Driving Databricks Platform With Revenue Intelligence ROI

Summary

• Revenue-focused implementation: Align your Databricks Data and AI platform adoption with revenue-driving projects like improving sales pipeline and customer retention to achieve executive buy-in and positive ROI.
• Measurable business outcomes: Demonstrate concrete improvements in critical business metrics rather than technology metrics—this drives stakeholder support and long-term platform success.
• Short-term wins, long-term foundation: Deliver quick ROI on your platform investment while simultaneously building the data infrastructure required for sustained competitive advantage.

Driving Databricks Platform With Revenue Intelligence ROI

Talk by Joel Fuernsinn, Senior Director, Data Management, Veeam
Demonstrating a real ROI is key to driving executive and stakeholder buy-in for major technology changes. At Veeam, we aligned our Databricks Platform change with projects to increase sales pipeline and improve customer retention. By delivering targeted improvements on those critical business metrics, we created positive ROI in short order while at the same time setting the foundation for long term Databricks Platform success. this video targets data and business leaders looking to understand how they can turn their infrastructure change into a business revenue driver.
Here’s more to explore:
Economist Impact: Unlocking enterprise AI: https://www.databricks.com/resources/webinar/unlocking-enterprise-ai
The State of Data + AI: https://www.databricks.com/resources/ebook/state-of-data-ai
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements

FAQs

How do we prove Databricks ROI to executives?

Align your platform implementation with revenue-generating projects (sales improvement, retention) and measure business impact, not just technical metrics.

What if our AI initiatives haven't generated ROI yet?

This video addresses the current "AI disillusionment" by connecting platform investments to concrete business outcomes and proving value quickly.

How long does it take to see Databricks ROI?

Veeam demonstrated positive ROI in short order by targeting high-impact business initiatives, while simultaneously building the foundation for long-term platform success.

What business metrics should we track?

Focus on metrics your executives care about: sales pipeline growth, customer retention, cost of operations—not just technical KPIs like query performance.

