Analytics: From BI To Prescriptive Analytics - Nick Jewel, Timo Elliot & David Wright

Estimated read time: 1:20

Summary

In this insightful panel discussion from the Data Innovation Summit 2018, experts from Alteryx, SAP, and FICO delve into the journey from Business Intelligence (BI) to Prescriptive Analytics. They emphasize the shift from merely collecting data to using it for digital transformation, highlighting the need for speed and collaboration across teams. The panelists discuss the democratization of data science, the importance of agile and iterative processes in leveraging data, and the role of cloud technology in enhancing data-driven decisions. With a focus on cross-functional teamwork, they emphasize how organizations can integrate analytic models into everyday business processes, striving towards a future where data automatically improves service and productivity.

Highlights

The importance of speed and collaboration in evolving from BI to Prescriptive Analytics. 🤝
Data's crucial role in digital transformation and creating new business models. 💼
Making data science accessible to non-experts is key for innovation. 🧠
Agile methodologies enhance the adaptability of businesses in a data-driven world. 📈
Automating data processes can lead to continuous, self-improving services. 🔧

Key Takeaways

Speed and collaboration are key in the transition from BI to Prescriptive Analytics. 🚀
Data is now central to digital transformation, creating new business opportunities. 🌐
There's a growing need for democratizing data science, making it accessible to non-experts. 🤓
Iterative and agile processes help in quickly adapting to changing data landscapes. 🔄
Prescriptive analytics allows for automated improvements in services and productivity. 💡

Overview

The panel at the Data Innovation Summit 2018 brought together experts from Alteryx, SAP, and FICO to discuss the evolution from BI to Prescriptive Analytics. They highlighted the importance of rapid adaptation in the usage of data, underscoring the shift towards digital transformation. The experts emphasized that data is no longer just for making decisions but is also driving new business models, making its role more pivotal than ever.

A major highlight of the discussion was on the democratization of data science. Panelists argued for making data and analytic tools accessible to non-experts, suggesting that this democratization is vital for innovation and practical implementation in business processes. This approach would integrate analytic models directly into day-to-day decision-making, enhancing operational efficiency and encouraging collaborative cross-functional teams.

The conversation also covered the application of agile methodologies to manage ever-evolving data landscapes. The panelists stressed that iterative processes enable businesses to quickly adapt to changes, making them more resourceful and competitive. They concluded with insights into how prescriptive analytics can automate improvements in service provision, thereby fostering environments where processes learn and develop autonomously over time.

Chapters

00:00 - 00:30: Introduction and Panelist Introduction The chapter introduces the topic of the Data Innovation Summit 2018 and the new panelists. The host welcomes viewers back to the summit and introduces Nick, Timo, and David, who are panelists from Alteryx, SAP, and FICO. The correct pronunciation of FICO is clarified.
00:30 - 01:00: Introduction to BI and Prescriptive Analytics The chapter titled 'Introduction to BI and Prescriptive Analytics' begins with a discussion on the transition from Business Intelligence (BI) to prescriptive analytics. It invites participants to introduce themselves and explain their roles concerning BI and prescriptive analytics. One of the speakers, Timo Elliot, introduces himself as an innovation advocate at SAP, a company known globally for its leadership in business applications. He shares his 30-year background in analytics, highlighting his experience in educating business users about the power of data.
01:00 - 03:00: Digital Transformation and Collaboration The chapter titled 'Digital Transformation and Collaboration' discusses the role of Alteryx, a leading company in self-service data science. The focus is on guiding users from being line of business analysts to becoming citizen data scientists. It highlights the trend towards optimization in this field.
03:00 - 05:00: Challenges in Data Science and Business Collaboration This chapter discusses the significant changes and upcoming trends in the analytics market, especially as they pertain to data science and business collaboration. The conversation indicates a lot of excitement around future developments, specifically looking forward to changes and innovations expected in 2018. Though not explicitly detailed in this excerpt, the context suggests a dialogue around evolving analytical tools and practices that could impact both industrial success and overall market growth.
05:00 - 08:00: The Role of Technology in Accelerating Business Processes This chapter discusses the pivotal role of technology, specifically analytics, in accelerating business processes. Traditionally, analytics involved the collection and storage of data to enhance decision-making. Today, data is fundamental to digital transformation, leading to the creation of new business models that leverage data more directly. These models aim to improve customer experiences and open new avenues for selling data. Consequently, there are emerging demands on professionals in the field to adapt to these changes.
