Is there such a thing as "modern BI"? Or is BI still falling short?
I'm of two minds on this topic. On the positive tip:
- We are indeed seeing more "modern" BI (and planning) solutions that are intended for business users - and reduce the "IT reporting bottleneck."
- Even the BI tools designed for "data geeks" are getting more powerful, incorporating a broader range of data. And so-called "data geeks," are getting better at collaborating with business/domain experts. (I use the "data geeks" phrase lovingly, as I am one).
- Yes, dashboards have limitations, and are not the end state of BI, but: I've run into plenty of companies that swear by dashboards - and derive value from them.
- AI-enhanced BI tools are starting to help, especially in the ongoing chores of data cleansing/quality, and anomaly alerts.
On the downside, we need to be more rigorous in our criteria of successful BI. Yes, equipping business users with better BI tools is progress (I think of BI and/or planning vendors like Domo, Yellowfin, ThoughtSpot, Sisense, Planful, and to some extent, even Power BI).
But it comes down to: are we doing our jobs better because of these tools? And: are we making better decisions? That last point is a high bar indeed - one that should keep all BI vendors humble. In his scorching review of BI's evolution, Will BI survive? Yes, but we may not recognize it, Neil Raden made a similar point:
The path to ROI with BI is better decisions, not better dashboards, and better decisions must be fast, consistent and agile.
That's the dashboard problem in a nutshell. Attractive visualizations may be reassuring; the red-colored alerts may be a wake-up call, but making proper decisions from this information is another matter entirely. Are the users in question even empowered to make those decisions? Yes, AI may be able to help, by providing "prescriptive" actions to yellow/red alerts, but that AI promise is still in its infancy. And, as Raden points out:
It's inconceivable that output from machine learning models will simply be accepted at face value, that "decisions" made in real-time will not be investigated after the fact.
AI explainability is a vexing issue. And: decision support, however intelligent, does not inherently lead to better decisions. Spreading more data across the organization doesn't necessarily provide proper context. Data storytelling, done well, can make a difference here: see: Numbers don't lie, but they can fib - without data storytelling, your data has no context, by Yellowfin's Tony Prysten.
Delivering on the value of BI - the role of embedded analytics
I've done plenty of use cases with BI/planning customers that found great value from external BI environments. But I remain convinced that "modern BI," if it wants to earn that phrase, must blur the lines between transactional and analytics systems - from a user experience standpoint. To me, embedded BI, where users are shown the right data for their roles, in the context of their day-to-day, is a big piece of the puzzle. As Raden wrote:
Data warehouses with data marts, ODS's, cubes and obscure metadata layers in BI tools are not agile. They can't close the loop. BI capabilities need to be embedded in applications.
To get another angle on the art of embedded BI, I turned back to Yellowfin. In 2021, some of my best conversations on BI were with Yellowfin's Michael Hollenbeck. Hollenbeck has a couple decades of experience taking analytics solutions to market, so he's seen BI's struggles - and evolution - firsthand. Yet in his last two years at Yellowfin, he's seen the company thrive. When I asked him to explain that growth, one of the first things he pointed to was: Yellowfin's embedded BI. As Hollenbeck told me:
The experience of the consumer matters a ton. I think we've got some great capabilities in our product that speaks to those people. The products that compete with us, they're very much designed for people who create reports, or people who transform and load data, or people who are administering databases. And if you do those things for a living, those products are great for you.
But if you're optimizing for the consumer, and people who don't do data for a living, people who do business stuff for a living, visualizations do matter. Presentation does matter. The organic appearance of our product within their appearance matters. They don't want people to be able to go, 'Oh look, they're embedding Power BI.' They want it to disappear - and appear to be part of their product.
Can embedded analytics address BI bottlenecks?
I asked Hollenbeck about BI bottlenecks. In my view, the biggest historical BI bottleneck was with the IT team - waiting a week for custom reports on customers, or profitability by store - there's no time for that now. But in the case of modern BI, isn't there another bottleneck, waiting for the data geeks to serve up what business users need? Yes, the analytics team is faster than IT, and they live and breathe data, but: isn't the BI equivalent of low-code empowering business users to serve up their own reports and queries? Why can't business users get what they need, and the data geeks do their thing alongside, handling deeper/custom dives, compliance reporting, data integration scenarios, and getting reports/dashboards in front of execs?
Hollenbeck agreed, but only to a point:
Well, the short answer is yes, but it's a nuanced reply. Your comments are very in line, if we're talking about enterprise analytics. By that I mean, I am a big company, and I'm trying to report on my operations to my people. Where that's the case, I'd say, 'Yeah, you're right'. I would probably expand on the bottlenecks a little bit, but definitely the data analysts are a bottleneck.