Full transcript

Good morning everyone. Uh thanks for coming and welcome to driving data bricks platform acceleration with revenue intelligence ROI. Um so just to go through the standard uh notes please complete your surveys. Uh you'll probably see this slide a hundred times this week but um all righty. So, one of my favorite books is Cryptonomicon by. And if you aren't familiar with the book, um it's basically set in two timelines. One is in World War II and one is in kind of the late 90s, early 2000s time frame, uh and focuses on a group of folks doing um cryptocurrency and things in that space.
So, in the book, there's an exchange between two of the engineers uh discussing why they need to put up with some of the things that aren't really of interest to them on the business side. And it boils down to this quote. Now, one of the reasons I love this book is that I see a lot of familiar traits in the characters in the book in both myself and the people I've worked with over the years. The reasons I got into data and engineering are that I love working with technology.
I love solving problems to try to make things better. And I love working together with people with that same type of mindset. The problem, in the words of in the book, is that if you and I are going to hack and get paid for it, some people have to hire us. And ultimately, the reason that companies look to hire and build things is that they expect to get a positive return on investment from whatever it is that they're doing.
Now, when we look at um the capabilities of AI and machine learning, there's no question that the last few years have been explosive, right? Everyone wants to get into the AI game and for us folks in the data world this should be hugely exciting but after the initial kind of wave of euphoria from the intro in introduction of chat GBT um that's starting to wne we've we've sort of reached that crest of the wave of expectations and I think we're we're going into the trough of uh of disillusionment and but why do why do companies think that right so for starters Last year, OpenAI had about 5 billion in losses, and they're projected to have as much as 14 billion in losses this year.
From a survey of technology groups, only about one in seven actually had budgets allocated for their AI initiatives for this year. And in another survey, almost three quarters of financial analysts expected AI to have no impact on profitability for the businesses they covered. So what I want to do today is give you a roadmap that's going to help you fight against that trough of disillusionment, align your data and technology strategy with your business objectives, understand how you can leverage data bricks uh and the data bricks platform to accelerate your work and ultimately demonstrate a clear ROI for your data bricks and AI projects so that we can all keep doing the work that we like to do.
So who are we? Uh if you don't know VH, uh Veh's the world leader in enterprise backup and recovery software. Uh we power data resilience from backup and recovery to endto-end ransomware protection for over 550,000 organizations in 150 plus countries. Uh VH has grown rapidly over the last decade with an average growth of 15% year-over-year. Who am I? I'm the senior director for data management at VH. Uh I've been with VH just over two years now. uh but I spent two decades working in data and software engineering for a number of different consulting and industry firms.
Now one of the reasons I came to beh uh is that my first real entry into data work was as a DBA and so backup and recovery is something that speaks to me on a fundamental level. So I have no problems telling anyone that yeah this is I love the work that I do and I I feel good about the work that I'm doing for our company. So there's there's my veh pitch. So for our our platform implementation, where did we start? If I roll a clock back about two years ago to when I joined BH, um we had an existing data warehouse and some data science capabilities, but there were some core issues with what were in place.
First, our ETL processes were customuilt in-house programs. And as you can expect, this led to some challenges in hiring and ramping up new employees. And basically, we'd written a bunch of custom Cotlin ETL. Now I like cotlin as a programming language but the pool of cotlin developers out there is relatively small and the pool of people who know know cotlin and know rel is about 10 people in the world which is not a great place to be.
Second our data was siloed and generally focused on just our day-to-day reporting. Um and from both a business and an ML AI standpoint right bringing together those disparate data sets is really where you're going to start generating a lot of the value for your projects. Our existing infrastructure had limited CI/CD capabilities. We we basically have had and have a single prod environment in place for our existing data warehouse.
And as you can imagine that causes all kinds of problems in terms of automation, code quality, deployment. It's it's just not a great place to be. And lastly, our data storage and our compute costs were tied together. So that's that causes a fundamental problem for larger data sets, right? especially when you're talking about um you know really large or unstructured data sets being able to store that data um and then do processing on it is having having those two things tied together just created a real bottleneck for us.
Now the good thing for me was that when we were talking about making changes to our data platform that aligned perfectly with our overall company goals. Every year at our internal beam launchpad event, our leadership publishes these set of core or a set of core goals that cross our sales R&D and our internal corporate groups. And so when I look at the goals that we have for last year, transforming our internal data analytics and AI into a world-class capability goes handinhand with that.
Right? If I look at accelerate innovation as a company, veh has we're we're launching 27 different product and platform releases on our roadmap this year. Um our our teams are rolling out uh capabilities and lines of business at an incredible pace. And so we really need to make sure that our data platform continues to scale with our business digitally transform. Um, so like every other company, right, we want to get into the AI and ML game um, and really use that to improve our business.