08:00 - 10:00: The Importance of Democratization in Analytics The chapter titled 'The Importance of Democratization in Analytics' emphasizes the evolving role of individuals working with modern technologies. It highlights the need for these individuals to be both technically proficient and business savvy. The discussion focuses on the crucial aspects of speed and collaboration in digital transformation. It cites the significance of cross-functional collaboration within organizations. Additionally, it references a survey by MIT Sloan that underscores the importance of these elements.
10:00 - 12:00: Innovation and Experimentation in Business The chapter discusses the significance of innovation and experimentation in business, highlighting the importance of collaboration within organizations. It cites that about 70% of organizations successfully undergoing digital transformation achieve this with cross-functional teams. The chapter emphasizes transparency and mass engagement involving data scientists, analysts, and information consumers, ensuring they contribute meaningfully to the business processes. It advocates for approaching challenges by focusing on business problems first, rather than being solely driven by data.
12:00 - 15:00: Prescriptive Analytics and Automation The chapter discusses the importance of identifying business challenges and utilizing data to address them, particularly in the context of digital transformation. It highlights that many companies, including those in the manufacturing sector, are still on this journey. The chapter notes the struggle due to a shortage of data scientists and experts.
15:00 - 18:00: Impact of Cloud and Open Source Technologies The chapter discusses the growing significance of cloud and open source technologies in various industries, particularly focusing on large manufacturers. A major challenge identified is the shortage of data scientists. This shortage highlights the necessity for business analysts who can translate business problems into terms that data scientists and IT professionals can address. The chapter emphasizes the importance of collaboration between these distinct groups to effectively tackle industry-specific challenges using data-driven solutions.
18:00 - 20:00: Future of Analytics in Business Processes The chapter explores the future role of analytics in business processes, emphasizing that the integration of analytics is an ongoing journey. It highlights the importance of raising awareness within companies regarding the benefits and importance of analytics. Additionally, the need for data scientists to collaborate effectively with business teams is discussed, along with the challenges businesses face in this integration process.
20:00 - 24:00: Challenges and Changes in Traditional Business Models The chapter discusses the challenges businesses face in adapting traditional business models to incorporate data science. The conversation highlights the difficulty in formulating data science questions, particularly for those not familiar with the domain. It mentions the importance of understanding the possibilities and capabilities of current technology to craft solutions. The rapid pace of technological change complicates this process, making it vital for businesses to continuously adapt and learn in order to leverage data science effectively.
24:00 - 26:00: The Role of AI and Machine Learning in Business This chapter explores the transformative impact of AI and Machine Learning in the business sector, focusing on how these technologies enable faster and more agile decision-making processes. It highlights the shift from the traditional, slow-paced methods of using data effectively to run businesses, towards a more iterative and flexible approach. The chapter emphasizes the competitive advantage gained by businesses that can quickly adapt and respond to changing conditions by leveraging AI and Machine Learning to make data-driven decisions in short cycles, typically ranging from a few hours to a few weeks.
26:00 - 32:00: Balancing Innovation with Risk Management The chapter discusses the impact of new technologies, particularly cloud technology, on innovation and risk management. It highlights how these technologies enable faster business operations and enhance data utilization, changing the traditional approach to managing risks and innovation.
32:00 - 36:30: The Future of Business Analytics The chapter discusses the evolution and growing accessibility of machine learning and AI technologies in business analytics. Although machine learning and AI have been around for many years (as evidenced by FICO's 25 years of expertise), the development of new tools now allows companies to adopt these technologies more easily. The chapter emphasizes that while many algorithms themselves are not new, the ability to quickly process massive amounts of data in the cloud is a significant advancement, enabling broader and more efficient utilization by businesses.
36:30 - 48:00: Ethical Implications of Advanced Analytics The chapter delves into the ethical considerations surrounding the use of advanced analytics in organizations. It notes the recent acceleration and transition of data analytics from theory to practice. There is a focus on the democratization of technology, allowing more individuals within organizations to understand and apply data analytics in relatable terms. The conversation highlights a shift away from complex jargon towards more accessible language, reflecting on the importance of data literacy among users.

