Hazelcast To Unlock Core Innovation,
While Saving $68K
While Saving $68K
Data and AI
Hazelcast
Hazelcast Platform uniquely combines a distributed compute engine and a fast data store in one runtime. It offers unmatched performance, resilience and scale for real-time and AI-driven applications. Over 50% of the world's largest banks trust them.
- ~130Company size
- 10Major sites worldwide
- 12Number of timezones
Summary of success
BrightRide has identified opportunity to remove 20,000 hours of waiting for its most advanced engineers, while saving $68,400 (over 400% ROI).
Challenge
In the time of AI revolution staying nimble and innovate is imperative. Economic downturn makes it even harder and forces everyone to save. Are both possible at the same time?
Hazelcast leverages workforce from all around the globe: USA, Turkey, Poland, India, Netherlands are the major locations. That helps them to get access to talent as well as optimize the cost. But does it stand in the way of innovation and speed, and how to unlock more. This is the question we had to answer with BrightRide.
In the time of AI revolution staying nimble and innovate is imperative. Economic downturn makes it even harder and forces everyone to save. Are both possible at the same time?
Hazelcast leverages workforce from all around the globe: USA, Turkey, Poland, India, Netherlands are the major locations. That helps them to get access to talent as well as optimize the cost. But does it stand in the way of innovation and speed, and how to unlock more. This is the question we had to answer with BrightRide.
Solution
BrightRide team collected Hazelcast's public communication data for a year (between the middle of 2023 and middle of 2024). Data included Hazelcast Community Slack and publicly available GitHub used to track and discuss issues and changes in functionality. We analyzed interactions between teams, calculated waiting times and visualized in a way that works best to highlight the improvement opportunities.
BrightRide app provides data adapters to automatically collect collaboration data. Data is then stored in BR Universal Format. Universal Format allows BrightRide to uniformly process peoples' interactions from diverse sources and to support any additional collaboration tools in the future. Adapters work via REST API and OAuth, and it took 1 week to collect all data while respecting API rate limits of Slack and GitHub platforms.
Organization structure of Hazelcast was obtained by looking up Hazelcast members on LinkedIn and titles of authors at Hazelcast Blog. We also used the following diagram from Hazelcast website to understand relation of modules and teams responsible for them,
BrightRide team collected Hazelcast's public communication data for a year (between the middle of 2023 and middle of 2024). Data included Hazelcast Community Slack and publicly available GitHub used to track and discuss issues and changes in functionality. We analyzed interactions between teams, calculated waiting times and visualized in a way that works best to highlight the improvement opportunities.
BrightRide app provides data adapters to automatically collect collaboration data. Data is then stored in BR Universal Format. Universal Format allows BrightRide to uniformly process peoples' interactions from diverse sources and to support any additional collaboration tools in the future. Adapters work via REST API and OAuth, and it took 1 week to collect all data while respecting API rate limits of Slack and GitHub platforms.
Organization structure of Hazelcast was obtained by looking up Hazelcast members on LinkedIn and titles of authors at Hazelcast Blog. We also used the following diagram from Hazelcast website to understand relation of modules and teams responsible for them,
Other non-@hazelcast.com accounts, after verifying messages on Slack, were categorized as customers. It took 2 weeks to collect and clean this data, which could be saved had Hazelcast provided a CSV file.
In the end, we obtained the following matrix of org units arranged by functions (vertically) and management layers (horizontally),
In the end, we obtained the following matrix of org units arranged by functions (vertically) and management layers (horizontally),
Hazelcast Communication Traffic Map
Waiting times are shown as follows,
Some numbers look unrealistically high, like 28987 hours waiting on the OpenMRS Clients (API) team. Let's dig deeper, 28987 hours for a year between mid-2023 to mid-2024 is 0.6 hours/day for each of the 130 members. Also these times are cumulative — they include transitive waiting from all upstream members. This is exactly why we see a huge opportunity here as small change here can have a drastic effect for productivity!
Widths of arrows represent the amount of communication along the channel.
For functions and managerial layers you can see the temporal diameter—average duration for message to go along the path from any one member to any other.
We used only publicly available sources to prove value and demonstrate how BrightRide works, hence the picture presented here is incomplete. We didn't have access to Hazelcast Jira and internal Slack workspace if there is one. Even with incomplete data we see that,
✅ the leadership team doesn't block the organization,
✅ there is no significant waiting by customer accounts, yet
🔴 there are other 4 units that create most waiting in the system.
Those blocking units are "Core Platform," "Jet Stream," "Clients/API" and "Tech Writing." And waiting mostly happens along the horizontal connections between those 4 units. This is especially manifest for the Core Platform team and must be affecting its performance. This team's inquiries and requests should 100% be prioritized!