And we want to make sure that we take away technology roadblocks and make our processes as seamless and automated as possible. And then on empower, engage, and empower, engage, and grow. This is for both our customers and our employees. Um, we want to put tooling in the hands of our customers so they can easily back up their important data and workloads and verify that it's correct. And then internally, I want our data platform to be accessible to everyone with proper governance in place so that everyone from our CEO all the way down to a day one salesperson can get value from it.
Now, I'm lucky enough that I've worked with data bricks at our at my last two companies and so I knew the capabilities of the platform and that it would go handinhand with what we were trying to do here and so I I it wasn't new to me. But I also know from two decades of working in the industry that anytime you have a major platform initiative like that, it's right, it's not free in dollars, time or resources, right? So someone's going to ask the question, hey, is it worth it for us to do this?
So then for me going to our leadership the question was right how can we align the data bricks transformation that we want to do with our business make sure that we've defined a critical capability and then forecast and track an ROI so that we can then continue expanding the platform for the rest of the enterprise. So how do we go about identifying our target ROI? Now, it's going to sound cliche, but the first step in identifying your business case is like actually know your business.
And at BEH, right, we're primarily a software as a service business. So, for us, customer acquisition and customer retention are two of our biggest drivers overall. I know that when I go to our quarterly quarterly all hands meeting, our execs are going to talk about sales pipeline. They're going to review our customer retention numbers. I know they're going to talk about customer acquisition cost. So for your respective teams and companies, right, hone in on what your execs and investors talk about.
If you can't align what you're doing to what they're talking about on a regular basis, that's a a I don't want to say a recipe for disaster, but being able to align that is going to make your life easier. Um overall it was also useful for us to look at where um data domains complement one another because that can help you lay out a road map that gives you uh quicker overall return. So for in our case, if I'm looking at sales pipeline, I know I'm going to need CRM data.
I know I'm going to need some HR data. I know I'm going to need probably some finance data. But if I look at customer churn, there's a huge overlap in those data sources, right? It's probably like 90 90%. And so that means that I can do multiple projects generating ROI um at once without doing everything serially. It's not that I have to go get a bunch of data from one place then a bunch of data from another place. I can do those things and uh generate that for multiple projects.
Now like I know I know this sounds obvious um but there's too many people in the data and AI space that are in a technology bubble, right? And if you're one of those technology folks, and I'll like I will 100% tell you that earlier in my career and probably still today, I am one of those folks. I love technology, right? So that's the fun part to me. But I can't emphasize enough how understanding the strengths and weaknesses of your business, the biggest revenue and margin drivers, and the biggest cost impacts will in turn give you a leg up in your career.
As one other side note, um something I'd recommend to everyone, if you can focus your initial project on business growth rather than cost reduction, um that is always going to be a win, right? If you're able to improve topline revenue numbers, everyone's going to be happy. Um no exec is going to say no to cost reductions, but they are going to celebrate that topline revenue growth. Targeting realistic ROI. So, it's easy to get caught up in the AI hype, right?
And say this is going to fundamentally change our business. And in some cases, it might be true, but for most of us, those changes are going to be, while they may be significant, they're also going to be incremental. And so, setting that realistic ROI target for you is going to be critical in terms of revenue intelligence projects, right? It boils down to what percentage of revenue margin are you impacting? At VH we're large enough that even small changes can impact millions of dollars for us.
So we we had some flexibility, right? But is that the case for your business? How do you make sure that the impact that you're going to have is actually going to be significant enough to keep driving that forward? What is your organization's technical maturity? Right? Do you have the folks who know this technology? Do they have experience at large scale data uh at large scale change management? Are you going to need to hire an external firm for consulting or staffing?
I was a consultant for a long time, so I can tell you right that will help, but it ain't going to be cheap, right? And so make sure you're factoring that into your uh your ROI calculations. Um, and then what's a realistic timeline for you? So in our case, not only were we starting our data platform, but we were also asked to build it in a cloud provider that was different from what our internal corp tech group had had typically done, right?
And so that cost us a significant chunk of time at the start in terms of initial cloud setup, right? All the networking security um you know all the all the things that cause a headache there, right? That was a blocker for us. So for your respective projects, what are those blockers on your front and then finally on understand the ongoing costs, right? Something that execs or folks love to forget is that it's not just nothing goes on working forever, right?
