iPredictt Data Science Labs provides a range of services using Data Science tools
& techniques. Having pioneered use of Anomaly Detector, we help companies
understand business health from terabytes of data and identify immediate
actionables. We have also simplified complex analytical requirements through
a set of docketed modules of data science tools, namely iPredictt Core
and iPredictt Buzz.
We offer our products and services to a wide range of industries including Mobile apps (Churn analytics), E-commerce (Customer scoring), Telecom (Recommendation engine), Health (Predictive engine), Consumer goods (Sales predictions), HR (Prospect filtering using scoring techniques), Travel (Demand prediction).
Cross-industry process for data mining (CRISP-DM) provides a structured approach in planning a data mining project. It is a well defined, industry standard methodology which provides a uniform and scalable framework:
iPredictt uses CRISP-DM in all projects as it is a robust methodology and has well-proven results.
iPredictt has five products, which provide end-to-end solutions for businesses
across industries. These platforms provide a user-friendly interface with
detailed deep dive capability, allowing faster decision-making.
These flagship products form an embryonic state into fully functional and automated solutions directed towards client's business problems and industry.
Businesses face major problems while attributing their sales to multiple advertising channels. iPredictt solves all the below problems using our proprietary tool iPredictt MMM+. It allows data transformation to best fit non linear relationships between advertising variables and gauges the delay time before a channel is able to convert a campaign to sales.
iPredictt MMM+ comes in where traditional media mix models fail and adds a lot more value. Using cross channel multi-collinearity exercises, the model provides a multi dimensional view to the media ecosystem. The model we use is a white-box model, thus allowing managers to view and play with the statistically calculated channel weights. The final output is in form of a simulator which uses channel-wise estimates calculated through the model and combines them with baseline spend figures to generate realistic profit-loss scenarios. This provides an end to end view of the existing media ecosystem and prescribes on how to channelize efforts for a new marketing strategy or a new brand launch.
iPredictt MMM+ comprises of 6 stages starting from data collection to final model deployment.Industry standard data mining process called CRISP-DM is deployed for the project of building the MMM+.
Data is collected from multiple sources using big data tools such as Spark and Hadoop. Social Media sentiments (Perception data) are captured using product accelerator of iPredictt Buzztm
Each variable is handled separately by using a data cleaning snippet designed specifically for that variable. The objective of this snippet will be to:
1. Handle outliers
2. Do exploration and multi-collinearity exercise
3. Imputation of missing values using algorithms such as decision trees
Each media channel has a delay associated with it while returning sales. The amount of lag is analyzed and quantified here. Data is tested to check for relevant lag associated with the corresponding returns and corrects it in the raw data feed. This handles delay in returns which can vary from channel to channel.
Variable correlations are studied to eliminate variables having strong correlations so that they don’t bias the model. Variable importance is calculated using decision trees to understand which variables show high significance to the dependent variable (Net brand sales). Factor analysis using multi-dimensional scaling and PCA is done to prepare list of variables for modelling stage.
Generalized additive model (GAM) is an industry standard white box regression technique used for attribution exercises and driver analysis. This iPredictt model handles 2 main concerns of sales-advertising relationship:
1. Diminishing returns for higher advertising spends.
2. Delayed returns to marketing campaigns.
The equation of GAM is defined below:
𝑌 = 𝛽1 ∗ 𝑓(𝑎)t + 𝛽2 ∗ 𝑓(𝑎)t−1 + 𝛽3 ∗ 𝑓(𝑎)t−2 + 𝛽4 ∗ 𝑓(𝑏)t + 𝛽5 ∗ 𝑓(𝑏)t−1 + ...
β – Represents standard estimates
𝑓(𝑎)𝑡 - Represents transformed variable “a” from time “t”
The variable transformations help create non-linear relationship between advertising and sales. Following are a few examples of variable transformations used:
Base spends to each of the channels are collected and finalized with the relevant client POCs. Based on the variable weight matrix, the base spend value of these channels are inflated, ensuring the sum of the spending across channels sums up to the total media budget expected. Using Monte Carlo simulation, changes to these base values or weights is studied to better understand implications of new marketing strategies. The model will generate weights for each variable input, by calculating Beta values for each variable. The sum will then be scaled to the budget value, thus adjusting proportions of inflated values to cover 100% of given budget value.
Based on the model and the simulations, scenarios are presented which are most favourable and aim for maximizing sales. Channel-wise attribution is shared with stakeholders in form of an interactive dashboard as well as a simulator, allowing them to create their own scenarios. iPredictt then co-creates along with client market leaders to generate an action plan for media mix in order to maximize profit / awareness. The model creation is generally a 3-5 month exercise depending on magnitude and condition of available data. Once the platform and data lakes are created, any subsequent model iteration takes no more than 1-2 weeks to run.
iPredictt MMM+ captures accurately the relationship between advertising variables and gauges the delay time before a channel is able to convert a campaign to sales. The model provides a multi-dimensional, cross-channel view to the media ecosystem while being a white box solution. The final output is in form of a simulator which uses channel-wise estimates calculated through the model and combines them with baseline spend figures to generate realistic profit-loss scenarios. This provides an end to end view of the existing media ecosystem and prescribes on how to channelize efforts for new marketing strategy or new brand launch. The tool is co-created to suit exact business requirements and is customizable to meet changing business rules and market conditions.
