Data Science in Keboola Connection

 

Why should you use ready-made data science applications in Keboola application marketplace?

Maybe because your company doesn’t have an in-house data scientist to write a custom code.
Or perhaps because the problem you are trying to solve doesn’t even require the magic hands of a data science wizard and, in no time, you will find the answer among the solutions we offer.
The best part? The apps with Keboola logo are completely free to use! For partner applications, fees may apply.
Let’s have a look at what the practical use of Data Science can bring to your day-to-day business:

Anomaly Detection
Anomaly Detection

What is it? This application can instantly find anomalies or outlying values. It is typically used in time series to uncover irregular events and also one of the most widely used techniques in data mining.

Where is it used? In a nutshell, everywhere. It can help analyse consumer behaviour on the website or in the store, by highlighting unusual behaviour such as interrupted shopping via abandoned carts in e-commerce, pointing to a larger issue. It can also be used to help with fraud prevention or even in healthcare. In Advertising, it can help notify you about spikes in spending across your advertising channels.

What do I need to run it? All you need to do is upload a table to Keboola storage which has two columns of data with time records and metric you want to follow.

What does the output look like? The output is simply a table highlighting records with exceptionally high or low values. This can then be automatically exported to a Business Intelligence tools of choice and populate otherwise empty section of a dashboard as a means of alert, all in very near real-time basis.

How long to configure, run, and test it? If your data is already in Keboola Connection, it will only take a moment.

Copy-of-keboola-kolecko-copy-NEW-800px

Want to know more? Contact us!
Basket Analysis
Basket Analysis

What is it? This algorithm helps find items or actions taken together. From this place it is much easier to predict action and recommend the next most likely step for your user/client.

Where is it used? This technique is most commonly used in e-commerce and any type of retail business. For example, when you place an item into the basket in an e-store, you will instantly see a customised offering based on the data of what other people had bought with the same item you are about to purchase. Naturally, this drives upsales and increases basket size and its value.

What do I need to run it? On input, you will need a table with transactions, identifier of transaction, columns to segment the table by and a value to express the minimum significant frequency.

How long to configure, run, and test it? If your data is already in Keboola Connection, it will only take a moment.

Copy-of-keboola-kolecko-copy-NEW-800px

Want to know more? Contact us!
Correlation Finder
Correlation Finder

What is it? Correlation finder algorithm scrutinizes your data to quickly reveal whether or not it is feasible to look for dependencies within your data.

Where is it used? Correlation is one of the most broadly used data mining functions. It can help to identify linkage in consumer behaviour. For example, you look for correlation in sales and weather trends or website up-time and trend in subscriptions to a service. When using this, please beware of spurious correlation.

What do I need to run it? Simply point the app to any table in the storage and define the column which contains the segments.

What does the output look like? This app produces a list of .png Correlogram chart.

How long to configure, run, and test it? If your data is already in Keboola Connection, it will only take a moment.

Copy-of-keboola-kolecko-copy-NEW-800px

Want to know more? Contact us!
Data Type Assistant
Data Type Assistant

What is it? This application is an enormous help to analysts and data scientists. It will take a good look on your data and will automatically find the majority data types within the columns (date, number, time). If there are errors and incorrect values in data based on the identified data type, this app will find them and replace them with the correct data type values. You can see this application as a simple data quality enhancement procedure.

What do I need to run it? You need an arbitrary table with the very minimum of 100 rows; the bigger, the better.

What does the output look like? The app will return, upon your preference, the same table you provide with either error rows removed or NULL value inserted where the error occurs.

Copy-of-keboola-kolecko-copy-NEW-800px

Want to know more? Contact us!
Grouped Histogram
Grouped Histogram

What is it? Histogram app automatically finds parts of the data which appear to contain statistical similarity. It creates segments with similar statistical properties. These can then be further analysed separately.

Where is it used? In retail, for example. Imagine a table of data collected in store: number_of_items a customer bought with customer’s age and shoe size being the properties of each customer while in the last column you have time_spent_in_shop as the number of seconds spent in the store with segments column being the age. This app will enable you to test different hypothesis about your customer’s behaviour segmented to different age groups.

