[VSD] - Adding a new metric - "AI (Code Suggestions) Usage Rate"
Problem
Engineering leadership (Director/VP/CTO) need to demonstrate the ROI from the investments in AI features. To quantify the impact of adding AI and to demonstrate the ROI, they need:
- To understand which metrics improved as a result of this investments in AI.
Proposal
- In Log First Daily Use of Code Suggestions by User (#441485 - closed) groupcode creation will
record the first time during a day, a user receives a suggestion
. -
groupoptimize will aggregate this record into
Monthly Unique Code Suggestions users
. - The
Monthly Unique Code Suggestions users
will be stored in PS, and can be later cloned into CH. - Using the new
Monthly Unique Code Suggestions users
calculate a new metric - "Monthly Code Suggestions Usage rate"Monthly Unique Code Suggestions users
/Total Monthly unique contributors
). - Results will be generated for each group, sub-group and project.
- Next iteration will be UX: [VSD] "AI Impact view" MVC - visualize the ... (#443698 - closed) (frontend)
Code Suggestions unique users Usage rate
vs Acceptance Rate %
For the MVC, we believe that comparing the Code Suggestions usage rate
month-over-month will give a more accurate indication of AI Impact than Acceptance Rate %
, since Acceptance Rate %
will be easily impacted by:
- The experience level of the developer, e.g. Senior eng will have a low acceptance rate, while joiners will have high.
- The type of project and the level of complexity will also impact the quality of the suggestion.
- Network latency will also be a factor.
Old proposal
Can we use the Code Suggestions Usage data to enrich the groupoptimize ClickHouse-based contributors table with a new "AI contributor event"?
Meaning that we will track on CH, the number of users that use the code suggestions. Once we have this enrichments, we can calculated the monthly "AI Usage rate" - Monthly Unique Code Suggestions users
/ Total Monthly unique contributors
).
And can we compare this new "AI Usage rate" with other SDLC performance metrcis.