UBIE is a universal user identification scoring mechanism. It allows you
to know more about your user’s financial position, social stand on various
issues, professional history and matching a right choice for your user.
In this era of data-driven industries, every company wants to get insights about their consumers’ preferences, to serve them well in future. It also helps in forecasting the consumer’s behaviour for the upcoming product.
User scoring can help companies to know their consumers, get to know about their shopping behavior and transaction patterns on their websites or products.
Universal Behavior Identification Exchange can transform the way we identify with our customers, especially in Ecommerce industry. A centralized database platform coupled with the scoring model delivers intelligence to B2C businesses. It helps in significantly reducing losses occurred due to false customers, location or supply.
UBIE is designed to provide an integrated approach across all the consumer based industries for scoring their consumers on a set of objects (financial position, professional qualities, e-commerce transactions, location, social media and much more). This is done using their users and social media data combined together into a customized data layer. It help industries by using their data to find behavioral user patterns. UBIE also provides an approach for uniquely scoring user as per industry attributes.
In finance industry, credit score is used to represent the creditworthiness of a person using their past transaction history and professional background. The data from NBFC is merged with professional data for checking their worthiness for future credibility. Scoring mechanism can help you to find if the user is qualified for a loan and whether they will be a bad debt. Apart from credit score, we can also score user identity about their professional past and present, their social identity using social media. Credit scores determine which customers are likely to bring in the most revenue. The use of credit or identity scoring, prior to authorizing access or granting credit, is an implementation of a trusted system.
In E-commerce, user scoring is used in context of cash transaction of buyer payments, as 80% of E-commerce transactions are through Cash on Delivery. There is a fair amount of risk in COD as buyer has no prior commitment about buying product. They can cancel order at anytime, which makes the seller pay two way courier charges without selling their product. Cancellations and returns are two big problem that all online sellers face. The risk is especially high in sales made through cash on delivery, which is one of the defining pillars of India's E-commerce boom. U.B.I.E provides location based scoring for users. It analyzes the E-commerce data which provides insight on a buyer’s feasibility to the seller, using their past transaction history and other location based factors.
In Human resource management or talent acquisition, a user’s professional background is a major criterion for their credibility. UBIE scores users by taking their educational & professional background into consideration and combining it with their presence on social media platforms like LinkedIn and Twitter. This will help to get more reliable knowledge about their skills and professional strengths, which can benefit HR consultancies to get better choices.
As matrimony industry is all about match-making, we can use social media for matching users according to their gender and their likes-dislikes. UBIE can integrate data from social media with matrimony profiles, using UBIE identity scoring match-maker to find a suitable fit for profiles with common choices, social stand and professional qualities.
Sentiment is an element of human behavior which tells us person likes, dislikes for an object, event or situation. So, Sentiment Analysis is the process of determining whether a piece of writing is positive, negative or neutral. It is also known as opinion mining, as it derives the opinion or attitude of a speaker. A common use case for this technology is to discover how people feel about a particular topic. We can use sentiment analysis for analyzing text, voice, video and image. Therefore the target of sentiment analysis is to find opinions, identify the sentiments expressed and then classify their polarity.
The applications of sentiment analysis are broad and powerful. The ability to extract insights from social data is a practice that is being widely adopted by organisations across the world. Some of the applications are listed below:
Identifying sentiments in text could be a difficult task mainly because of the ambiguity of words in text, complex meaning of words and various factors such as sarcasm, irony, politeness, writing style, as well as diverse language from person to person and culture to culture. iPredictt’s sentiment analysis process flow is defined below:
A brand manager needs to know about the response of fans with respect to the marketing campaigns being run for the brand. It becomes quite easy to gauge the influence it has on the audience. In such cases, Sentiment Analysis can be a blessing. It allows checking the sentiment behind the comments and posts of people about a brand or topic. It comes up with a segregated differentiation between positive, negative or neutral comments, thereby making the task much easier.
Sentiment analysis can also help companies develop and refine their public relations strategy. For example, companies can use sentiment analysis to identify sales leads and spot industry trends. As previously mentioned, sentiment analysis can also be used to identify influencers in your industry with positive sentiments toward your brand, which can be leveraged in a PR strategy.