Analytics: From BI To Prescriptive Analytics - Nick Jewel, Timo Elliot & David Wright Transcription

00:00 - 00:30 [Music] welcome back to data Innovation Summit 2018 glad to have all the viewers back here with me and we have some super interesting new panelists with us we have on a first-name basis Nick Timo and David from Alteryx s AP and FICO is it FICO stressor fi Co if I go all right there we go so this panel is
00:30 - 01:00 about from bi to prescriptive analytics what is next please introduce yourselves and say a little bit what you do with regards to BI and prescriptive analytics good afternoon everyone my name is Timo Elliot and I'm an innovation at Angeles for sa p SI p is world leader in business applications and my background is in analytics for the last 30 years I've been talking to business users about the power of data what fantastic
01:00 - 01:30 so I'm Nick Jewell I'm a director of product strategy at Alteryx so we're a leader in the self-service data science space and really we take users on that journey from line of business analysts all the way through to citizen data scientists but certainly seen optimization being
01:30 - 02:00 very successful and growing in in Satya now certain events events are outstanding well a lot of the dead delegates have sent in questions and many of them have in common that they're wondering what's coming up next I think it's very excited to be easy to be excited about what's coming up next so why don't you give us a little primer on what you think is gonna be relevant in 2018 and what's you know what is everybody talking about now well the big change for anybody who's followed the analytics market at all
02:00 - 02:30 analytics has traditionally been about collecting and storing data then using it to make better business decisions that's still incredibly important but now data is much more than that it's now at the heart of new digital transformation so new business models built more directly on data than ever before to either improve the customer experience or quite literally to sell data directly to new customers in new ways this is putting new requirements on the type of people that have to be
02:30 - 03:00 working with the technologies that you all represent people that need to be more technical more business savvy two of the words that have come up several times today have been speed and collaboration how do you see those two words being central to the new type of person that is working with these tools I think it's a really good point I think when you talk about digital transformation as Timo just did we're really talking about cross collaboration between different functional groups within an organization I think it was MIT Sloan did a survey and there was
03:00 - 03:30 something like 70% of the organizations that have successfully achieved digital transformation do it with cross-functional teams so they need to collaborate there needs to be an open transparent process but we need to engage the masses so we're not just talking about data scientists we're talking about analysts bringing in information consumers knowledge workers and making them part of that process I think one of the things has happened is I want to come at it from their perspective which is to start with the business problems first because Ron just looking at data for data's sake and
03:30 - 04:00 saying well what value is there in that data FICO we try to get the clients think about the business challenges they've got and then looking at the data to see can they did their be used to help address some of the business challenges they caught and so for them digital transformation in many cases many comes is still on the journey think manufacturing is is struggling struggling because there's a shortage of data sciences there's a shortage of experts at I spoke through a couple of
04:00 - 04:30 very large manufacturers in this part of the world and they sons we can't get data scientists it's so desperately short of them so I think you need not just data scientists but I think you also need business analysts need people who understand how to translate the real problems they're having in terms that data scientists business analysts and those working with data in IT will understand and it's the culmination collaboration between those different groups that I think helps to bring some
04:30 - 05:00 value that's what we're seeing it's a long slow process it has a long way to go I think this event is great in it in raising awareness I think we're still very much in raising the awareness stakes for many companies so you mentioned that it's hard to get a hold of data scientists and that these days scientists need to collaborate with the business in a certain way you know something we touched on in an earlier panel is the challenge that businesses
05:00 - 05:30 face in even formulating a data science question I mean if they're not experts at the domain of well yeah we don't even have to explain it in complex terms if you don't know what's possible you can't conceive of a solution making use of a certain technology so how do you guys see businesses coping with with the need to formulate it in terms of data science but that technology is changing so rapidly it brings it back to that point
05:30 - 06:00 you said about speed and agility what's really it's really the fundamentally the biggest change is we've always known how to use data effectively to run the business better but it was typically a very slow process taking many months we don't have time that and to do that in the modern world but also people as you said don't know what they don't know don't know what they want so the big change has been this ability to iterate in more agile cycles taking just a few hours or a few weeks how about this how
06:00 - 06:30 about this did you mean this we now have the technologies that allow you to do that much more effectively in the past notably with the cloud and so that's having a much very powerful change in how businesspeople appreciate the abilities of data so the tea it's not just that people used to be slow it's that these new technologies like the cloud are enabling people to be faster than they could before what could I say that these technologies are not new what summarization technology is not new
06:30 - 07:00 neither is machine learning or AI I mean FICO he just celebrated our 25th year of machine learning expertise so they're not new but I think what we're seeing is is tooling develop around some of these technologies it's allowing it now to be consumed by companies yeah they consume by debt assign so an algorithms are not new someone new but most of them are not new but access to huge amounts of data that you can spin up in the cloud very quickly little power to crunch through
07:00 - 07:30 all of that data with those algorithms that's all very very new and that's there's been a huge tipping point just in the last year where it's gone from theoretical to absolutely real and and accelerating but I think the exciting point though is that democratization so David you talked about almost about literacy within the organization this is now enabling users to be able to grasp the terms and actually the applications in a language they understand so maybe moving away from nouns where we're saying we want to do a support vector
07:30 - 08:00 machine or a neural network to I want a predictor value I want to find a category I want to find something that's similar to something else those are terms that regular business users understand and they can formulate their problem in that language and now data scientists are almost getting in the way of progress in some cases on the one hand they're expensive they're hard to find they need to be very skilled so there just aren't enough of them the same time though there is an attitude with like well I'm going to take the data and show you the perfect model whereas increasingly we do need to democratize it and it's being built into
08:00 - 08:30 business applications for example where the scope is very well known you know the decision you need to take you've got lots of high quality data so you don't need all of those data scientist skills for each new project you can do it and just provide it yeah so I think one of the things that probably as vendors we all need to learn is its how to properly engage with some of these clients because I've done a lot of worried manufacturers and do you know what they're just not interested in turning down their operations to talk to
