Scaling Executive-Ready Insight With Databricks AI Query on AI/BI

Summary

Databricks' own sales analytics team automated daily pipeline summary generation using AI Query within an AI/BI dashboard, replacing a manual process that previously took approximately 30 minutes per territory and an estimated 36,000 hours annually.
The key principle established in this video is that 80% of the impact comes from clear strategy logic and trusted data, with AI delivering the final 20% by expressing that logic as consistent natural-language summaries at scale across hundreds of sales leaders.
The Pipeline Smart Feed delivers predefined, AI-generated daily summaries to leaders the moment they open the dashboard, while Genie handles conversational follow-up investigation for deeper ad hoc analysis.

Scaling Executive-Ready Insight With Databricks AI Query on AI/BI

Creating executive-ready metric summaries is harder than it looks. Leaders expect insights that are accurate, consistent, action-oriented, and concise, yet many organizations still rely on manual reporting — driving high effort, slow turnaround, and inconsistent interpretation. In this video, we share a real-world example of sales strategy and analytics teams automating daily pipeline summaries using Databricks AI Query within an AI/BI dashboard. Each summary previously took ~30 minutes per territory—an estimated 36,000 manual hours annually. By standardizing business logic, strengthening the Databricks data foundation, and applying AI selectively, we scaled executive-ready insights to hundreds of leaders. The key lesson: 80% of impact comes from clear strategy logic and trusted data — AI delivers the final 20% by generating consistent natural-language summaries at scale. Attendees will learn a practical model to encode executive expectations, govern AI outputs, and reduce time to insight.
Talk By: Kimberly Zee, Business Intelligence Manager, Databricks ; Marina Zhou, Strategy & Execution Manager, Databricks, Inc. ;

Chapters

FAQs

What is Databricks AI Query and how is it used in AI/BI dashboards?

AI Query is a feature within Databricks AI/BI dashboards that generates natural-language summaries of metrics using governed business logic embedded in the dashboard. In this video it is used to produce daily pipeline summaries for sales leaders without requiring manual analyst interpretation.

How did Databricks save 36,000 manual hours annually with AI Query?

Each territory's daily pipeline summary previously required approximately 30 minutes of manual effort to read and interpret multiple dashboards. By automating the summary generation with AI Query across hundreds of territories, the team projected savings of roughly 36,000 manual hours per year.

When should you use AI Query versus Genie in a Databricks AI/BI dashboard?

AI Query is best for predefined, repeatable summaries where business logic is fixed and consistent output is required at scale, while Genie is designed for conversational, exploratory queries where users want to ask ad hoc questions. This video uses both: AI Query for the daily Pipeline Smart Feed and Genie for follow-up investigation across the broader platform.

What is the Pipeline Smart Feed built by Databricks?

The Pipeline Smart Feed is a predefined daily summary delivered to sales leaders that uses AI Query to translate pipeline metrics into natural-language insights the moment they open their dashboard. It answers the two core questions every sales leader needs each morning: whether pipeline generation is sufficient to cover revenue targets and whether open deals can close on time.