Another use of sentiment analysis is the qualitative evaluation of resumes documents. Resumes can be distinguished based on different criteria like experience, education and location. However, the main section that would help distinguish each document from the rest is the skills section. Each job posting has some special requirements defined in terms of the skills that an employer is looking for. This specific requirement can be handled by extracting the skills and running sentiment analysis on it. The word cloud, thus produced, can make the scanning of resumes easier.
If Government agencies can constantly keep a tab on pulse of its citizens, it can pave the way for better governance. Social sentiment analysis can be a very useful tool to achieve the same. It can address the following questions which Government agencies would be very interested to get an answer:
In today’s world, social media platforms like facebook and twitter have gained much popularity among people to vent out their feelings. This data can be exploited to provide reliable early warning for a class of extremist-related, real-world protest events. Monitoring social media to spot emerging issues and trends and to assess public opinion concerning topics and events can be helpful.
Every B2C company spends considerable time and resources to understand their ecosystem of customers better. "Who are the customers?", "What do they really do?", "What do they like/hate?", "How to increase the share of their wallet towards my business?" are some of the few questions a company asks themselves. But all of these questions are only substitutes for the most fundamental question: Which of my customers will attrite? This field of study of customer attrition is known as Churn analytics. The primary goal of churn analysis is to identify those customers that are most likely to discontinue using your service or product.
Churn Analytics is often used as an indicator of customer satisfaction. Banks, telephone service companies, Internet service providers, pay TV companies, insurance firms, and alarm monitoring services, often use customer attrition analysis. Some of the applications of Churn Analytics include:
We use a 3 phase iterative modeling process with the following stages:
Mobile Number Portability has eased the mobile user experience of switching telecom operators or telecom circles. But, it has inadvertently aided in Customer churn for the telecom companies. In such a scenario, where the customer can switch operators at the drop of a hat, it is important that the telecom industries invest in Churn Analytics to control the Customer Attrition rates.
In the Mobile App industry, it is even more necessary due to the large number of apps that are being introduced every day. Large competition, size of the app and data usage are just the few factors that influence churn of mobile app customers. Churn analytics will aid you in retaining these otherwise difficult customers.
In the E-commerce industry, been able to make the customers to repeatedly purchase goods and services is the success of an e-commerce company. To avoid the opposite, churn analytics can be used to earmark customer micro-segments in the ‘danger zone’ which can be targeted specifically using target- oriented marketing actions like personalized offers, cash backs, etc.
A Brand takes time to be produced and introduced into a market of intense competition. Brand health, or brand equity, is a measure of your brand's performance relative to competitors. Managing brand health has become more complicated as the expansion of offline and online channels creates a vast array of consumer data that can be used for analysis in the post-digital age. This is where analytics comes into play. We help customers create accurate indexes which help monitoring relative performance of their brand across various stages of its life cycle.
In essence everyone wants to understand what the health of their brand is in relation to their competitor ecosystem. Some of the applications of Brand Health Analysis include:
iPredictt builds a 5-dimensional model to quantify brand health. By using various technologies and tools such as regression analysis, scoring, predictive modeling, Adstock and seasonality corrections etc, we create a robust model which then is used to understand related problems such as Cannibalization or Brand inception. This model consists of 5 aspects of a brand’s life cycle:
These 5 dimensions allow us to track health of a brand relative to that of its competitor brand. These latent variables comprise of a large number of variables that are refined by our data handling processes before inputting them to our 5-D model
iPredictt follows an industry standard project model called CRISP-DM for deploying brand health tracking
Invaluable insights to plan your Social Media campaigns.
The analysis of feelings behind the words using natural language processing tools, technically called opinion mining or commonly called sentiment analysis (emotions, opinions, attitudes).
This is about looking far beyond the likes you get on a product release campaign, blog post, ad campaign. As marketers to adapt our strategies and driving the actions we have planned for we need to determine and capture the complex emotional response. One of the biggest challenges that is seen in today’s online campaign is understanding the impact achieved, figuring out if the set targets were achieved and if future targets and goals could be rationed based on these results.
We will analyse the brands reputation and will uncover new insights such that identifying new targets will be easy. Our algorithm compares every possible data attribute combination of those relationships and visually ranks the strength so that you can instantly chalk out future targets to focus on. This can lead to a strong knowledge of a single data attribute like location, gender or age, or even attributes like credit score, college degree. Based on historical observations of similar person’s interests, our engine automatically predicts the interests of a person.
iPredictt Buzz would not only give you insight on campaign performance, it would also give you insight on social media sentiments. With this insight, you would be able to sharpen the edges of your existing campaigns and even chalk out alternate goals and targets. This data can be invaluable to plan further detailing on your future social media campaigns too.
We use Topological Data Analysis to find solutions to business problems
Modern science and engineering has led to the production of a large amount of data at an unprecedented rate. However, this large amount of data needs to be analysed in such a way that only that information which is relevant to us is retained.