08:30 - 09:00 you about working with their data they're like throwing huge amounts of data away being the sensor technologies improve dramatically many sensors now are doing analytics onboard for example so getting data and getting it put in and use that expression I don't like but data lake yeah let's just say they're collecting data and storing it somewhere that I think is least to the problems I think the challenge then is well what do we use do we know what we're looking for what there are things that are just we
09:00 - 09:30 just don't know what some of the problems are right we can see the problems out in the other products in the field but are they due to problems in the way we make products are they to do with challenges we're having with processes about the way make our products or something else so we have to learn to be flexible and have have talling that allows you to look in those sorts of challenges in an unsupervised way so that you can find things that we just don't even know if you know you're
09:30 - 10:00 looking for something looking for a supervised fashion then I think all of our companies have got the right sort of data science experts that can do that but I think companies are certainly manufacturers are going strongly down the digital twin route that's exactly what they go in that room so they can be safe they can replicate an environment and use that as a just try and showcase any and the anomalies and show where they might be able to improve their operations so we're seeing the digital train route is quite interesting manufacturers it's a low-risk way I
10:00 - 10:30 think that's what we have to look for we're going to bring this technology on faster and you mentioned speed custom companies do want solutions quicker and they want to be able to user solutions as business users they don't want to be expert so don't want a tool that means they've got to be a data science expert or not tomorrow's Asian expert it's gonna be something very usable for them that gives some business value you know they don't necessary want to change at all sorry us no no no I agree I just think fundamentally all of our
10:30 - 11:00 companies are here today to try and sell to people the idea that companies want to run more experiments they want to do it at that data science level they want to look at telemetry from Internet of Things and sensor devices but they also need to run business experiments that can be running that agile way to know that you mentioned so that now means getting closer to a business problem continually delivering and being able to talk to a product owner of the of the objective and refine as they understand over time I think it was Jeff Bezos that said you know an experiment that you know is gonna succeed isn't an experiment at all there is gonna be an element of failure and we need to do
11:00 - 11:30 this in a safe productive environment it's actually one of the big changes for vendors like ourselves is that sa P is known to be a valuable product with typically relatively expensive but to do this kind of experiment people say well no we're not going to pay for a big upgrade before we see any results we want to buy and small experimental packages so that's what we've been doing is providing like focused smaller solutions where people can experiment with a bundle of products so they can
11:30 - 12:00 adapt over time so it really is about digital transformation for us as well adapting how we're selling these technologies so people consume them in easier ways mm-hmm I think if you go back five years or more if you are deploying an optimization application you would model the application I don't think too much is changed it turns the modeling and the speed of modeling what's changed is our the the tooling that allows users to use the application
12:00 - 12:30 whereas what's changed and I think now we're used to seeing proof of concept that in a matter of a few weeks where you're demonstrating value very quickly yeah rather than in months or even years you know we often have to wait for a first iteration after six months and if that worked you didn't know those duration under six months later I think those I think we're seeing the end of those days and you've seen you were seen with our in the pretty advanced its world we're seeing with pot new
12:30 - 13:00 technologies like Python so these these new techniques are all moving a user interface development forward very rapidly and and tools that are able to work with these in some cases open source technologies a range of technologies your tools need to be open they're the ones that I think are gonna succeed at the same time companies are really struggling to change the way they work in order to take full advantage often there's a small department whose job it is to go and do some innovation
13:00 - 13:30 but they rapidly run into the rest of the big corporate machine that is just not designed around that kind of most companies are not designed around agile flexible iterative innovation so it's about much more than technology they're risk-averse on they companies are risk averse if they see risk they smell risk they step away from it well I think which is necessarily a down bad thing but we're really talking about making the risks bite-sized yes if you take enough small bite-sized risks you're
13:30 - 14:00 actually optimizing your opportunity of finding the big net fair well I heard a nice anecdote about why there is a need for conservative and for progressive people and that is that you have the progressives to start up the new companies and try risky ideas and then when you want it to run on Rails you bring in the people that you know have best practice and then you they run it conservatively but some of these technologies that are changing and are disruptive right now are non mainstream and the business advantage you're getting from using something non
14:00 - 14:30 mainstream is palpable to say the least well nobody's using it so I'll call that out so what do we mean by non mainstream I honestly believe that open source is one a gigantic part of this battle so our Python there now established utilities to build data science applications on top of true it's it's certain there we understand what it is and we understand what it does it's now whether it can be deployed into a culture that needs to mature blockchain for example I would argue that a blockchain is not mainstream yeah so you're not in business but then it's not
14:30 - 15:00 necessarily shown message massive advantages for traditional companies yet either no I think it's because it's such a hype thing that people aren't using it for what it was designed for it to begin with they're using it like a database yeah but there'll be a tipping point really really soon so for example in the banking world ripple being that blockchain to to help replace bits of the Swift Network for payments and get closer to real time when that tips other organizations are there verticals will really start to see that benefit we have some very real solutions in the pipeline notably track and trace for pharmaceuticals you want
15:00 - 15:30 immutable chain of knowing exactly where your drugs are at any moment so I think with many companies maybe the ones that have used BI in the past what where do we take them on I think the road we take them on we help them to understand what that wider analytics picture looks like now that road right from descriptive what is through predictive at the use of optimization you know and at each stage you can show how you can benefit with a
15:30 - 16:00 certain level of automation what we can show is that the more of the analytics pie they use more complete your decisions get and I think that's really important and I think customers are many customers have maybe done think bi is pretty well so now it's 80% of businesses use bi whereas they once you get down to predictive it's like it's less than 50% then we get to optimization it's more like less than 25% so you can still see
16:00 - 16:30 there's a there's a long way to go in terms of getting this these technologies accessible and being used by businesses but I think we have a responsibility in the business to show customers the consequences of not going down that road and that is the competitive branches they're going to lose because they're in all markets there are leaders there are those that follow what's a good way to quantify that advantage because that seems to be something that you know oil and gas and others that have a tangible
