Sunday, November 27, 2022
HomeBig DataSunstone: Knowledge-Powered Buyer Success - The Databricks Weblog

Sunstone: Knowledge-Powered Buyer Success – The Databricks Weblog

Bringing Knowledge+AI to Buyer Success

At Databricks, we need to carry knowledge and AI to all components of our group. This 12 months, the Buyer Success group has partnered with our in-house knowledge scientists to create data-powered instruments that improve the best way we information our clients on their journey to the Lakehouse. Considered one of these initiatives is Sunstone, our inner suggestion platform.

Sunstone permits us to know how a buyer is leveraging the Databricks platform and customise suggestions amongst 4 totally different areas:

  1. Knowledge engineering
  2. Machine studying
  3. Knowledge evaluation
  4. Platform administration

On this weblog, we are going to discuss in regards to the historical past and motivation for Sunstone, how we construct our suggestions, and the way we have used it to make data-powered impacts for a lot of of our clients. We are going to conclude with a quick part on our future plans for the mission. We hope this might be an inspiration for different data-driven buyer success groups trying to implement knowledge merchandise for their very own clients.

Why we created Sunstone

Beforehand, serving to our clients undertake greatest practices required a number of cycles of knowledge gathering and discovery periods. From these discovery periods, we might have interaction with people to be taught what function they’re in, what options they’re utilizing, and the way they’re implementing their use instances. We relied on these engagements to glean details about the place we might successfully assist clients. It took on common 2-4 weeks to finish the cycle from scheduling a session to the shopper implementing the suggestions. We wanted to discover a strategy to velocity up the engagement and widen the breadth of suggestions and floor insights by way of telemetry.

Figure 1. Previous engagements required customer facing teams to conduct multiple discovery sessions.
Determine 1. Earlier engagements required buyer dealing with groups to conduct a number of discovery periods.

We began with creating a rule that helped us determine the workloads operating on older runtimes. It is necessary for our clients to remain present with a purpose to unlock the frequent enhancements we’re making to the Databricks runtime. Our knowledge scientists used our inner Lakehouse platform to instrument the characteristic wanted to gasoline the advice. We then materialized this suggestion on inner dashboards, permitting our buyer dealing with groups to present this suggestion with their very own clients. We have since expanded our suite to incorporate many different suggestions, extending this functionality to our different core teams of customers together with, Knowledge Engineers and Knowledge Scientists, with suggestions on Delta, Structured Streaming, ML Movement, and extra. Sunstone has positively impacted our clients as they’re implementing their use instances and we are going to discover extra on that within the subsequent part.

Large advantages for our clients at scale

By constructing a suggestion platform, we’re in a position to serve our clients a constant technique of making certain they’re getting essentially the most worth out of the Databricks platform. Previously, it will have taken a number of discovery periods to be taught that the shopper did not have IP entry lists configured to limit entry to workspaces or weren’t utilizing cluster insurance policies to implement cluster creation constraints throughout their group. Now with Sunstone, we are able to generate this diagnostic earlier than we even meet our clients for the primary time. This strategy has enabled the Buyer Success crew to onboard a buyer quicker, in addition to allow our clients to develop their use instances quicker, resulting in an earlier and extra environment friendly go-live.

Figure 2. Sunstone-powered engagement enables customers to onboard faster and develop their use cases better.
Determine 2. Sunstone-powered engagement permits clients to onboard quicker and develop their use instances higher.

Let’s convert knowledge into actions

On this part, we are going to talk about our learnings on working with telemetry knowledge and influencing human habits. Telemetry knowledge in its uncooked type is just not actionable. As a way to render the information to our function, we take the telemetry knowledge and mix in greatest practices to create a human readable rating. We wished the rating to be simply interpreted pretty much as good or dangerous, so we designed the rating to be on a scale of 0 to 100 with 0 being least optimum and 100 being essentially the most optimum. A great rating (increased than 80) is a constructive reinforcement that reveals the shopper is leveraging all of the really useful options. A much less optimum rating (decrease than 50) prompts the shopper to research what’s inflicting it and opens the door to implementing adjustments.

Figure 3. Combining telemetry with best practices is the winning formula for influencing behavior.
Determine 3. Combining telemetry with greatest practices is the profitable system for influencing habits.

Our greatest practices are compiled from years of trial and error from buyer dealing with groups interacting with clients. They’re knowledgeable, clever, and much from arbitrary. We take greatest practices from the shopper groups and refine them additional with Product Supervisor and Product Specialist enter. Once we are creating a brand new suggestion, it isn’t unusual that we see this rule growth section debated over many iterations.

