The Few Facts About SAP Leonardo, SAP HANA and Machine Learning

https://iytwist.blogspot.com/

THE FEW FACTS ABOUT SAP LEONARDO, SAP HANA and MACHINE LEARNING

The term machine learning is being used very frequently nowadays. The term is known to boost the salability of software. It appears more frequently across Google as well; the frequency has scaled by four times over past two years.

SAP now has a history of using technology terms that do not associate with their business at all. The terms could be related with collaboration, good UI design, IoT, marketplaces, inventory optimization, databases and HTML5. When SAP partners with a company, it highlights the partner company’s capabilities as its own.

Let us try and understand how SAP has used the term Machine Learning. SAP has started using the term machine learning all of a sudden. But the problem is that the way SAP presents machine learning is much different from how machine learning actually operates. They never used the term earlier. But they now use it frequently.

The best part is that as soon as SAP started using the term Machine Learning, they claimed to be market leaders in the technology. They came up with a video for their Leonardo product, in which they claim that they are the only company in the world that helps their customers by implementing machine learning.

The term is used very frequently by SAP, and it has resulted in SAP employees using the term for their skillsets over LinkedIn. A few of them have no background in statistics or forecasting. They may have a background in CRM, Hybris or Fiori, but now attach Machine Learning to their titles. This makes the combination of skillsets very interesting. This goes to show that SAP employees just add the term to their titles to make it look catchier.

There are cases wherein SAP functionalities based on ARIMA (The auto regressive integrated moving average algorithm) are categorized as Machine Language.It is difficult for someone knowledgeable about machine learning to understand why SAP has classified their functionalities as Machine Learning, because the methods they are using are not a part of Machine Learning.

ARIMA (The auto regressive integrated moving average algorithm) is a univariate statistical approach, which may be multivariate? SAP has come up with a video which clearly shows ARIMA univariate analysis in action, and the same has been classified as Machine learning.

Going by that particular video, SAP proposes that ML as a way that automates intelligent decision making within an organization. This very much sounds like Artificial Intelligence, but it is being expressed as Machine Learning.

The way they are expressing Machine Learning has nothing to do with Machine Learning. Machine Learning algorithms, on the contrary are very specific. It is analysts and data scientists who run them. These must be set up, and they must be run over a narrow space.

Another notable factor about SAPs claims about machine learning is that they bear a resemblance to earlier claims made about analytics. There are cases wherein nothing about SAPs claims is apparent in present day environments. They keep making claims with trending terms, time and again.

Let us do a quick check over predictive analytics claims by SAP. SAP makes claims in form of videos. They earlier used the terms predictive analytics and predictive algorithms. They now substitute it with machine learning. They keep coming up with modified sets of analytics solutions which are never implemented in practice.

SAP Lumira was initially cited to be a Tableau killer, but it has more or less disappeared from the markets. Even while there is still some interest in the product, it is on a decline.

Let us now consider the case of SAP HANA predictive analytics. Analytics that came with HANA never attained its potential. The companies that actually use HANA use it with BW, which is as complicated as HANA to operate. The initial idea was that S/4 HANA will be used for analytics as Embedded Analytics. Since Embedded Analytics never really developed on a large scale, it is only a few companies that use them.

As per SAP’s claims, it had ML Algorithms even prior to 2013, but used to call them Predictive Analytics. SAP has used terms from ML, such as K nearest neighbor, K -, C4.5 decision tree and ABC classification in their advertising videos. They are used together without mentioning ML, under Predictive Analytics.

SAP has therefore used ML in a way so as to rebrand predictive analytics. The screens that they show are fascinating in their videos. But what the clients receive has no similarity with this. The videos are best classified as confusing. Predictive Analytics are discussed with HANA in SAPs videos. HANA however is a database.

In general, as well, SAPs way of explaining things create confusion. Boundaries defined between different technological terms are blurred. When one listens to an SAP’s outlook of things, one knows less about a subject than one did before starting out. When SAP talks about analytics, they are not talking about the analytics application which stays over the database. This is similar to saying that the ERP one is running is running with Oracle 12c or IBM DB2.

In practice, it is Analytics and ERP that are the applications. A user works in the application. It is the application which sits over the database. SAP has over the previous years attempted to co-opt the term Machine Learning. This is not the only incident. The company has a history of co-opting in other areas as well. Earlier, by using SAP HANA, SAP used to pretend that they were better at databases, as compared to Microsoft, IBM and Oracle. Over a period of 7 years, we have come to realize that this is not true. Now SAP is trying to promote the notion that even thought they have zero history, even in linear regression, for machine learning, they are the best company.