Full transcript

[00:09] Hello everyone. Thanks for coming to join the last session of the day. I'm pretty sure you guys are very tired, but I promise this will be a fun session. Uh so by the end of the session, you will learn how to save hundreds or even thousands of hours of your team's time
[00:23] and imagine what you can use that time for otherwise. So I'm Marina Joe from data bricks go to market strategy and execution team focusing on pipeline analytics. Today my colleague Kimberly
[00:36] who will be joining me later and I will talk about how we turn mundane routine dashboard reading insight generation work into highly scalable way using AI saving hundreds or even thousands hours
[00:49] of your team's time. And we'll talk about in three parts. First the challenges of our team and how we solve it using two solution. One is creating a prescriptive insight summary that
[01:02] delivers instantly at your team's fingerprint. We call that pipeline smart feed. And the second way is to use conversational AI genie. You probably heard tons of time already today. And how we use it to scale the insight
[01:14] across our platform to all the system we work with. So before we move on, I'm going to take a quick poll. How many of you feels like there are too many dashboards in your daily work? Raise
[01:27] hand. Oh, almost everyone. Nice.
[01:42] the same uh pain point our leaders used to share. We hear this on repeat all the time. Every morning I need to click through multiple dashboards, five dashboard, 10 dashboards a day in order to figure out what I need to do next. Um
[01:56] to understand the challenges more I'll give a little bit more business context of what we solve. I focus on a daily basis on pipeline analytics. So this pipeline is not data engineer pipeline but sales pipeline. So in case you don't
[02:10] know a pipeline simply means any potential deal a seller is chasing through different stages. It can be close one or close loss at the end. Fundamentally, all our sales leaders want to answer two questions. One, what
[02:24] is pipe gen? Meaning, do we have enough pipeline today to forecast and cover our revenue for tomorrow? And second of all, pipeline execution. So for all the pipeline we have today, can we close
[02:37] them on time? And then if not, what can we do about it? It's very two simple questions on the surface, but the challenge is in order to understand one region, a human need to read tons of
[02:50] data points. I don't know if any of you feel the same, but for us, we have 10 plus pipeline metrics defined by different teams for different usage and all of them are interdependent. You cannot understand one without reading
[03:03] the other. We have 10 plus regions. Every region has their own nuance and we have five and even more dashboards to click click through in order to gather this data points and if you want to do
[03:15] your extra due diligence looking through things period over period that lead that makes your numbers up to 30 plus. So imagine one human staring at all the walls of number and trying to figure out what change what is the insight what
[03:28] action take. It's a lot. And to further quantify that, let's say for one region to write a wellthought through summary after understanding all the metrics changes, that takes about an hour and we
[03:41] have 360 leaders on the field, that's 360 hours per week. Imagine if we use AI and instantly cut the time into zero and our sales leader will be able to use that time um meaningful activities like
[03:54] team coaching or spending more time with customers and we can use AI to just give what sales leaders really want to answer meaning like just tell me how healthy is
[04:06] my business and what should I do next without giving me all the dump of numbers. To do that we use AI and then we want our data bricks AI solution to cover three points. One is instant the second
[04:20] explorable and trusted. For instant what we want is every time the data refreshes your status your next step summary are also automatically updated as well. So we call them our smart feed. We also
[04:33] have Genie where our leaders can explore and follow up uh doubleclick questions. And the third pillar which is the most important piece that we want to cover is how to get trusted data. Um today AI can
[04:46] cover the first two pillar really well. It can give you quick answer and allow you to explore very easily. But the trusted data piece is actually require the most hard work and that's what makes
[04:58] your insight summary you deliver to your executive most valuable. And what we suggest is for all the work you spend actually spend 70% of the time honing your data and strategy foundation right
[05:12] meaning spend the time with your data engineering team align the metrics definition the logic the decision logic with your strategy and operation team everyone work together to get the first 70% correct and only brings AI for the
[05:26] last 30% simply taking the data the insight expressing it in the natural language format that way and scale it at once. Um, if there's any one slide you we'd like you to take home today, it
[05:38] will be this chart. Meaning just spend most of your time with your data and analytics and strategy question first before looking through the latest fanciest AI tool. Because without the
[05:50] first 70% data foundation, your AI will just scale all the mistakes you have. So for us you can give you the wrong data by pulling the wrong region target analyzing the wrong physical quarter
[06:02] because you didn't have the right quarter end definition properly uh given to AI and what's even worse is you can write a perfectly well-ritten summary but the leaders are chasing the wrong
[06:15] owner of the action all of these will cost you time credibility and also the quality of your decision which are everything that we want to avoid. So enough principle and why does the
[06:27] data analytics foundation matter? Uh let me quickly walk you through how our sales leader use our pipeline smart feed every day. So to remind us pipeline smart feed a smart feed is simply a governed insight that is delivered to
[06:42] our leaders fingerprint every day the first thing they wake up and open up the dashboard. So I'm a sales leader. I every time I open up my dashboard, I'll see the smart feed next to my dashboard
[06:54] with all the metrics I see on a daily basis. So, it's a four paragraph smart feed. So, the first is simply just a quick two sentence summary on the health of pipe generation and pep execution.
[07:06] And I know as a sales leader that all these numbers are trusted because these aligns exactly as the numbers are showing in the dashboard as well. And the reason it can do that because underneath AI knows to pull each number
[07:20] from a trusted govern table and from the field that is asked to pull. So there's no guesses from AI. Next, we have the highlights and low light section. Notice that we only have
