Here is an article by one of our Top 5 Budding Data Scientists, Nitesh Srivastava. Read about the scope of Machine Learning in Implementations Risk Assessments.
Objective:
Gone are those days when there used to be an on-premise solution and their implementations used to be even lazier. In the current fast-growing and fast going technological world, if you wait till these kinds of slow processes add value to your business, you will be far behind…far far behind. This is era of cloud where the project’s lifecycle has been drastically changed from 1 year to 3 months. The widely adopted Agile methodology over the waterfall model again proves that nowadays business wants IT to start contributing without any delay. The below image tells how cloud added value to business:
No gain comes without Risk:
The cloud implementations are no doubt fast, secure, not so complex (in vanilla format) but there may be many risks attached to it. For e.g. will the project overshoot the budget? scope will go beyond agreed points, timeline will be breached, implementing the team’s competency, project complexities. In a cloud implementation these points must be thought over within the ideal time of 3 months of implementation. Sometimes we cannot measure the risks properly resulting to project failure and CSAT (Customer Satisfaction) score least.
Artificial Intelligence can help:
Now is the era of digitization. It is a time when we give most of the supporting work to machines. Machines have different capabilities to learn with historical data and predict things. Machine Learning can train our system based on feedback and history data and help us to measure the risks with project implementations.
How to start:
To get predictions from machines we need to train it i.e. feed data in and for training we must have previous data and feedback from relevant stakeholders. The people contributing to this are the main end-users of this predictive model. The Pre-Sales Team, Customer Success Team, Account Managers like teams who meet with customers and implementation teams to understand the depth and risks associated with project implementations. Based on their previous experience data can be collected and can be used for feeding in Machines.
How the data looks like:
There are many aspects which can affect project success and cause damage too if not averted in the early stage. Few can be as below:
1. Product maturity: How much a product is stable and mature to fulfill customer’s needs is the key to success. There may be cases where no products are involved (which usually don’t happen in Cloud Technologies), in this case, this feature can be ignored. It can be measures based on ‘High, ‘Medium’ and ‘Low’.
2. Implementation Complexity: How complex the implementation is? Has the customer asked for stars from the sky as that’s their right? This complexity is vital to be measured and discussed as per need. This can be measured like ‘Very High’, ’High’, ‘Medium’, ’Low’, ‘Vanilla’.
3. Implementation Team’s competency: How good the team is when it comes to technical and functional aspects of implementation, is the key question to be asked. It can be measured as rating as numbers between 0-5.
4. Stakeholders Engagement Level: If all the stakeholders for the implementation are engaged actively, it reduces the risk a lot as many things can be sorted by discussion within. Can be measured as “High”, “Medium”, “Low”.
5. # of technologies involved: It is important to understand that how many technologies, integrations, systems are involved in the implementation
The final output variable can be like “Risk Associated” as Yes and No. There may be many factors that can be considered to this Risk measurement prediction.
Next Step:
Based on the data provided, the EDA can help us to get insights and action point which can be taken as measurements to prevent any risks. For example:
1. Which factor is highly affecting the project risks?
2. Which factors are not so much to care about?
3. How factors are related to each other? And many more insights can be seen.
Model Training and Predictive Modeling:
By looking at the nature of problem statement, it is classification problem where based on data model need to predict “Yes” or “No” whether Risk associated with implementation. The inputs will be given as factors mentioned above and the model will predict the risk in binary i.e. Yes or No.
Model Evaluation:
The most important stage of any Machine Learning is to evaluate how the model is performing as to take business or strategic decisions one should be very much confident about the outcome of it. The evaluation metrics can be as below:
– TP When model predicted Risk and it had Risk
– FN When model predicted No Risk, but it had Risk
– FP When model predicted Risk but there was no Risk
– TN When model predicted No Risk and there are no Risk
Acceptance Criteria:
As it can be seen from above image to fine tune the model we need work towards minimizing FN because the stakeholder cannot afford to have Risk and it is not predicted. So, we can accept the model which can give some reduced FN value i.e. resulting giving high Recall value and good precision value.
Conclusion
The above approach of predictive modeling for Project Risk measurement can be used in any kind of projects and across verticals like Manufacture, Healthcare, Education etc. only the factors affecting can be reworked as per need. This will defiantly help business to cut down on time and budget which is caused by unwanted reason like budget and scope creep, non-competent team or higher expectations from deliverable.
1 comment