Everything is going to need care and feeding. And so, how do you understand when your model is no longer effective or if there's business changes you need to account for or if there's just technology changes that need to go uh go into it? Right? It's never going to be just fire and forget. All right. If you take nothing else away from this discussion, I hope you remember this one. So defining your measurement and proving you've delivered that ROI is the most important thing you can do, right?
What are the specific KPIs you're going to change or you're going to impact with your change. Is it revenue? Is it margin? Is it customer satisfaction? Is it something else? And then how do you programmatically identify the changes caused by your work? And how do you make sure that you are uh you can attribute those changes to the work that you've done? There's nothing more powerful than being able to provide the receipts.
But it also means that you make have to make sure that you have the data and processes to back it up. Just one story to illustrate the point. So I was working on a consulting project for a railroad transportation client um probably about 10 years ago now at this point and we had a bonus writer built in that said if we delivered $20 million in revenue uplift for them, they owed us a million dollar bonus. Now even with $20 million in hand, no one likes writing a check for a million dollars, right?
That's not fun for anyone. But because we had built that telemetry in, there was no question that we delivered and we got paid, right? So what can you do to make sure that you get proverbally paid on your end or actually paid on your end? So with that overview out of the way, right, what did we actually build? Again, we're a SAS company, so sales pipeline and customer turn at the heart of our business. So no surprise that's what we focused on first, right?
And our drivers for SAS are easy, right? It's sales growth and uh right, how do we how do we uh keep driving new sales growth and how do we make sure that we keep our existing customers on the sales side should be easy, right? Um as of 2024, BEH is the number one uh enterprise backup and recovery company in the world. But even with that, our total market share is just over 15%. So there's 85% still out there for us to go grab.
We also, as I mentioned, have 27 new product launches coming this year. So, that's going to address even more of our customers needs and workload. On the other side of the coin, if we want to keep driving 15% year-over-year growth, that means we need to generate $250 million in additional revenue in additional revenue this year. Right? The laws of compounding start working against us. On the churn side, we want to make sure we keep our existing customers, right?
In a perfect world, we'd never have someone leave us, but we all know the world ain't perfect. So, we have 550,000 customers in about 150 countries, and we're also working with thousands of different channel partners. So, for our customer success teams, it's incredibly hard to keep track of all of that. The good thing is that again, the scale of our business means that even small reductions in churn can drive a huge amount of revenue, right?
Every 1% insurance reduction for us is a million dollars and change, right? So 3% is $3 million. That starts adding up in a hurry for our work. We also then started with the data needs. Um we we wanted to make sure that we were uh focused on delivering that initial product and initial ROI as quickly as possible. So we wanted to start with that MVP but then also have a roadmap to layer in more data as we went on. Right?
Um I mentioned before uh we were looking for those places that we could get economies of overlap when you're thinking about your own projects right where do you have those economies of overlap in your respective companies. Um, now if you are someone who's not uh more a businessincclined person, not a technical person, um, like I I want to call this piece out. Historical data availability and need are what I've seen trip up more projects than anything else, right?
It sounds like it should be easy. Be like, oh, give me the last three years of data. But there are so many things that can cause a problem with that, right? Have you had system changes, right? Have you changed a CRM? Have you had an upgrade? Have you changed your HR system? Um, do you have your systems keeping track of transaction history? And if not, where are you going to go get that? And lastly, do you have enough history to both train and test your model?
For most models, right, you typically need somewhere between two and three years of data to start providing accurate results, right? So, where do are you going to be able to provide that? And it's okay if you don't know the answers, but if you don't, go have that conversation with your respective data teams first thing on Monday morning. and figure out what the answers to those questions are. So what did we actually deliver?
Um the V- account IQ tool was delivered in Q4 of last year uh to an initial PC group and then we rolled it out to our enterprise uh in January of this year. Uh it's basically a sales propensity tool that uses a cross factor score to combine our customers fit, need, and intent. The scoring is adjusted for different sales plays. So you can look at new new and existing logos differently and focus on specific product lines.
Here you can see our sales people can select from BDP which is our traditional data platform. Uh BDC which is our cloud offering, Salesforce, any number of other products. The thing that I want to highlight on here is the Salesforce button. This button is what automatically creates a task in Salesforce for our respective SDR or salesperson and leveraging that leveraging that automation is what allows us to specifically track the opportunities and revenue that's generated by this tool.
So going back to tracking the metrics, right? This is how we're doing that. If you look at the screenshot here, right, I can see the first account has a cross factor score of 87 for VDP. So seems like a pretty good candidate for us to go after. So what do I do next? If I click on the customer, it now provides detailed information for the seller uh for from our CRM and from other data sources. I can see the details on how the score was generated.
