What next? – Now, after the prediction has been created, the scorecard has to be reviewed. The scorecard is a place one may go to find prediction performance and to identify any setup or modelling issues. The scorecard also tells that if the performance of the prediction is too high, when it’s too high, it means that most likely it is set as hindsight bias and have some potential leaks. Removing these leakers, will make the prediction work as the stakeholders want it to. Also, if the performance of the prediction is too low, it hints that there is not enough relevant data for the prediction to work appropriately. In this case, more relevant data need to be added.
Asking the business experts may help in identifying the information they need to make these decisions. Artificial Intelligence helps by trying to think like humans, so it is good if a human analysis making can help it make better decisions. To increase the prediction quality, there are various choices which can be determined to have an added value for it to work better. Some examples are – if this is a red account, % change in number of cases, number of severity 1 cases, customer success manager and solution engineer sentiment or assessment score, Account Health Score, Account Tier, average NPS score, lead product, and much more. Cloning previous prediction is an option which makes a copy of the earlier prediction and holds all the setup. With a little adjustments, a better version can be created with ease.
Also, one might go to the details of the scorecard to see the top predictions and have the stakeholders check if those predictions make sense for them from a business perspective. The predictions may seem very obvious sometimes, not to get discouraged as it is only the confirmation that the Prediction Builder is following the right patterns for the business.
When the stakeholders are happy with the quality of the predictions, it can be enabled to get the scores. To find out the predicted values, add the prediction field or opportunity score to the list view and relevant page layouts for the sales reps to see them. After a few weeks or few months, real life data will be there and it will be known which opportunity laded up being closed won and which of them landed in closed lost. Then the comparison between prediction results and the actual results can help in determining how the prediction is working for real life data. If needed, the prediction can be revisited for betterment.
How to determine if the built AI project worked and succeeded? – This is where the original goals and KPIs can be revisited. It can be compared if the goals are achieved, the set revenue targets were met and if this scoring helped in adding the values to the company.