarounds of the process and also might have to pitch in by helping formulate the dataset.

They will also make sure that the data collected about the domain can be trusted to be the basis for insights.

8. Communicate when redefining your goals and milestones.

As you progress through the project, you will discover new ways to achieve your goal or maybe even end up modifying the end goal. You might find that something you had planned is not a feasible way to move forward anymore.

Apart from this, any changes to the pre-existing processes are going to disrupt the usual flow of things and you will have to account for the same.

It is essential to communicate all these changes with the stakeholders.

9. Defining metrics that measure the success/failure of the project.

Keeping all the stakeholders involved is very important with respect to the direction the project is taking. You can communicate the progress made much more easily if you have a good metric that all stakeholders understand to quantify your success

10. Identifying where AI will fit in your organization.

Different companies use different data science team structures; some of them have a data scientist for every department and some have a unified data science department whilst others have some hybrid of the two.

Something to consider will be at what level AI will augment your processes and which departments will benefit the most with the introduction, resulting in high impact with relatively low effort.

Eventually, you will find out what works for you but it is important to have the end goal in sight.

11. Having a long term vision.

It is very important to have a vision of what the future state of the company looks like, with all the abovementioned points combined. The steps you will take to get there might vary, and it is extremely probable that so will the vision.

There will definitely be modifications as you go along and realize what is necessary for your organization, but as long as you have an end goal in sight, it will prevent you from going around in circles.

Simply following these 11 steps will help you avoid many pitfalls when first introducing AI. While all these steps might sound simple, don’t confuse simple with easy.

In future articles, I plan on elaborating on this high-level overview. To stay updated you can subscribe to my newsletter You can also get in touch with me on twitter @akshanarayan.