This large amount of information is therefore analysed through the method of Topological Data Analysis. This method provides the fastest time to getting insights into large volume data, thus leading to its liberation for analytics.We employ several Geometry and Topological algorithms to analyse the data. iPredictt Core rapidly finds all the significant patterns in your data and this is owing to the immense depth and reach of its algorithms.
Insights, those which are invisible to other approaches are easily revealed by iPredictt Core by creating topological summaries using the geometry of the data. This very data breaks down several complicated scenarios in today’s businesses and can help you analyse the different predictions.
Complex data sets tend to obscure solutions to today’s profound business problems and it is these insights which provide a solution to the problems. The insights that can be traced using iPredictt Core would give you a competitive edge to understand and take critical decisions in your business. The varied spectrum of uses for iPredictt Core includes finding new customers and markets, prediction of program failures, fraud detection and cyber security, risk mitigation and several more applications which define its uses. In Short, iPredicttCore is our top of the line of products which uses Topological Data Analysis to find solutions to business problems.
Analysed Medical Data to create new pathways
Over the past decade, hospitals and medical care providers have shifted to electronic medical records. This has resulted in the accumulation of a wealth of data, which is information regarding to a patient. The medical history of a patient can be traced back to several years which in turn can help the doctor in analysing and to pin point the medical condition.
Using iPredicttHealth we take things one step further. What if disease prediction can be done well in advance? If the disease can be predicted even before its onset then not just time, but also money can be saved which otherwise would have resulted in massive hospital bills. Data mining is the fundamental principle behind disease prediction. Using data mining technique the number of test should be reduced.
When the data of a patient is recorded in EMR, the practitioner can access this data even after several years. A new type of treatment can be started by collating the information and trying to target nodes in this information which indicate the possibility of some disease like a heart attack, or a stroke. This reduced test plays an important role in time and performance.
We use the Topological Geometric Algorithms to analyse this large amount of data, form recognisable patterns which can show tell-tale signs of the onset of a disease. Medical practitioners can then use this analysis to create pathways and utilize their medical knowledge and latest of technology to enhance these pathways leading breakthrough in disease diagnostics and treatment.
Analysed data to produce a usable dataset that is easily comprehended by HR personnel
Changing employee expectations, new technologies, increasing globalisation and a need for agility in the face of a turbulent business environment mean that tomorrow’s workplace will be barely recognisable from today. HR will need to respond accordingly.
Research by Accenture has identified 10 business trends that will radically reshape HR in the next five years. Instead of managing a workforce with a one-size-fits-all approach, HR will treat each employee as a “workforce of one” with unique needs and preferences, and will customise employee incentives accordingly. Skills gaps are widening and HR will be increasingly hard pressed to ensure their organisations have the right people. HR will need to develop initiatives to be able to quickly tap skills when and where they are needed. These trends are happening now and will only get more real and impactful.
A very different set of HR and talent management practices will be required, which are better suited to a highly volatile, global and knowledge-oriented age. HR functions that recognise this and react will have an unprecedented opportunity to help organisations and people become leaders in the new world of work. For those companies that don’t heed the call, HR risks irrelevance.
Enterprise and social data is mined by iPredictt HR and analysed using our in-house algorithms and machine language interpreters to produce a usable dataset that is easily comprehended by HR personnel. iPredictt HR is the go-to platform for all employers who would like to completely understand the hire, in other words to determine if their investment in the employee would be an asset or a liability.
Data mining plays a very important role in predicting sales.
“Why bother with a sales forecast?” is what we all say when we start a new business. It is after all, at the very best, a guess. But without a sales forecast the target is zero, and how do you know what to do? These are common questions.
Using iPredictt-Ecom we collate copious amounts of data which shows trends. Sometimes called affinity analysis or Basket analysis we evaluate the patterns of credit card use, patterns in phone usage as well as evaluate fraud insurance claims. Look at your purchase data with an eye for patterns. Do you see people who buy item X also buy Y? Which item did they buy first? Why? Could you encourage people to buy X, Y and Z, thus boosting point-of-purchase sales? Depending on the purchase patterns of customers it is possible to predict as to when that customer will make another purchase.
Our machine learning algorithms working hand in hand with our pattern recognition engines using the Topological Geometric Analysis will mine this wealth of customer purchase information and provide a set of organized data that will clearly tell you the trends in which sales will progress. Sales forecasting using iPredicttEcom therefore becomes very handy to new enterprises and entrepreneurs. By examining customer purchasing patterns and looking at the demographics and psychographics of customers to build profiles, you can create products that will sell themselves. For example, as you collect this data, start to look for opportunities like best days to run a discount promotion.
For the offline, a company looking to grow by adding stores can evaluate the amount of merchandise they will need by looking at the exact layout of a current store. For an online business, merchandise planning can help you determine stocking options and inventory warehousing.