And in this case, right, I can see that the account has a competitor product installed, but it also looks like they've been there for a while. So, they could be coming up on a renewal. So, there's potentially an opportunity there. We also have multiple scores in the fit and need categories. So it's likely they can b even if we have a competitor there, there may be something else in our portfolio that we can uh sell to them.
This information lets our sales team really tailor their pitch accordingly. And as the cherry on top, we've integrated VH's internal GPT instance with the data, providing a tailored summary and recommended next steps. So from a business standpoint, what we've done is shortcut a lot of the manual work that our SDRs and sales teams need to do and given them a playbook for them to go after their next best lead. We also wanted to put in place something for our execs and leaders.
Um right for any tool change management is always a huge issue. So we wanted to enable our leadership and sales operations to see both the users of the tool as well as the return. The management dashboard we have lets us track how our users are using it and make sure that we've done the training and change management appropriately. Right? We actually had an issue where we saw a big uh initial usage in in our air region but AMIA didn't pick it up as much and so going back and making sure that we done the training for the AMIA group and then we immediately saw a spike in usage after that.
Right. So, how do you make sure that you can do that same kind of tracking? And then also being able to tie our opportunities and dollars back to the tool. So, with the initial results um from what our our leadership has seen, right, this uh this also let our sales leadership include account IQ into their weekly meeting cadence, which then in turn further drove adoption of the tool for us. So what did what did we actually achieve with our tool?
Um as I mentioned we we deployed account IQ to our enterprise in January and what we've seen so far is an overall increase in sales pipeline of about 4%. Right? In the grand scheme of things 4% you know it doesn't sound that much right but again that's millions of dollars for us. The other metric we specifically targeted was an increase in our first contact conversion rate by two basis points. Uh so in SAS overall that metric is uh the best folks in the business are typically in the 15 to 20% range for that first contact conversion.
And I won't tell you where we are but it was definitely not 15%. Now the good news for us is that even um after only being deployed for five six months now we've already seen that metric tick up by one and a half basis points. So we're on a good track to hit that by the end of the year. um on our churn model side which we dubbed it churn IQ of course um so we've actually just deployed that to our mayor teams about a month ago so it's a little too early for us to get a result on that front but our target for that is an overall churn reduction of 3% year-over-year and again that 3% is millions of dollars for us ultimately being able to show these types of returns and attribute it to the work that we're doing that makes it real easy for me to now go to my executive leadership and say, "Hey, I just got you $10 million.
I need a little more money to go do something else cool." Um, all right. So, I want to pivot a little more to the technical side. Um, why do we choose data bricks and what did it solve for us? So, the two themes I want to focus on are unity and simplicity. Um, I'm sure everyone has seen this slide or something very much like it, right? It's straight off of the data bricks reference uh site. But the reason I want to highlight it is that one of the principles for our project was to try and leverage as much of the standard technology as possible.
Right? I I don't want to go back to the business of custom building everything in-house. So unity catalog ingest and transform automation all the advanced ML and AI capabilities right that was part and parcel of what we were trying to do on standardization front. Unity catalog um right I'm sure if you're at data bricks you know about unity catalog but for us that takes care of a lot of our needs in a way that's integrated and with minimal setup right access management is standardized with our existing enter groups um so our business can manage their own permissions um based on their respective job roles and it makes sure that those updates get handled automatically I don't need to go and manually add a bunch of people to an ad group um we also have the row and column level security pieces Right?
We're pulling in HR data. There's some sensitive stuff in there. We can't be broadcasting that to everyone. Right? So, having the built-in security there uh was was huge for us. On the data discovery front, um again, I mentioned I we want our platform to be open for to everyone, but most people are not are coming into data at a more beginner level, right? So, how do we make it easy for them to get started and start using the platform?
You've got the data catalog, you've got quick search, you've got that organization that gives people a good starting point to say, "Hey, what am I looking for and where can I go get it?" And then lastly, the data lineage piece helps everyone answer the eternal questions, right? Where did this come from? How'd you get this calculation? Um while there may be some additional investigation we need to do um being able to trace that lineage is a huge help for our for our engineer engineers and analysts and helps answer those questions uh much more quickly.
On the ingestion side again we want to let data bricks do as much of the heavy lifting as it can. Um, you know, we I've got a bunch of smart and capable folks on our DE team, but we are never going to be able to throw the same kind of resources at ETL that data bricks can, right? Why would I try and go and do this when I don't know how many data bricks uh people data bricks has, but it's way more than we do. So, let's let them handle that.
Um, you know, autoloader is right like every uh almost every organization, S3 is a source of data for us. Before autoloader, right, you had to keep track of files. when did they come in, what got loaded, um you know, where's it going? And autoloader takes care of a lot of a lot of that for us. So, we can be more focused on the actual contents of the file and the quality of the file rather than did I get a file on the streaming side.