[07:32] like four numbers shown here instead of the 10 plus metrics that we mentioned that are relevant. That's because we want the leaders to focus on things that are truly matter that are varied significantly above the threshold that
[07:45] are aligned by strategy. So this is where we spend a lot of time working with the field to make sure we come up with the threshold and guidance that take into consideration of the region variance, the time variance and um how
[07:58] each team can be different from each other. So without a wall of number the the sales layer only need to looks at this four metric particularly to pay attention to. Next we have our action item. For each
[08:12] action we have a embedded link that takes the leader directly to the either the direct or the account executive they need to follow up question with and a list of use cases that might be relevant. So again the same idea of
[08:26] everything happen in one place. The leader doesn't need to go to other dashboards to recreate the granular level view that ties up to the number they see. Everything happens in the same dashboard al together.
[08:38] Last is we have a drill down uh use case section to highlight the big mover opportunity that explain the changes of your metric. What's even better here is we embedded AI summary inside for each
[08:51] opportunity. So it reads through all the notes, all the data add that opportunity to explain for example why the use case is lost, why the use case is stalling longer than expected and again
[09:04] everything all at one place the leader will understand what is the pattern that is happening in their book of business. So with a few click and scroll of our smart feed very simple very concise uh
[09:16] our team knows what is the health of the business and what I need to do next. uh underneath that we actually spend a lot of times constructing it to be to make it correct. We have 125 plus
[09:30] decision point pre-coded along with the assessment framework of the threshold the logic altogether send it over to AI. In this case the narrative truly stays analytics and strategy owned and we have
[09:43] no human in the loop to double check every number that are presented in front of our sales leader. And what's even better is all of these are simply part of our data engineers daily workflow. So
[09:55] the AI that runs uh so a data engineer simply just open up a notebook in data bricks call on a SQL function and one of we have many AI functions but one of my favorite is AI query and within AI query
[10:09] you can call the model you use and for all the 125 decision point and assessment framework we talk about construct that into a prompt instantly. AI knows how to write 360 versions of
[10:21] the same summary all at once and saving them into a table process in batch and in parallel. So none of this leaves data bricks platform that means you have the same benefit of governed uni unity
[10:34] catalog system with lineage tracing version control and access control all together. Uh one last thing from me. So what we just walked through is just a very prescriptive insight summary that is
[10:48] constructed by AR query with a lot of pre-processing. This is perfect when you want to set something that the logic is predefined. You don't you don't expect a lot of changes. The data is well defined and you need to process things in batch.
[11:02] However, if you want to do explorative analysis, uh air query won't be your best friend. This is where Genie will come into place and then talk about uh turning all the same govern insight into
[11:15] a conversation and scale it across different system. So with that I will hand it over to my colleague Kimberly to talk about how we use Genie in our goto market team.
[11:36] Thank you Marina and 125 plus inputs. Thanks for doing the hard work so I can you know talk about how we can leverage Genie for the easy part. So like Marina mentioned she brought it from zero to one. She's created all these governed inputs, these trusted outputs and now we
[11:50] can take it using Genie and really go from one to a thousand. This is the tagline that was given to our teams. We're trying to make the
[12:02] field more effective. So our objective and how we're measured here is, you know, we want to make sure that the sales team has more time in front of their customers. We want to make sure
[12:14] they have faster uh access to insights. And this third bullet point, we want to make sure we're saving the sales operations manh hours. Now, I think this feels contradictory to me. How do we
[12:28] make sure we reduce the operational burden on our sales operations team, but at the same time get the salespeople out in the field more? Normally, that's kind of the tenuous relationship that you have to work with. But with these
[12:42] governed inputs, we're able to keep that strategy layer fixed. We're not increasing that operational burden and we're just simply changing the delivery service using Genie. So, at the center
[12:54] of all of this is the governed inputs. It's this 125 decision points that we've put that we've worked together with the strategy team to define. So this way judgment on the performance is very standardized. You're not going to have
[13:07] one person out there that says 5%'s actually amazing. At the same time somebody else out there is saying 5% is not very good. So it's all fixed. The metric means the same thing everywhere
[13:20] because it's all predefined. And finally, with the smart feeds available on the dashboards, the numbers and the narratives continue to stay in the same place. There continues to be a source of truth for people to reference and
[13:34] leverage. So on this left hand side, we've had these governed inputs, that foundational data set that we've created. We're leveraging the Genie platform and we're
[13:46] able to scale it to all these different pieces. Genie is able to identify when an audience changes. It's able to create push notifications. It's able to analyze structured and unstructured data. And
[13:58] I'll show you how we leverage that in our day-to-day here. So, starting with the audience changes, you have the classic dichotomy of a senior senior uh field leader as well as
[14:12] what an AE wants. The senior leader, the sales leaders are going to be talking and working with a thousand different AES. So what they're looking for is really that 10,000 foot view. What is
[14:25] the entire country of France performing? What is the quarter quarter view? What is that six quarter run rate look like? On the flip side, an AE wants to talk about the opportunity, the use case, the
[14:38] workspace level, the product. That level of granularity is completely different. And with Genie, it's able to identify, hey, this person is actually a sales leader. What we should be providing them
[14:51] is more of that, you know, longtail historic view versus, hey, I know now that I'm talking to an AE. They have six accounts in their book of business. We should be talking very granular levels.
