Actionable Insights

Data science at work in Pinpoint.

We use machine learning to make meaning out of the millions of data points lost in the development of software.

By bringing together activity data from the systems used to build software and mixing it up with all the interactions you have in Pinpoint every day, we are able to use machine learning to unlock all sorts of actionable insights, make recommendations on how to improve execution, predict risks, and forecast work effort. All of this, saving you from manual effort and gets your team back to coding.

Our Principles

Built transparent.

Machine learning shouldn’t be a black box. Here are some principles we use when creating our models:

Explainability

Explainability

We “show our work” with customized explanations for each metric or prediction.

Personalized

Personalized

When building out new ML features, we want them to be contextual to the user viewing it.

Constantly Learning and Improving

Constantly Learning and Improving

As we get more data about how engineering teams work and build more models that rely on each other, the accuracy of our models and ultimately our product improves.

Global and Custom Data Models

Global and Custom Data Models

We have two data models to ensure there is enough data to train the models for the highest accuracy of a prediction.

How AI is built into Pinpoint

Issue Forecast

Predicts the time an issue will take to complete.Expand

Estimating how long an issue will take is crucial to properly setting expectations and delivering products on time. Current methods such as story points are ambiguous and unreliable. Our Issue Forecast predicts the time an issue will take to complete; capturing the entire lifecycle from first in progress to final closed date.

To the extent that epics represent projects and are more than just a collection of issues, it is useful to estimate when epics will be completed. The model we’ve built for this purpose starts with the Issue Forecast for each issue in the epic. Using the Monte Carlo method, we then simulate an array of outcomes for the epic, based on the confidence levels of each child issue’s forecast. The collective outcomes then provide our estimates for the range of completion dates for the epic.

The “epic card” includes a small blue graph detailing the probability of completing an epic vs. date of completion, along with other information relevant to the epic. Notably, the chance of finishing an epic increases more slowly as you approach 100% – even at Pinpoint we have trouble predicting the future with absolute certainty.

The state of a sprint can change dramatically as work progresses. To help get a handle on how a sprint is faring, Sprint Risk represents the likelihood of finishing all issues in a sprint by the end of the sprint. We use the Issue Forecasts for each issue to estimate, on a user-by-user basis, how confident we are that all issues will be completed. This allows us to see if some users are overloaded, and thus the sprint could benefit from redistributing work, or if the team as a whole may have trouble. To help visualize whether the sprint is on track, we provide a visualization of the current progress on a sprint, compared against historical completion rates, broken down by day of week – in this way it’s easier to see how the remaining work relates to the time left while considering that some days just aren’t as productive as others.

Sprint Health helps teams track and improve their sprint planning and efficiency. To derive Sprint Health, we examine a sprint’s work at three points: the initial plan, the final plan and what is actually delivered.

Teams can spend hours planning out work every week or two. This sucks up a lot of valuable time away from building the software. Pinpoint uses information such as what your teams work on, what should be prioritized, what was unfinished from the previous cycle, and who has the most experience with certain tasks to make recommendations about your plan. Pinpoint will auto-fill your sprint with items that are aligned with your product priorities and auto-assign issues to the people best suited to handle those tickets.

We bring all the necessary context such as PR and issue data together in one place to make stand-ups more efficient. Our AI uses team activity to generate predictions for capacity issues that will impact your team’s ability to deliver and identifies blockers and dependencies that typically go unnoticed.

Teams struggle with capturing and following up on action items that are identified during a retrospective. Because Pinpoint knows what you’ve worked on and has the historical context of the team, Pinpoint will baseline your sprints and tell you how you're doing against your average performance and tell you why you did above- or below-average so that you can improve in the next iteration.

A personalized activity feed to view and collaborate on the activity related to your work. This model creates an interest score on activity that is built off the past volume of interactions and types of interactions a user has had with a specific entity (another user, epic, project, or repo).