Um I talked about this earlier. We have we have Kafka as a input source but um our old platform actually caused a big problem for us because we didn't have the storage capability to keep track or keep all of the raw messages and so this caused a problem for us when we had actually had a bug in how we were doing some of the ETL because we couldn't go back and rerun our logic on the older messages right our our message uh Q retention was at a week anything older than a week was gone So now we're able to keep the raw data. we're able to store that and if hey if something goes wrong we can always go back and reprocess that if we need to and then lakeflow connect um so last year conference when they talked about lakeflow connect and announced the things that were coming out uh it was it was like they went into my brain and said hey what's your wish list of stuff right so Salesforce workday service now SQL server those are all things that are um critical data sources for us I in full disclosure, right, a lot of these things are still in preview.
Um, but we've been working with closely with our technical our account uh technical team and uh have been have been testing these things out. And so while we haven't put all of them in place at this point, I feel I feel pretty good that when we have this uh conversation next year that a significant chunk of our ETL is going to be handled by that again taking load off of our engineering staff. On the automation front, uh again you have built-in jobs and workflows uh that makes your orchestration easy, right?
It's all there. I can see what ran, when it ran, did it fail. Um, we have you have that centralized monitoring and logging and you also have integration for a host of thirdparty tools, right? We happen to do our uh sort our error handling through a combination of teams and pager duty. I have plugins that make that easy for both those things, right? There's and there's a host of other things out there uh plugins and just base connectivity that is going to make your life more easy.
On the CI/CD front, um, data bricks asset bundles are right helps centralize that deployment configuration and how your deployments work. Um, and I would recommend that everyone if you're not using it, go take a look at it. Um, it's got that standard integration with all your major CI/CD platforms and really just helps make uh make Git the default process for all of your uh for all of your work. And that has benefits all on its own.
And then on the MLOps side, it again you've got standardization in how you are doing deployment and management of those models and it helps build some good habits for your data science and ML people, right? I uh data scientists are really good at what they do, but they tend not to be great engineers, right? And so there's a lot of like one-off I built this thing in a notebook and or you know I've got it locally on my machine or something like that, right?
So, uh, just helping build some of that framework and good habits is good for everyone across the board. Um, and then also the monitoring for accuracy and drift and right, how do you keep track of what your models are doing over time? Uh, and see see that something is changing before it completely implodes. And then finally on the data warehouse front. So this is a place where right anytime you have a major change doing that change management is going to be a big part of it but data bricks data bricks makes this super easy for the vast majority of folks who have any experience with BI or analytics previously right you've got SQL compatibility it's like NCSQL is NCSQL I can write something on pretty much anything else and it'll probably work here like maybe I have a couple little things I need to change but that translation is easy um you've got a familiar familiar database style UI.
So you've got an easy starting point as a user. Oh, this looks like every other database tool I've used. Not much to uh to get trained on there. You have supported connectivity for Tableau, PowerBI, as well as the integrated AIBA AIB capabilities that again make that transition easy. I can hook up Tableau, I can hook up PowerBI, I can hook up um you know a host of other tools, get results without a ton of overhead. I now I will fully say that uh we are probably we are we are using some of the BI capabilities less of the AI right now but what I really like about the BI side is it gives an easy place to uh even if you're using another uh BI tool do some initial frameworking get those results be able to see something before you go through the process of creating a dashboard somewhere else and then lastly you've got the in integrated data bricks assistant so hey how do I do something folks can just type that in and um at least in my experience, you will probably get the right answer, right?
Nothing's 100% but it is it has been pretty good from what I've seen. All right, so coming out of the last 12 months, um what did we learn and what can you learn from our mistakes on the architecture and engineering side? Um so DT is easy but limited. What I mean by that is right it they are great for getting up and running. It's gives you a good starting point, but it also has more limited configuration than some other options.
And if you're really looking for a vast majority of things that you're doing, it's probably fine. But if you're really looking to kind of eke the last bit of performance and optimization out of something, DT might not be the right uh way for you to go on that particular process, right? Look at some of the other options you have. Serverless is a default. Um, if you're not using serverless, I like just go use serverless.
That is uh uh it has made everything so much easier. I don't have to worry about um turning clusters on and off. I don't have to worry about sizing and management and all of that. Using that as your default is hugely simplifies everything that you're going to do and I would absolutely recommend it to everyone. Um and in fact, we only have basically have one uh still assigned cluster that's running for our data science team and they just use it for specific things.