[15:04] So being able to identify that context instead of us needing to create a new dashboard, a new analysis, a new sort of visualization for every single persona, we're able to keep it all in one place.
[15:18] As an example, we have these push alerts, these scheduled tasks that Genie can do. And in one place in the same skill we're able to have it say hey if it's a leader analyze you know the
[15:30] overall portfolio of the region what percentage of their use cases is going what is the sort of trend that we're seeing in terms of losses of use cases. So looking again at that 10,000 foot
[15:43] view and then schedule alerting to a sales rep. We're getting down to the nitty-gritty of what's on the account, what's in the use case, what are the next actions that they should be taking.
[15:57] Again, the foundation here is the govern inputs, the 125 inputs that we've created. And now we're able to very easily say, hey, if you understand the context in which you're, you know,
[16:09] getting this information, we're able to push it out in a more tailored way, in a very specific way, so that people don't have to read through information that's not relevant to them. They don't have to uh read through again like a bunch of
[16:21] different dashboards. This is sent to them right to their emails, right at their fingertips, tailored to exactly what they need to know, decreasing that operational burden and putting sellers out in the field more.
[16:35] Now, focusing more on that operational burden, you know, how many times have you been asked like various de data definitions? I sure hope I have a room of data professionals in here because otherwise this question is a little awkward. Uh but I have a couple of
[16:49] screenshots here. Um and this is just on the definition of pipe genen. This happens almost every morning that we log on. We go into slack and it's hey this use case why is it not pipe genen? Hey
[17:02] this use case why is it pipen? And I get it. It's a little bit of a complicated concept. Here's the definition. It's pipe genen uh when it's the dollar amount associated with the use case when
[17:15] it enters stage three for the first time. It took me a few reads of that sentence to really understand that conceptually, but what it comes down to is there really is a codified and a
[17:27] prescriptive way to understand what is pipe gen, what use case counts as pipe gem, what qualifies, what doesn't. And every morning when we have to log on to all these different questions, that's
[17:39] probably one to two minutes of our day that's taken away because we have to just do some double-checking, just make sure that everything is here and really run through a predefined set of logic
[17:51] that we have in our head. That's all changing with these predefined uh governed inputs because now especially in Genie, you're able to follow up in the same chat that they're sending these scheduled messages and just ask for the
[18:04] data definition. So Genie coming to a sales rep saying, "Hey, by the way, this use case is stale and you might not have enough pipe genen." The AE is able to immediately
[18:16] have this question here. Hey, I actually have this use case. Uh here's the link. um and why is it not pipe genen? So now Genie knowing that context, knowing that information is able to take that
[18:28] operational burden, go through that set logic and get to the answer for them. This doesn't increase any of myself or Marina's time. And this one also immediately provides an answer to the
[18:41] AE. There's no latency. There's no needing to wait for, hey, when is Kimberly free for a Slack message? This is available to them at all times. And next there, Genie is also able to
[18:54] identify, understand, and learn unstructured data. What that means is it's able to take all of your company documents um and all of the information and synthesize it and identify exactly
[19:06] what is relevant to the AE. So instead of needing to read through a bunch of different case studies or a lot of different customer stories, they're able to go through and understand, hey, this
[19:19] person that's asking this question, they're in this enterprise and emerging business unit and they have these types of accounts. Now, what that means is these are the exact playbooks that are most relevant and most helpful to them.
[19:32] Again, cutting through all that noise, there's a lot of things out there out there. There's a lot of information. We need to make sure that they get it at their fingertips immediately. And speaking of, now it's available on
[19:45] mobile. So before needing to pull out your laptop, needing to go deep dive into a dashboard or maybe five different dashboards. We've solved that because now dashboards scale into mobile and
[19:59] then mobile also has Genie. So the data foundation here is able to help us get to this delivery p uh platform and able to create these
[20:12] trusted insights that we can now give the field at their fingertips. Coming back to this I think we've answered this prerogative make the field
[20:24] more effective. I would like to add one more line here. make the field more effective while not increasing operational burden. It's a little less catchy, but we have 6,500
[20:39] sales individuals. And if I count the two of us, uh it's a ratio that doesn't really make that much sense unless we can use the power of AI in a trusted foundational way to scale the work that
[20:53] we've done. So again, number one, you have to start with the governed truth. Otherwise, you don't know what's what AI is going to be spitting out. All these sorts of uh the
[21:07] narrative needs to be pre-coded. Strategy analytics has put a lot of thought into making sure all the inputs are true. The second piece is really leveraging AI because of the delivery
[21:22] system that it currently has. it's able to be a tailored workflow for each individual. And thirdly, this does uh this does result in operational relief. So myself and Marina were able to
[21:35] support the 6,500 field users by using these three steps. Now, a couple takeaways I'd like for you guys here. We've heard it a lot. We've heard it multiple times. Foundational
[21:49] data is really the first starting points. garbage in, garbage out. So you need to make sure that every single one of your decision points are codified. It's make making sure that the source of truth data comes first. The second piece
[22:02] is using AI for expression, not truth. You don't want AI to be the one creating that judgment. You want to make sure that logic is prescribed already. Um, and that is where you get the strongest
[22:14] result. And thirdly, distribute it with Genie. There's so many capabilities out there now with Genie that this delivery platform all in one place is so powerful
[22:27] scaling all of our operations. So thank you everybody. We will be open for Q&A now.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.