When it comes to embedded analytics, Hollenbeck sees another BI bottleneck:
My US team operates almost exclusively with embedded analytics situations. Our bottleneck is a little bit different. Let's say I'm the company producing the digital product, and I have a successful product - people like my product... That means that I'm either sitting on a lot of data that my product originates, or there's a lot of pass-through information that is enriched by my product - data that people want.
One example would be, if I'm Uber, and people are using my service, there's going to be a lot of people who are sending their sales force out into the field, who want to keep track of the rides they're taking, and there's going to be a lot of information they want back.
So how does that bottleneck emerge? What is the embedded analytics pain point for customers? Hollenbeck:
In those cases, the bottlenecks that we see principally are: if they're doing it in a very inefficient way, it's the developer. Their customers are going back to whoever their contacts are at the company. They're like, 'Hey, we got to see this data. Can you give us an export? Can you give us a dashboard? Can you give us something?' And let's say they go back, and they create a dashboard for their customer. The customer goes, 'Oh, yeah, that was great. I love that. Now I want to view it like this, and like that'.
If you're not using the right tooling, all of those are requests back to R&D, back to someone writing code.
Ah, so the embedded BI bottleneck gets stuck on app development?
Yes - it's an app development bottleneck. There's a frustration on the part of the customer. There's a frustration on the part of the product manager, where they go, 'Oh, crap, we've got to be way more agile with this. Every time somebody asks for a report, it can't have to go through R&D, right?'
This is not just a report request - there is real business urgency behind it.
The product manager recognizes that this data, it prevents churn; it creates loyalty, and it probably is a huge differentiation factor versus our competitors. And if it's not a differentiation factor, it's an 'Oh my gosh, we're way behind the other guys' factor. So they're trying to hit parity.
If your product doesn't offer embedded analytics, you're falling behind:
Product management recognizes that the reporting capabilities are a huge part of the value proposition of the product. Typically, the folks that are being asked to do this, it's kind of like a side project - and so you have to elevate that information feedback loop.
Putting embedded analytics to work - healthcare scenarios
Hollenbeck cited customer examples from the healthcare industry:
Some of our healthcare customers are taking the information that they get, and making sure that they're distributing it back to their users in a highly consumable, highly integrated way. There's one CEO that I'm thinking about, they look at this stuff, and you can almost see the cogs spinning... We couldn't move fast enough for them. [They were telling us], 'My users - this is what they do. We're going to make sure that we're much more agile, that we can put the power in their hands. Oh, you guys are doing NLQ.'
Just at every phase of their product, they're trying to figure out ways where users can hit the easy button to just ask questions, or get the information they need.
Dashboards still have value, but Hollenbeck loves to see customers push further:
The step that gets me most excited is when people are thinking beyond the dashboard. They're really going, 'Okay, how do we make sure that people are informed in the stream of thought, in the context of their workflow?' And, you know, make sure that those folks can do the smartest things possible.
So how would you move beyond dashboards? Hollenbeck shared a story, about a Yellowfin prospect that's considering the moment where a doctor is prescribing a drug to a patient. At that time, the algorithm is looking at the diagnosis, assessing the possible drug options. The doctor then hypothetically receives a popup that says something like, 'Hey, you're asking for this, but check these options out' - providing the doctor with alternatives, and risk scenarios.
Arming everyone in the organization with better information is great, especially if it can be done in (near) real-time - but it's not an end in itself. The power of intuitive information is another matter. Providing alerts, data points, and prescriptive options in a role-based way, similar to Hollenbeck's doctor example, is not a guarantee of better decisions either - but I do like your chances better.
Hollenbeck flagged challenges buyers should be aware when evaluating embedded platforms (e.g. make sure the analytics are fluid from a UX standpoint; and make sure you aren't going to find yourself in an app-feature-request bottleneck). Embedded analytics is not a cure-all, but I particularly like it from the vantage point of SaaS vendors, embedding performance benchmarks directly into their product screens (Coupa is the farthest along on this of any SaaS vendor I've seen). Hollenbeck says that type of embedded benchmarks use case is definitely viable in Yellowfin.
We opened up a good debate here on the pros and cons of dashboards - a debate we didn't fully resolve. Hollenbeck argues that with a proper design, you can make dashboards more interactive, and ensure a good user experience, between dashboards and other analytics environments. He also believes customers should expect "very rich actioning into the context of those dashboards." Example: press a "call" button to contact an affected client - right from your healthcare dashboard.
That's a good place to wrap this one, but the pros and cons of dashboards, or the art of the interactive dashboard, is worth revisiting - as is the impact of NLQ. Next time.