What do I need to run it? Point to a table in your storage that contains at least two columns, where one column represents behaviour and one column can be used for segments. Both columns need to be numeric.

What does the output look like? The output is a table loaded to the storage of Keboola Connection containing segment boundaries. The segments will never overlap. Visualisation can be in a graph.

Copy-of-keboola-kolecko-copy-NEW-800px

Want to know more? Contact us!
Linear dependency finder
Linear dependency finder

What is it? This algorithm finds linear dependencies between all numeric columns and a designated target column.

Where is it used? You can use it to look for relationships and similar behaviours in two separate data sets. For example, imagine this being used on analysing sales/revenue and marketing spend to help prove causality and effect of marketing effort on business results.

What do I need to run it? You will need to point this up to a source table in your Keboola storage and mark the target (typically revenue figure) and segments columns.

What does the output look like? This app will produce scatter chart .png files which can be found in Keboola storage for further upload and analyses into the BI system of your choice.

Copy-of-keboola-kolecko-copy-NEW-800px

Want to know more? Contact us!
Next Event Prediction
Next Event Prediction

What is it? Pretty self-explanatory, this data science technique helps you look at a series of events within a group and anticipate the next most likely occurrence.

Where is it used? This has a great use case in B2C and B2B space where you can, for example, anticipate next customer’s purchase with personalised campaigns or micro-targeted promotions.

What do I need to run it? Upload a table from your CRM or point of sales (POS) system with orders, event identifier, group identifier and a date.

What does the output look like? The app produces a table where column types will contain customers segmented into groups, such as regular, irregular, occasional and random. This isn’t real prediction but rather a computation of plausible regularities. As a first step this is great but we recommend having the app customised to include calculation of confidence intervals for more precise results.

Copy-of-keboola-kolecko-copy-NEW-800px

Want to know more? Contact us!
Segmentation
Segmentation

What is it? Segmentation app finds a segments or ‘groups’ of similar rows in your data placed on scatter plot chart so it is easy to understand. It will work for multiple dimensions.

Where is it used? This is used to identify common properties of the data set based on all relevant attributes. For example, it can look at your customer’s transaction and reveal one segment where they most commonly buy your service 5x per year and also have in common shoe size 43. This can then be used for better campaign/communication targeting.

What do I need to run it? On input, this app just needs a table with customer ID and order ID (for the above described scenario) and other relevant attributes that can help define the segments. Optionally, you can specify the number of segments and the algorithm will outfit the output accordingly.

What does the output look like? The output will look exactly like the input.

Copy-of-keboola-kolecko-copy-NEW-800px

Want to know more? Contact us!
Table content overview
Table content overview

What is it? This app analyses the input table and infers the most suitable data type definitions based on the content of each column and detects possible invalid values.

Where is it used? Use this recipe if you need help defining data types but also wish to have an overview of potentially invalid values without doing alterations to the source table and also wish to have a deeper insight into the characteristics of the data contained in the data set.

What do I need to run it? Simply connect a table you want to analyse in input mapping.

What does the output look like? The recipe produces the following outputs: a table with a list of detected column data types, a table with potentially invalid values found in the table, a table containing the column characteristics and visual representation of the values/a picture for every column. The .png files are stored in the “File Uploads” section and tagged HistogramEstimate, ProbabilityPeaks and the list of the produced .png files is stored as finalResults in the default out bucket.

Copy-of-keboola-kolecko-copy-NEW-800px

Want to know more? Contact us!
Omni Channel marketing attribution (partner app)
Omni Channel marketing attribution (partner app)

What is it? Marketing attribution app is a set of algorithms which analyses your data from different marketing channels in order to monitor and compare the performance of each channel. It is able to look at your data from Facebook Ads, Adwords, Yahoo and any other display or video advertising platforms.

Where is it used? It is used by marketing teams to find the channels and campaigns within these channels that perform best conversions you have targeted as your business goals; this may be a sign up, upgrade to premium, purchase, or simply a webinar registration.

