C4C Machine learning -Deal intelligence- (Opportunity & Machine learning)
This document shall help to understand the general Features and Functions of the Machine learning Scenario ‘Deal intelligence’ as well as the necessary steps to implement the same.
Machine learning / Deal intelligence 1802
1. Description of features and functions of Deal intelligence
a. General overview
Sales Managers are going blind on at risk opportunities
Only 40% of forecasted opportunities actually close
High volume of low-propensity opportunities leading to inflated pipeline
The deal intelligence feature is a feature in the C4C Opportunity management which automatically scores deals and ranks them based on highest propensity to close.
The ranking should help to:
– Close deals faster
– Increase win rates
– Improve the pipeline
– Better prioritization of high propensity opportunities
– Better backfill and planning for at risk opportunities
– Sales acceleration and predictable revenue/deal flow
How does it work?
Historical data is used to train the machine learning model. As a result, we get a customer specific predictive model which is applied to provide end users a scoring of opportunities. The higher the score, the better the chance of winning the opportunity.
Historic data is taken from existing opportunities in C4C
b. Detailed features & functions
i. End user experience
The end user will get a new pane in the opportunity workcenter which gives an overview on the status of each opportunity. Also a new column ‘score’ is introduced which shows the score for each opportunity in the opportunity list. The score field can also be used as filter criteria to select certain opportunities.
The score and the additional info in the side pane is updated once a day based on the changes made to all opportunities in the system. This update is triggered at midnight (timezone of your datacenter)
Can I see the time of the last update? Currently not (1802), this feature is planned in a future release
Can I manually trigger an update myself? This is currently not possible.
After the first activation of a model- which opportunities will get a score? All open/in progress opportunities will be scored. Opportunities created after the activation will be scored with the next update at midnight.
Which information shown in the sidepane is predicted/calculated by the machine learning feature? Only the opportunity score is predicted. All other information in the sidepane is data collected based on the opportunity and shown in addition for a quick overview.
Whether an end-user can see the side pane and the score column depends on the assignment of the corresponding authorization role to the user which allows the user to see the side pane.
2. Prerequisites for the implementation
a. Customer must have a C4C Enterprise License
b. Data volume
The more historical data the better.
To create the prediction model, the opportunities of the last 12 month are used.
The absolute minimum needed to create the prediction model is 5000 opportunities. The number is heavily dependent on the factors listed next.
c. Data history
The model calculates the chances of an opportunity to be won or lost based on historical data and the evolvement of the opportunity over the course of time. The better these changes are recorded, the better the prediction. For example changes in the sales phase, changes in the probability, attached activities etc…
d. Data quality
The prediction model identifies key fields in an opportunity which are considered for the prediction. These fields should consistently be filled (ideally, they are mandatory from the beginning)
Close date, probability, sales cycle/sales phase, status
A full list of relevant fields will be provided soon.
e. Data consistency
In order to have an accurate prediction it is important that opportunitiy fields are consistantly filled. Example: if you have 10000 opportunities and for half of them the field ‘probability’ is not filled, the accuracy of the model decreases.
f. Data balance
The data balance plays a significant role. If for example sales reps have only created data for won opportunities in the system, the prediction model will not be accurate. Same for example if won opportunities were not set to ‘won/lost’ but remained open even though they were won or lost.
The absolute minimum is 1000 opportunities with status won and 1000 opportunities with status ‘lost’ in order to get a good prediction.
g. Customization/Custom development
Heavy customization can influence the prediction model. The customization of an opportunity has to be checked individually if it contradicts the machine learning feature.
Prediction model does not consider extension fields!
Are extension fields supported in the future in the prediction model? This is on the future roadmap.
In general the implementation can be divided into the following phases:
a. Precheck data
Check customer data with regards to topic 2 a-g
Currently these checks are to be done manually, in the future special check tools should support this step.
b. Activate/configure deal intelligence
i. Tenant decision
Before activating the Deal intelligence feature it needs to be decided in which tenant the first implementation is done.
The recommendation is to activate it directly in the production tenant since there is usually the most accurate data. Since the feature is only a ‘read’ feature – there is no risk of harming production data. Also, the feature can only be made available to some test users at the beginning.
Usually there is no sufficient data in the test tenant, so it makes no sense to create the prediction model for testing there.
Copy of production tenant
Another option is to test the feature in a copy of the production tenant with the ‘productive’ data records.
additional tenant required, depending on your license this might lead to additional cost.
Testing of the dynamics in the model is not as good as in the production tenant since there is no constant change of data.
Model has to be recreated in production and might lead to other results than in test due to different data after some time
The deal intelligence feature has to be activated by SAP.
An incident has to be created in the tenant where it should be activated.
Please request the activation of the Machine Learning scenario
In a future release the feature can be switched on via scoping.
In order to assign the authorization to see the new side pane / deal intelligence related data in the opportunity workcenter a new business role has to be created.
The business role has to be created by the customer/implementation partner and it must be specified in the incident to request the activation of deal intelligence by SAP.
SAP will enhance this role in order to show the new features on the UI.
The role will then have to be assigned to the relevant users by the customer.
This proceedure will be replaced in a future release and the authorization can be granded by the Administrator just like any other authorization in C4C.
c. Train machine learning model
Before you can make use of the deal intelligence feature, you have to create your customer specific trained model which is used to predict the scores for your opportunity. To train your custom model go to the admin workcenter and choose the view ‘prediction services’ ‘configure’
The creation of the model is done in simple steps
- Choose opportunity scoring
- Add a new model
- as a result you see a new model with your chosen description
- Mark your new model and choose ‘train’
- As a result the historical opportunity data is collected and sent to the Leonardo machine learning component and your customer specific model is trained/created based on your historic data.
- This may take some time- wait until the status is ‘finished’
d. Validate model training results
As a result of the model training the SAP Leonardo component returns an accuracy which tells you how good your model is.
As a rule of thumb a model with an accuracy below 50 is not good enough to have a satisfying prediction result. But this must then be checked in the individual Project.
It is necessary to activate the model once the validation has been done and the accuracy is satisfying.
After the model is activated the open opportunities get a score. The first scoring happens over night after the first activation of the model.
As a first test you can validate if the score matches your own expectation for the individual opportunity.
As a second step you can change opportunities and see if the score update ( which happens only over night) matches your expectation. So currently it is not possible to Change an opportunity and see the effect on the score prediction right away,
f. Go live
Once testing is complete you can make the feature available for all users.
In case you started in the productive tenant- just assign the business role for deal intelligence which was created before to all users who should use the Feature.
In case you started in a test tenant: Repeat steps described in chapter 3 in your production tenant.
4. Productive usage
If required, you can retrain your model by creating a new model and activating it.
When/Why would I create a new model?
A retraining of the model could be necessary if your opportunity data history changes over the course of time for example:
– After uploading new opportunities caused by a data migration
– Enrolling new processes for handling opportunities in your company
– Different user behavior
Will the activation of a new model deactivate the old one?
Yes, only one model can be active at one point in time.
Can I reactivate an older model? Yes
Thanks Ralf, great summary. Thanks for sharing these experiences!
Ralf Kammerer I have two questions for you.
Thanks in advance!
Thank you for a great Summary.
Could you please let me know, for Lead scoring what should be the Status for Lead data?
Awaiting your response.