SAP is comparing itself to SAS in Machine Learning and claiming that it is better than the latter. If we take a look at the history of SAS, they have been into mathematical programming, right since their inception. Mathematical programming is the very basis of machine learning.

When HANA was launched, SAP claimed it was better in databases as compared to Oracle. Now with machine learning, SAP is claiming that it is better in algorithms and statistics than SAS.

Let us resume the topic Machine Learning.

Machine Learning is a term which can be very misleading. For non ethical software vendors, it is a term which gives an opportunity to put forth a proposition for radical transformation using technology.Among all such vendors, SAP is a vendor which is the farthest from reality. Over recent times, they are discussing all their conquests regarding machine learning. This phase is only temporary. When a new trending term comes to fore, SAP will stop discussing Machine Learning and start talking about the new term.

In practice, Machine Learning is actually an entire series of mathematical approaches for prediction and data analysis. A majority of forecasting that is done in businesses in the present date is univariate. Sales history is an example of the same.

But Machine Learning involves use of multivariate data sets. So, sales history alone does not suffice. It should be more like Sales History + product family + economic factors.

Let us now try and understand how new machine learning is across different spheres.Hardware and software are improving at all times and their progress is continuous. The transformations are radical and the differences that come about over time are tremendous.

Around 20 years back, it may be possible that 3.5 pages of code and functions was required to be written, for something that requires just a single command in the present date. But if we observe ML, it is being put forth as something which is entirely new. This is in spite of the fact that regression is a factor that has been around for a significant time.

Just as an example, regression is available in SAP DP, which is a supply chain forecasting application. But it is not used very frequently, for the reason that it can be very complex for implementation in a business setting where improvements are expected over shorter times. Complex methods call for more time for training, which is already inadequate.

Let us now take a quick look at the environments which use SAS for forecasting.

While SAS is a high investment kind of a forecasting product, it is complex as well. Even while the SAS has another more standardized forecasting product, they essentially go beyond the same in terms of products. It is not this product that SAS is limited too. The industries that use SAS most frequently are insurance and finance. In these industries, the budgets are bigger as compared to other industries. There is also more time to lay the focus over producing complex forecasts for a smaller number of items.It is safe to conclude that Machine Learning has been around for some time now, even though a number of market vendors have now started using the term. This gives ML a higher profile than it used to have earlier.

Machine Learning undoubtedly offers an array of advantages for organizations, such that they are empowered to make predictions that pave way for making better business decisions. Knowing about Machine Learning gives an organization a better idea about where it should or should not be used.

Let us take a look at prime advantages of Machine Learning:

Machine learning is essentially a very powerful tool, which withholds within itself the potential to bring about a paradigm shift in the way things operate. Identification of trends and patterns is simplified with machine learning. It is used for analysis of significant amounts of data. With machine learning, one comes across trends and patterns that are otherwise difficult to identify.

Amazon uses ML to get an insight into purchase decisions along with browsing histories of its users. This helps them come up with deals and products relevant to the consumers and the company displays the same to the consumers when they go to the portal.

Another prime advantage of ML is that human involvement is minimal. As machines are empowered to learn by themselves, they can make predictions and improve algorithms as well. ML helps recognize spam. Antivirus software use ML to filter out new threats, as and when they are recognized. With experience, algorithms for ML enhance in efficiency. In weather forecasting, amount of data improves with time and empowers machines to make better forecasts. ML algorithms also handle multi variety and multi dimensional data for environments that are uncertain or dynamic.

It is also in e-commerce and healthcare industry that ML holds a vast potential for delivering a personalized experience for consumers and targeting the right consumer cross section. ML however comes with its own set of disadvantages as well:

For training on ML requires massive data sets. They should be unbiased and of a good quality. There are times wherein an organization may have to wait for new data sets to be generated in order to make ML operational. Apart from data, investment of time too may be required to make ML algorithms operational. It lets the algorithms develop to a degree such that they can implement their purpose. This may call for investment of resources and extra computer power.

Error susceptibility is another feature of Machine Learning which is noteworthy. In case the algorithm is trained with small data sets which are not fine enough to be included, the accuracy of the results of the algorithms may not be up to the mark. The results are then biased, as resulting from training which is not up to the mark.

As resulting from the same, commercials that are displayed for the consumers are not accurate. It may also occur that a chain of errors is initiated which goes undetected for long periods of time. This may further result in losses for an organization. Such problems take an even longer time to get corrected.

So, while machine learning can be the right tool to go for when used at right places and in the right ways, all avenues are not feasible for ML. An entity must be very cautious and prudent in putting ML to avail.