Dangerous Suggestion Good Suggestion
Cluster 1 = Runtime 7.3
Cluster 2 = Runtime 9.1
Cluster 3 = Runtime 10.4
Cluster 4 = Runtime 11.2

Really helpful Motion: Hold your clusters up to date.

Your rating is 75/100. DBR 7.3 is coming as much as the top of help whereas 11.3 permits Unity Catalog and quite a few different enhancements to the Databricks platform.

Really helpful Motion: Improve Cluster 1 to Runtime 11.3 LTS.

Determine 4. The distinction between a nasty suggestion and an excellent suggestion is being particular and prescriptive.

It is necessary to supply suggestions solely when an motion could be taken on account of it. A nasty suggestion states the information and presents little path on what to do with the data. A great suggestion provides you with an concept of the severity of the issue and intuitively communicates what it’s good to do to extend the rating.

Moreover, suggestions needs to be bucketed and filtered primarily based on whether or not it’s related to them. A knowledge engineer could not care about Audit Logging so we do not give them that suggestion. As a substitute, we give the information engineer suggestions on utilizing Vacuum when creating tables of their knowledge pipelines.

Position Really helpful Motion Advantages
All The shopper ought to be utilizing the newest Runtime model. Customers may have entry to the newest options and safety updates.
Platform Administration The shopper ought to activate Audit Logging on all of their workspaces. Clients will be capable of audit person exercise on their workspace.
Platform Administration The shopper ought to be making Secrets and techniques API calls to all of their workspaces within the final 28 days. Customers can conceal delicate credentials like passwords in notebooks.
Knowledge Engineer The shopper ought to be operating Vacuum instructions on their tables. Use Delta Vacuum to take away outdated recordsdata to not incur pointless cloud storage prices.

Determine 5. Really helpful actions are given to related roles and advantages are clearly articulated.

How a buyer just lately benefited from Sunstone

In working with one in all our clients, our Buyer Success Engineers consulted Sunstone to determine a set of suggestions. They realized instantly that there have been a number of necessary options that they may advocate. These have been:

  1. Secrets and techniques: The workspaces listed right here have jobs and different workflows that aren’t using the secrets and techniques API, which prevents the publicity of delicate credential info in Databricks notebooks and jobs. Utilizing the secrets and techniques API is a straightforward approach to make sure you’re securely using credential and related info in your Databricks workflows.
  2. Audit Logging: This rating measures the implementation of the audit logging characteristic throughout workspaces. Audit logging is a service clients can activate which sends low-latency logs in JSON format for any workspace through which it has been configured. Each quarter-hour, Databricks will pipe buyer workspace-level metrics to a desired cloud storage location. They comprise a wealthy schema of knowledge together with particulars on accounts, secrets and techniques, dbfs, secrets and techniques, and extra.

Of their common syncs with the shopper, they addressed our suggestions. Over the following few months, the purchasers’ use of each Secrets and techniques and Audit Logs hit the highest of the mark.

Figure 6. Customer achieves maximum score after following recommendations generated by Sunstone.
Determine 6. Buyer achieves most rating after following suggestions
generated by Sunstone.

By implementing these options, the purchasers elevated their safety and decreased their compliance threat.

What’s subsequent?

In conclusion, Sunstone has enabled our buyer groups to raised perceive and serve their clients with a diagnostic that intelligently qualifies and offers actionable suggestions. We’re constructing data-powered instruments to make it simpler for our clients to achieve success on Databricks. On the horizon is a transfer away from our present self-serve mannequin to 1 that immediately makes suggestions to clients.

Sunstone has been adopted broadly throughout the Buyer Success group. It’s leveraged each day to supply over 100,000 actionable suggestions thus far this 12 months. On common, we’ve got additionally seen a 60% discount in ramp up time for our clients.

In case you are fascinated with constructing or leveraging instruments like Sunstone, we’re hiring! In case you are fascinated with how the Databricks platform can allow your use instances, please e-mail us at [email protected].


We want to thank our knowledge heroes Francois Callewaert and Catherine Ta from the Knowledge Science crew for creating Sunstone with us and accommodating Buyer Success’ many characteristic requests. Additionally, thanks to Ravi Dharnikota and Francois Callewaert for serving to overview this weblog publish. Lastly, thanks Manish Bharti for offering an awesome case examine to stroll by way of!



Please enter your comment!
Please enter your name here

Most Popular

Recent Comments