16:30 - 17:00 foundation they don't they don't have that fire in their butts right now from knowing that well either that there's a lost revenue stream or that there's perhaps some disruptive startup they could change change the shape of things so I see this coming down to a maturity in say the chief data officer within an organization and if some research is sort of emerging in this space but we might call it in phenomics so the idea of treating information as an asset so you can use analytics and defense or in offense modes defenses risk mitigation optimization of
17:00 - 17:30 costs that kind of thing we use it in offense mode where we're trying to monetize our data maybe create data products or actually embed the analytic models we produce to the public make them customer-facing make them more visible and I think that's something that's really gonna emerge over the next couple years yeah I was gonna say I completely agree the first step to take with these new technologies is a big step back at one level and to think again about the big picture because what happened is that we're used to having processes and businesses that generate
17:30 - 18:00 data and we use that data to generate the processes but now we've almost flipped it the other way around in that the data is being used to create new processes I think about the customer experience it used to be sort of fairly linear I'd see an ad I'd go to a store I'd buy the product and lots of people would be doing the same thing now everybody expects a very unique customer experience but guess what that's powered by data so each customer is getting a very unique set of offers and channels
18:00 - 18:30 based on the context their demographics so that each customer is essentially following their own individualized customized process powered by analytics and it's so it's a big change because that's part of what customers companies are selling it's part of the customer experience so the business people are more interested in data than they've ever been in the past because it's not just about I've got some reports of my profitability by-product this is
18:30 - 19:00 something that I need to iterate in near-real-time because I can affect a decision I can change a customer's behavior by using they traded guess tumor attention for example classic example using algorithms now we can we have something called a customer a pulse so like a pulse owner in a hospital we can see if there's something out of the ordinary with their their interactions with us over the last 40 days 100 days and if there's something that's out of the ordinary can predict whether that means it's going to turn into a customer turn on earth so that's an area I'm
19:00 - 19:30 fascinated in so the idea that as we get closer to real time as we get closer to streaming so that could be sensors or that could be individuals on that personalised journey and on detection yeah understanding when something's about to go wrong or has just gone wrong and with digital transformation we're starting to see I think these prescriptive analytics getting embedded directly in the process so for example machinery can now phone in and look itself a service purely based on the fact it's likely to fail in the next X number of hours right we're currently working on an application with called adaptive scheduling where actually it's involving both the use of
19:30 - 20:00 prescriptive optimization and some predictive analytics to have a constant feedback loop to constantly improve the processes automatically so as you as you submit your factory to new production constraints or new demands the models will produce the data and reopen and that will reauthorize the operation continuously and that's so that is something I think is going to be the future for for the process industries
20:00 - 20:30 we're seeing similar sorts of challenges in additive manufacturing as well and something else I wanted to add work we're thinking about the nuts and bolts of how companies use the technology maybe also we need to think about what's driving that so the sort of things that we're seeing for example in in in retail just as an example and in telco where you're seeing now the desire to use personalization services using digital products right that's gonna change just
20:30 - 21:00 think about that in retail for example about the potential there for using analytics to personalize services for individual clients so now we're thinking about more the revenue end obviously what customers are interested in so again they're they're saying okay so these are transformations and apples be competitive and computes at costs in exactly the way that you say in a negative not negative defensive defensive app I want to pass away what where can it really help us
21:00 - 21:30 now improve what we do for our customers so for example we're talking to one client at the moment that has there are white goods manufacturer and they've got reliability problems with things like washing machines and tumble dryers in the field that they think are related to something they're doing in production so they'll they want to look at the data to try and find is there is there a problem it's like an unknown we're looking for right but again it's related to this it's costing
21:30 - 22:00 the money to keep going back servicing the same machines is that component it was it we're we're manufacturing a piece of fabrication what is it so it being driven by the revenue side being driven by the business in that way I think is going to be one of one of the big drivers that sometimes we ignore we think others data there's push there's push from the data side because we've got all this data there's a lot of pull from some customers as well they're going to drag this kicking and screaming this digital transformation process
22:00 - 22:30 they're going to have to accelerate it so that we're sitting here in five years thinking you know what's the next revolution gonna be alright I mean what are the technology enabled us for that process we're talking cloud we mentioned that earlier on so we're talking about effectively unlimited compute easy storage all of us as analytic vendors need to be aware of that and make sure that that's really part of that solution as well so we can not only store the telemetry but we can run the analytics in exactly that same environment and I think the big takeaway for prescriptive analytics is that as you said every
22:30 - 23:00 product service and internal process we now have the possibility of it being automatically better over time and as more people use it right right we're used to looking at a process gathering data analyzing it and a human being coming along and seeing how where they could tweak it with machine learning a lot more of that can happen at automatically in almost every aspect our business from finance to HR to analytics itself right well I'm glad
23:00 - 23:30 we're circling back to prescriptive analytics because I'll contend your point that this is not something new there are a couple of algorithms that have been you know developing and being refined in the past 10 years that are able to do things that were fairly unimaginable only 10 years ago and if you look at how robotics is handled today a lot of it is still programmed with signals and systems algorithms manually rather at yeah so for example the way a joint moves in an arm there
23:30 - 24:00 are some singularities where no matter what currents you apply it won't Lock itself and you need to remove those manually and it's all signals and systems work yes that can be handled by deep learning much much better it be optimized sure and so what kind of an advantage wouldn't a company have if they used this entirely new so this enables a new kind of business it's not just the refinement of what existed before what's made a big difference is processing power so precious empire has made a huge difference to the amount of the techniques linear programming
24:00 - 24:30 techniques yes they've improved but the technology the core technology being around a lot so your neural networks go back to my 150 you couldn't train a bigger network there was a look deeper than five labor well exactly so but it wasn't the algorithm was so so so you didn't have activation and the tools around the algorithms right it's more things fast it's the data you need an enormous volume of data to train a deep