Following actions are also to be explored to reduce waiting,
In addition to that, the Clients/API team creates 28987 hours of waiting for others while waits for 10257 hours. Such imbalance is likely result of either the team being understaffed, or low quality of product requirements, or lack of documentation causing a lot of questions from other teams about team's functionality. This should be addressed.
- From Left — from other org units in selected unit's business line or management layer,
- To Right - org unit's own waiting on other units in its business line or management layer,
- From and To the Top — communication with unit's management and control units.
- From and To the Bottom - communication with unit's subordinate units.
Some numbers look unrealistically high, like 28987 hours waiting on the OpenMRS Clients (API) team. Let's dig deeper, 28987 hours for a year between mid-2023 to mid-2024 is 0.6 hours/day for each of the 130 members. Also these times are cumulative — they include transitive waiting from all upstream members. This is exactly why we see a huge opportunity here as small change here can have a drastic effect for productivity!
Widths of arrows represent the amount of communication along the channel.
For functions and managerial layers you can see the temporal diameter—average duration for message to go along the path from any one member to any other.
We used only publicly available sources to prove value and demonstrate how BrightRide works, hence the picture presented here is incomplete. We didn't have access to Hazelcast Jira and internal Slack workspace if there is one. Even with incomplete data we see that,
✅ the leadership team doesn't block the organization,
✅ there is no significant waiting by customer accounts, yet
🔴 there are other 4 units that create most waiting in the system.
Those blocking units are "Core Platform," "Jet Stream," "Clients/API" and "Tech Writing." And waiting mostly happens along the horizontal connections between those 4 units. This is especially manifest for the Core Platform team and must be affecting its performance. This team's inquiries and requests should 100% be prioritized!
Following actions are also to be explored to reduce waiting,
- Inform the teams about impact of their response times;
- Label messages from those teams as high priority, so they don't wait so much to respond. BrightRide supports such automatic labeling for Slack, it can be enabled later;
- Clearly define boundaries (ownership, contracts and interfaces) between teams. Read more about ownership assignment in our blog post and this one on Conway's law;
- Use those boundaries when planning releases to minimize required communication.
In addition to that, the Clients/API team creates 28987 hours of waiting for others while waits for 10257 hours. Such imbalance is likely result of either the team being understaffed, or low quality of product requirements, or lack of documentation causing a lot of questions from other teams about team's functionality. This should be addressed.
Customer Success
BrightRide has successfully identified constraints to Hazelcast performance and clear steps to remove them. We estimate that implementing recommended steps could reduce waiting of Hazelcast's most advanced engineers by 20,000 hours, as well as to unlock faster innovation in its Core component vital in the age of AI revolution.
Not all of the 20,000 saved hours directly convert into cost saving as people switch to other tasks when blocked. Sometimes wait happens overnight or through holidays. Assuming that people cannot switch between more than 4 different projects and work 8 hours per day for 250 business days in a year, 20,000 hours of waiting convert to 1140 hours of labor which could be saved. At $60/hour rate for a senior software engineer, it amounts to $68,400 of savings or over 400% ROI.
BrightRide has successfully identified constraints to Hazelcast performance and clear steps to remove them. We estimate that implementing recommended steps could reduce waiting of Hazelcast's most advanced engineers by 20,000 hours, as well as to unlock faster innovation in its Core component vital in the age of AI revolution.
Not all of the 20,000 saved hours directly convert into cost saving as people switch to other tasks when blocked. Sometimes wait happens overnight or through holidays. Assuming that people cannot switch between more than 4 different projects and work 8 hours per day for 250 business days in a year, 20,000 hours of waiting convert to 1140 hours of labor which could be saved. At $60/hour rate for a senior software engineer, it amounts to $68,400 of savings or over 400% ROI.
Next Steps
Given that BrightRide analyzed only publicly available data, the next obvious step is to work with Hazelcast team to get access to internal communication, including their Atlassian Jira account for a fuller picture and improvement opportunities.
Analysis to be done again next year, after steps recommended by the BrightRide team are implemented and automatic labeling of messages coming from the most blocking units is enabled. Then actual impact on the organization and further improvements will be clearer.
Given that BrightRide analyzed only publicly available data, the next obvious step is to work with Hazelcast team to get access to internal communication, including their Atlassian Jira account for a fuller picture and improvement opportunities.
Analysis to be done again next year, after steps recommended by the BrightRide team are implemented and automatic labeling of messages coming from the most blocking units is enabled. Then actual impact on the organization and further improvements will be clearer.
Pilot with us?
Our Mission — Unleash creative energy in
today's organizations for what truly matters.
We achieve this by equipping business leaders with tools
to forge efficient communication structures.
today's organizations for what truly matters.
We achieve this by equipping business leaders with tools
to forge efficient communication structures.
contact@brightride.io
© 2024 Bright Ride, Inc. All rights reserved.
© 2024 Bright Ride, Inc. All rights reserved.