What do I need to run it? You need to connect data from marketing channels you wish to analyse and our algorithm takes care of the rest. It will even work out delayed effect of conversions as sometimes your customers take time to act upon being influenced by ads.

What does the output look like? The result is a budget allocation recommendation for following time period based on previous analysis. Keboola can further automate distribution of these recommendations directly to the advertising platforms enabling to act on it in fully automated manner.

How is it different to other attribution techniques? We put the emphasis on customer quality, so together with our other app for estimating CLV (customer lifetime value) we are able to help you plan your spends in the most effective way in order to attract only the best segment of customers (i.e. the highest spenders).

logo-nahled-4

Want to know more? Contact us!
Customer Churn Prediction (partner app)
Customer Churn Prediction (partner app)

What is it? This is a set of machine learning algorithms which help predict an end of customer life cycle in advance. This insight helps prevent the customer to churn.

Where is it used? It is hard and expensive to acquire new customers and even harder and more expensive to keep them. This predictive model will automatically look at behavioural and purchasing history and identify the customers that are most likely to leave in the next X months. This Information is then used to drive micro-targeted campaigns aimed at preventing the attrition.

What do I need to run it? This app utilises a set of CRM data, can work well with purchasing history and web behaviour or customer actions based on different marketing campaigns

How does the output look like? A very simple additional column with information about percentage probability of this customer churning within a set time frame.

logo-nahled-4

Want to know more? Contact us!
Recommendation engine (partner app)
Recommendation engine (partner app)

What is it? This app utilises a combination of several traditional recommendation algorithms (UBCF, IBCF, Deep Learning etc.); a combination which is most suitable for business context. It offers personalised recommendations for users or customers based on the purchase history of the entire client base.

Where is it used? We first used it for an e-commerce customer with 500K customers and more than 60K products. The model makes use of the bundle of customer’s past purchases as well as product searches. The key to the success is personalisation. Our client has great mobile app and email marketing execution so they can reach customers when they want to. Combining product recommendation together with known customer preferences serving as the filters has a direct positive impact on the campaign’s ROI.

What do I need to run it? This model works with information about products offered, the inventory and, of course, the transactional data about customers.

logo-nahled-4

Want to know more? Contact us!
RMF Analysis
RMF Analysis

What is it? This algorithm helps find items or actions taken together. From this place it is much easier to predict action and recommend the next most likely step for your user/client.

Where is it used? This technique is most commonly used in e-commerce and any type of retail business. For example, when you place an item into the basket in an e-store, you will instantly see a customised offering based on the data of what other people had bought with the same item you are about to purchase. Naturally, this drives upsales and increases basket size and its value.

What do I need to run it? On input, you will need a table with transactions, identifier of transaction, column to segment the table by and a value to express the minimum significant frequency.

How long to configure, run, and test it? If your data is already in Keboola Connection, it will only take a moment.

sormen-logo

Want to know more? Contact us!
Geenea Natural Language Processing
Geenea Natural Language Processing

What is it? Geenea is an app developed by a third party partner company and additional fees do apply. They specialised in analysing text-based data and can provide out-of-the-box as well as custom solution for: Sentiment analysis, Lemmatization, Entities Extraction, Hashtagging, Language detection, etc.

Where is it used? Since it can perform different tasks, there is no singular use case. The first example is to use it in research and understand comments your product users say around the social web and work the feedback in form of analysis of unstructured text which is personalised and fine-tuned to your industry’s vocabulary. The second example: You already analyse data from customer care – mostly numerical. What if on top of NPS we could also layer sentiment of the actual customer comments, detect rising issues before they erupt and more?

What do I need to run it? Unstructured text in any form will do. Just use Keboola Connection to connect the data from your social media monitoring tool or customer care console like Zendesk or Kayako. Have data in other form like PDF? We have solution for that too. Just talk to our team.

What does the output look like? The output will vary for each function of this analysis but essentially it will be a table which you will be able to use for more advanced analysis in Keboola, such as looking for correlations between business result and customer feedback sentiment/topics. You can also just straight up send it to your favourite BI tool.

geneea-logo

Want to know more? Contact us!