And then lastly on the CI/CD piece, this is actually a place where I wish I had um sort of put my foot down a little earlier in the project. Uh we essentially started with just a single environment and it just it creates some bad habits. Uh, and I really wish what I would recommend for everyone is especially if you're starting out, even if you don't have like a full dev test stage prodized fashion is something that I would recommend for everyone.
So um it's I knew it and I I should have been uh been more emphatic around it uh when we started but um yeah definitely look at that for your for your own projects on the model side. Um so on our propensity to buy model uh for us we basically do a prediction at the account domain and country level right we are a world worldwide company we're selling in all these these different countries so for us it's not just Coca-Cola but it's Coca-Cola and Germany and um you know is there Coca-Cola or whatever the right so that's the level where our predictions um work the best um CRM our technographic and our customer sentiment are by far the most significant input factors, right?
CRM is is probably 80 to 85% of what is dictating the model. The fmograph or sorry, the technographic and customer sentiment is the other, you know, 5 to 10% each of those. And then everything else is a far distant like 5%. And so for your for your own projects, right, that's that's where I would start if you're looking to do something similar. Um, we actually do our primary school scoring still with a rule-based engine, right?
It's it's always fun to say, "Hey, we're using this fancy crazy algorithm." But again, get started, get something in place, and a rule-based engine a lot of times will give you that 80% and start getting you that return. And hey, if you still need something more um fancier after that, you can always do that later on. Um and then but to that point, right, we all we do have second secondary scoring that's coming from uh from an upsell model, which is basically a random forest model. on the churn side um we use a binary model there.
So we initially were looking at any dollar changes on our customers, but right there as a SAS company, there's any number of reasons that a customer workload could change. It was just too noisy for us to get good predictions out of that. And so we ended up with right did a customer turn or did they not? And that really helped stabilize the model overall. Um uh cat boost was what ended up providing us the best combination of performance and results there.
I am not a data scientist so I will not go into the gory details there but uh I I can get you to someone if you have specific questions. Um but then on the scoring side, we do our scoring from a range of four to one. In talking to a number of different companies, this is this is extremely common, right? Depending on who you talk to, some are 3 to one, some are 5 to one, but they're all using the same kind of logic of what is the uh what is the confidence that you have.
And then I do want to call out the two in one score, right? for any model having the model say either I don't have enough confidence or I don't know enough to give an answer should always be a valid outcome for you right and you can you can go um delve into that and figure out why that's the case but in the same way if someone asked me a question I don't know the answer I'm not just going to hopefully spout off a bunch of nonsense I'd be like I don't know let me go figure it out so for us next steps um again leveraging the data because of the data that we have the easy next steps for us are right customer lifetime value segmentation pricing uh product development feedback right those are all right within the same data sets that we already have available so again just how do we keep boosting that ROI without sort of linearly boosting our effort um it is the data platform for our enterprise right so we want to uh expand to some of the other uh other areas that we don't currently have right marketing finance uh pulling in more of our thirdparty data etc And then also how do we expand our user base right?
Uh again we we've literally opened this up to every single person in the company but that also means that we need to provide them with some standardized training. It means we need to provide them with some onboarding especially for folks who are on the analytic side right how do they get up to speed on this more easily and then something that we are doing internally we call it beam data days but it's essentially our internal data community building effort um and we did this for the first time last year and I will like I'll call out the team that did that like they did an amazing job and we're really looking to build on that and especially coming out of uh data bricks conference how do we take the things that we are learning at conference and spread that to the rest of the folks who are like either aren't here or um aren't able to attend for whatever reason.
All righty. So this is a quote from our COO and um comes right out out of our last board meeting and what this tells me is that right if RCO is talking about this at our board meetings that means the work that we're doing is having an impact and that probably means that the work that we we're doing is going to be able to continue and that we're going to be able to continue to build on it. And so seeing this and hearing this from him was a you know validation is not the right word but it's always right you get that type of quote and it's like all right we've done our job well.
Um so again thinking about this how do you get this quote from your respective uh sea suite lastly um all right so we we've covered a lot in the last I don't know 35 or so minutes um what I'd like you to do uh before we close it out is again think about how you can use what we've talked about today if you're on the business side what are those problems that you can leverage data bricks and your technical teams to help solve.
If you're on the technical side, how can you align your technical work to those business problems? And what's the most impactful ROI that you can go after so again that we can all keep doing the work that we like to do. So, thank you very much. Thank you for coming.

Learn more about the Databricks Data and AI platform.