24:30 - 25:00 learning system and that's also what's been able with that telemetry well the EM nest data set was what word propelled image recognition in deep learning and I was a bunch of students that they asked to write a bunch of numbers down he seems important to me for example and if you if you build an optimization model and it's not right it evens this very simple error it will run very slowly and often just changing something fairly simple and fundamental or tuning and tuning the
25:00 - 25:30 the the way the application runs in a better way what will give you big a big uplifting performance yeah so there's a lot of that there's a lot of that sort of road testing if you like of applications most most companies now they're looking seriously at optimization they want to evaluate evaluates first evaluate first it's still very much starting at the operations research level now they're guys that want to come in and test all right to do it it's fun if you got to come in test let us know we're here to help and every I think every time we we
25:30 - 26:00 go through an evaluation process we come out of it pretty well so the technologies let me follow up on this we're talking about accessibility to data scientists and the delegates are wondering let's see here we had a nice one what is the relationship between coders and clickers so our people do we need people B programmers in order to take advantage of these technology yes I think all those folks is the right answer speaking from Alteryx perspective where you know we are a code free and a code friendly solution so working with analyst and data scientist they work
26:00 - 26:30 nicely together in the right environment so in terms of have we made life easier for the clickers absolutely there is more power at their fingertips than ever before drag-and-drop connected up get the results but the key is interpret ability and explain ability there's no sense using a black box if you don't understand the number that's coming out of it in most regulated environments you can't put that thing into production from a code friendly point of view the coders we want to bridge that gap so I see data science organizations sitting in ivory towers they struggle to get their work operational and used by the
26:30 - 27:00 rest of the business then what's the biggest hurdle so number one they don't have the right support structure they don't have the right chain in the organization sponsorship exactly at that level but number two they don't communicate very well outside the ivory tower so they build things they get very very excited about their results but the analysts on the ground don't get to use their contents so by taking coding putting it inside a code free environment we start to get feedback and we start to get iteration and everything improves the executive sponsorship is something people have mentioned several times today what kind of executive sponsorship do
27:00 - 27:30 you need to put okay now we need to have a CEO we need to have a center of excellence like what kind of executive sponsorship is necessary to get these off the ground somebody that cares about data honestly it's as simple as that if you have a chief executive that will for example only look at the reports and data generated by the internal analytics systems and will refuse to look at the spreadsheets they've been hacked together by people with all of the exceptions that alone is worth its
27:30 - 28:00 weight in gold in getting the rest of the organization to align in actually making that analytic system high data quality easy to use and so on I think as well as a sort of view that I think customers Harries that I think most executive c-level people in businesses they're so busy running their businesses successfully I don't think that they necessarily want to know about data science or think they want to know and understand that use
28:00 - 28:30 productively can help bring business value to the organization I think I think they I don't think they've changed I think they need to know what bad it was gonna bring I can imagine their gut feeling has served them well in the past yeah so what what are convincing arguments and you're all representing technology companies what kind of things can you add to somebody the head of a conglomerate that feels that there are ecommerce or no there there brick-and-mortar stores are doing just fine the way they are now we mentioned
28:30 - 29:00 Toys R Us was shutting down at the earlier panelists I don't think they were nervous a couple of years ago but perhaps they should have been what if what a convincing argument I'll be honest I don't meet many of those people anymore there used to be more of them now it's typically the head of a small or medium-sized family-owned company that's it pretty much the only time you find somebody who doesn't care about data now there's been an entire generation today's CIO CEOs the first generation of CIO CEOs that grew up with PCs in the home they take computing for
29:00 - 29:30 granted yeah and I'm gonna say I'm gonna say the same with chief greater offices as well so generation one of that CDO probably the last five years we're really talking about just cleaning up the mess it might be master data management it's getting your reference data in order is setting the groundwork to do better things the second generation is starting to come through and this is now where we're seeing advanced analytics and data-driven experiments leading the agenda so execs aren't as blind as they were a few years ago and they're not as trusting as they were a few years ago they're gonna run the experiment and they're gonna get multiple points of view there's much
29:30 - 30:00 greater maturity and awareness of data so to answer your question the relationship between the coders and the clickers is much better than it's ever been because there has been an absolutely determined effort at bringing them together around these new themes and both try to get it I think when it's get to word as well I mean a question I was asked recently was well look our businesses has moved 50 percent to the into the Internet in the last four years and the way that's completed SWAT to my
30:00 - 30:30 supply chain I've got I've got distribution centers that serve my my online business and dis be Center to serve my stores business and they're completely different they're driven by a different dynamic how can I put them together right to get to to give me benefit because at the moment they both seem far too costly that's the sort of conversation you're gonna have with somewhere in your business and
30:30 - 31:00 those are sort problems they're gonna present you with now it is possible to solve that using the technologies that our organizations have but I think you have to have a competition at that level and it's a sight to provocative conversation right because you know comes that get to grips with it first they're gonna get competitive advantage I'm gonna talk about Amazon not pushing Amazon but Amazon are one of the biggest consumers of analytics they use optimization widely as well
31:00 - 31:30 and they are now been seeing I think they're the second largest company now I think in America that could become in America but they are not frightened to try and embrace new technologies in their organization they've got a supply chain that while the companies would be very keen to get their hands on and they paid lots of things they're even using analytics to anticipate when we might buy something and Stocking for it and there and they've painted those sorts of applications so we should be looking
31:30 - 32:00 that's our successful businesses being run out and lots of people are looking at that as a potential model but then you think about the supply chain and then suddenly it's had that neck back one stays further then to your manufacturing so if your supply chain changes all the signals and where that supply she doesn't bring chain all the way so the way that the signals about the way that supply chain runs change then the manufacturers have also got to get instead as well otherwise can't help but wonder if Walmart get to work if Walmart and the others saw Amazon as a
32:00 - 32:30 threat in 2005 and thought that they should start applying these principles as well yeah nobody even saw AWS as being a threat when it came out it was laughed out of the room several times back in the last decade you know and look at it now it's gonna be moving Amazon towards being a trillion dollar business speaking of the cloud many cloud providers meaning Amazon and Microsoft provide really nice and and solutions that was a job there really provides really nice and end-to-end solutions that they mark
