The strategy outlined below gives step by step instructions on how to train the algorithm. We found that this strategy is the most effective as time invested vs. recognition improvement.


PART 1 | Determining the narrative quality

Before training the algorithm, it is helpful to get a sense of the narrative quality of certain practice groups or timekeeper levels. This makes the training of the algorithm more detailed and focused. Here is how you can find the narrative quality of certain groups within the firm.


Step 1: Go to the Drilldown

Step 2: Partition by Practice Group

Step 3: Scroll down and click on Visualization

Step 4: Select the Comparison list

Step 5: Click on Compare and select Narrative quality

Step 6: The visualization displays the narrative quality of each practice group (see screenshot below)


You can make a similar overview for each office (replace Practice Group in step 2 with Office) or Time Keeper Level (replace Practice Group in step 2 with Time Keeper Level).



What can I do with this overview?

The practice group with the most red (in the example above, it's the practice group Notaries), has the most uncategorizable entries. As a start, you can focus on this practice group to train the algorithm.


PART 2 | Training the algorithm

You can now start identifying and reporting narratives to train the algorithm. 


Step 1: Go to the Drilldown

Step 2: Partition by Narrative Quality

Step 3: Click on Selection and add a filter of the practice group, office or level you have identified in part 1

Step 4: Add a time limit filter, for faster navigation (for example, Year: 2016) 



Step 5: In the donut, click on the red part to filter out the uncategorizable section

Step 6: Scroll down and click on Visualization. Select Time Cards as the visualization



Step 7: Order descending by recorded hours by clicking on that column's title



Step 8: You can now start identifying narratives that should have been recognized or are incorrectly classified

Step 9: The categorization buttons have different colors, each one explained in the screenshot below



Step 10: Create a list of terms that are uncategorizable in Clocktimizer and note the key term of how the narrative should be classified

Step 11: Also have a look at the orange Vague section where Clocktimizer identified general activities. You might find more specific activities in there that are not yet categorized

Step 11: You can send us the list and we will review and update the algorithm based on your feedback!