32:30 - 33:00 it as we handle everything from the data ingestion to you know your Excel and all the Hadoop and everything in between what do you say to customers who talk about software software as a service in the cloud switch ON switch off and that kind of thing where are you other technologies fit in today I'll take it first then so I think Amazon and and Microsoft offer amazing utilities I question whether or not they're a completely integrated platform because a
33:00 - 33:30 lot of them don't necessarily look nicely together so that interconnectivity is what really defines the platform for me and also the openness to bring partners into that ecosystem to work with the platform to enrich it is what makes the difference so I think they've got some great capabilities and we love building off many of their utilities but is it that integrated platform that's gonna speak to a line of business analysts I would disagree with today's that hmm I think it's a lot for coming to swallow the end-to-end story and I think you need to start somewhere I'm biased on tight
33:30 - 34:00 because I'm folks not summarization but that's it's a good starting point and one of the things that we're seeing as you stop that with optimization particularly saying am in a production environment or near logistics operation is that you can then very easily take the next step which is to say well maybe five years ago you use demand data your demand data to optimize the way you you bring your plant or you where you you
34:00 - 34:30 run your or your machine processes actually what we can add there's a lot more data available now so maybe you can take some enriched data or tell you what let's build some predictive models and actually do some predictive analytics before we actually optimize someone use a classic one in a plant and there's a there are solutions out there doing this today which are pretty maintenance so you've got thousands of machines in a plant and you want to understand when
34:30 - 35:00 those machines possibly could fail now there are solutions out there at the moment we're being offered we're coming off of the sensor technology they'll collect the data they'll do some very very raw analytics on it and it will come out a metric that says a traffic light system I've seen traffic light systems red green or yellow yes that machines working okay if it's green if it's art if it's yellow then maintain it someone else saying maintain your
35:00 - 35:30 machine in six weeks or whatever whatever it is but we're not seeing a huge amount of granularity right and detail in what's provided and we're also not seeing the next stage which is it's not just enough to know when your machine's going to fail you've got thousands of machines in a modern plant right so you need a an optimized plan as to when you're gonna maintain them yeah okay so if you start with this idea that we can use optimization for better
35:30 - 36:00 planning and you can look at multiple time frames for planning then augment it with predictive models augment it with some of this data and you can suddenly start to widen the the surveys of influence then the next question then is okay well maybe that's your now before we actually predict what might happen here I bet if you actually be more selective about about the data we're actually using an objective models so let's apply let's fast some banks of
36:00 - 36:30 business rules to take this massive data and be more selective stream it it's been more selective about the data we think we're going to use so each time it's about enriching at each stage you are enriching the output so the bottom knee gets to optimizations using a very very rich data set and your optimization then so much better and so the whole thing gives you better support but their systems better decision support throughout the process it also completes the picture I think that's just my
36:30 - 37:00 opinion of a good way to start and that sense the only way to start you could start the other end you could start by enhancing your bi it doesn't matter but I don't want to price that we thumb works we should focus on that again the title of this panel is from bi to prescriptive analytics and you could say that there is a step in between bi or rather prescriptive analytics could be divided into two steps one is what has happened and two is what should I do about it and then you can automate that as well if you're if you're fairly confident that's the what should I do about it is is any good and you can do
37:00 - 37:30 prescriptive analytics without having done bi at all well my question was going to be I think a lot of companies are under the impression that you have to lift the whole company at once like the ride you feel they have to lift the entire company at once or can you do different so here you're feel free to debate I I say you can start wherever you want it you know you across the spectrum you can start wherever you want and the key thing is that if you've got if you've got tools that can embrace
37:30 - 38:00 maybe bi predictive an optimization similar tools that your business users can use as well that's a help so what you want to avoid is having completely different tools an algorithm so every single stage of the fat of the process so that companies feel like it's taking too much on and I think we've successfully sold too many out summarization technology to many companies but we now find that some of those companies are now coming back and
38:00 - 38:30 they're saying okay we've got that what else can you do and I take a step back I'll actually go back even before bi and say do you even understand what your analysts and data scientists are accessing so why don't one of the concerns that my customers really have is around gray data so if I work in a financial services company I'm looking at a dashboard as an executive what's sourcing this data is it coming from one of my enterprise data warehouses from the lake from somewhere else or is it somebody's spreadsheet that they've manipulated and it's somehow flowed into the enterprise context so curating cataloging bringing in I guess tribal
38:30 - 39:00 knowledge from across the organization and then telling people what can be trusted and its lineage that's the number one starting point for me so get that right across the board automated as much as possible sound foundation for bi predicts for that foundation needs to be in place before the other ones I'm saying you can definitely have behind you can have prescriptive without it I'm just saying you have a stronger foundation if you understand the assets I'd say so bi is fundamentally about people looking at data and making big strategic decisions and because it's about people the biggest barrier is not
39:00 - 39:30 the technology its culture but it has been for a long time prescriptive analytics can truly be much more lower-level and automated and still provide massive value for example we did a project with a large chemicals company they for an invoice matching so they send out an invoice they get a payment back they get that information from the bank 70% of the time it doesn't match the reference numbers the difference there's two payments for one invoice two invoices for one payment and so it has
39:30 - 40:00 to go to a big room where there's lots of people shuffling papers around trying to figure out which invoice goes where we so it was a well good that one company they got it up to 70% with a whole bunch of hard-coded rules that need maintaining but with machine learning applied to the same opportunity we got it to 94% within two weeks and this is a organization that has hundreds of thousands of invoices so that's a massive savings and time money and effort but nothing to do with bi really
40:00 - 40:30 not how we think of it right but absolutely prescriptive analytics using machine learning on a complex data set to make optimize every one thing that's it and so the other is you gots 94% but it's still improving because every time there's an exception it gets kicked out to a person that says this is the right answer the algorithm learns from it so again it's one of those self optimizing processes that's getting better automatically over time so I'd like to expand on that 94% thing is I think a lot of people are under the impression that a solution are a machine learning
40:30 - 41:00 algorithm or some prescription needs to be perfect in order for it to be valuable yep I think this out really one of the big challenges actually in terms of communicating what you've built as a data scientist so there is going to be a confidence level and I don't think us as an industry as as vendors do enough yet to communicate the uncertainty in our models so there's a level of empathy I think within the outlets that we produce well we need to tell people this is not a dead cert there is uncertainty here and you need to understand the risks you take if you take up the recommendation
41:00 - 41:30 or there's a walk before you run approach I think that we need to adopt the other thing we need to recognize is that the pace of business change for organizations is like it's never been before it's a continual to change so you might build a model that's fit for purpose now in six months that bit that business changes fundamentally it's no longer fit for purpose so again there needs to be flexibility in in being a mature modify and change your applications and so it's
41:30 - 42:00 getting harder and harder for what I call package vendor solutions yeah to survive I'd say well this is the opposite depending on how you define package so so to get the most value out of AI you want large quantities of high-quality data yep so you typically got that as part of a business application for example you want a nice tightly defined scope of decision a repetitive complex decision that you're making hundreds of thousands of times a day supply chain optimization what product do I offer to a customer
42:00 - 42:30 that's just about to purchase something those kinds of decisions and then you want to ideally be able to take action directly as part of the business process so we were actually building a machine learning into everything we do that that invoice matching that's just one example of one small branch of Finance there are thousands of those that we can optimize Gartner believes that let's say half a billion users will save two hours a day a day this year thanks to AI power tools this is what they're talking about that's that's sexy that adds up to half
42:30 - 43:00 a million years of increased productivity this year alone but just that is am anything I was saying was hard coding is is no no no yes it absolutely needs to be the Flex is the DES is the discipline that's now spreading out of our world that continuous integration that right and then machine learning algorithms absolutely you know hard-coded there they're constantly learning air bottles and into bi and prescriptive analytics because I think when people hear DevOps they think IT systems whatever the term
43:00 - 43:30 ml ops or science ops whatever they call it right but the idea sounds not right the idea is it's those same disciplines we talked about we mentioned what is agile when it comes to an analyst right so it's just all about iteration it's about experimentation and it's about continuous test-driven development so whether you're right get that feedback faster in the cycle see my I have a question based on what you're just saying around you're predicting so many things at such a volume of such a velocity now how do you provide the user with a confidence on what you predict or do you do you analyze it afterwards in a
43:30 - 44:00 bi environment so every your every process where you're applying machine learning you're essentially outsourcing decision making to the machines yeah so it's absolutely essential to have safeguards governance checks so the most of the time we're using the the qualified people to train and guide the process the Machine the invoice matching so it starts by saying we think this is the answer with this amount of confidence are we right and the more
44:00 - 44:30 often we it says it's right we'd say well you know what would you just like us to do this automatically next time but it's finance its money it still has to go to audit you still have to be able to show what's happening so crucially domain expertise absolutely vital AdWords yeah could I just bring another fact that's important as well as we are I think we touch that we're not really developed yet and that is change management which I think we've probably all scenes stuff fail because people didn't want to embrace it I want to go biggest barrier when I go back to the cult of the collaboration
44:30 - 45:00 thing again change management is absolutely vital particular when you're changing processes it may well be better for the business but if they all go on strike for six months because they're not going to have that process anywhere near their plant right and that happens a lot is that many digital transformation is is worrying employees they're thinking am I going to have a job tomorrow is automation gonna spell the end of my career you like that I am busy so so change management to show that actually bringing these new
45:00 - 45:30 technologies will actually allow people to do different jobs more productive jobs better jobs I think is important it's next your buzz right we have to be honest there's gonna be some as good as some roles is there is some attrition definitely difficult it's mostly about displacing work rather than replacing exactly I like to think of it as augmented intelligence right so we get we're gonna allow he would be so much more productive by having systems of intelligence supporting that job yes cuz again it to go back to manufacturing what's the big problem in the UK it's
45:30 - 46:00 not do you care other countries of low productivity and low productivity in men in many cases are held back by the failure to embrace change rather than knowledge yes I'm not in furiously puri because of the reason it's not just manufacturing I'm gonna take an excel analyst somebody that's working twenty eight hours a week inside the spreadsheet new data comes in I'm gonna spend nine of those hours rebuilding that Sochi right look about low productivity old enemy spreadsheets McKinsey study said exactly the same thing their data processing data
46:00 - 46:30 collection is by far the most automatable area and using these music I think it's mentioning to the customers when they're asking why is it so expensive to hire a data scientist to fix this stuff that's because it's so valuable to get rid of those five hours per person per week doing all that work that's so valuable last question we have a couple of minutes left this is from bi to prescriptive analytics interpretability of the output is something we touched on a little bit and I'm of the opinion that some of these
46:30 - 47:00 new methods in deep learning are able to model problems that are perhaps more complicated than what we understand ourselves I think driving a helicopter is a good example we've never been able to manually guide a helicopter with signals and systems but we've trained models that have copied humans to do it it's because the domain is just so complicated and a lot of people say that the model is opaque but perhaps the business reality is opaque society is complicated your business is complicated
47:00 - 47:30 how do we deal with those environments in banks for instance when you're calculating risk it might not as be as simple as saying while you're wearing these pants and so your risk is five I I wouldn't let algorithms anywhere near that level of complexity that's just not what they're best at and there's so many massive opportunities with far simpler more repeatable processes that we do know that it's let's take the exhaust leave that to the researchers and that it's about giving the human being more
47:30 - 48:00 expertise more more knowledge to make a better decision it's decision support ultimately with the the idea is not yes we can automate mundane tasks but in areas like that risk you've still got all of the the the human being the ability to be able to decide at the end of the day and I think any decent application will allow many levers to be changed and pulled to make sure that you can shape and understand the problem present it in the way in which you want
48:00 - 48:30 to do it and it so there's a massive ethical connection yeah anytime you use these algorithms that touch human experience there's huge huge dangers I recommend a Booker weapons of math describe destruction yeah anybody was in this industry that hasn't read this kind of book or these kinds of articles absolutely sure because it's gonna be essential in the future we've seen Watson I think with the Chicago Police Department discriminating out-of-the-box based on the data we've got to be so so careful right thank you very much
48:30 - 49:00 [Music]