TIMi Suite

The “TIMi suite” is mainly a suite of predictive analytic tools dedicated to the creation of high ROI marketing campaigns (typically for cross-selling, up-selling and churn prevention).

The TIMi Suite has four main unique Added Value :

1. Higher ROI than other tools using the same datasets

As illustrated by many business-cases, the ROI of a marketing campaign is directly proportional to the predictive accuracy of the predictive models that you used to design the campaign. The higher the accuracy of your predictive models, the higher the ROI of your campaigns.

As demonstrated by our outstanding performances at various world-level data mining competitions, the predictive accuracy of the predictive models built with TIMi is significantly higher than the accuracy of models built with other classical predictive tools (such as SAS, SPSS and others). Marketing campaigns designed with TIMi have an ROI that is usually from 5 to 15% more than the same campaign but designed with a competitor tool.

This small difference in accuracy (and thus in ROI) usually means for Banks, Telecom and Insurance several hundred thousand euros more at each marketing campaign.

2. More ROI again because TIMi is able to exploit larger datasets than the competitors.

One way to still increase the TIMi model accuracy is to create your predictive models using larger databases, possibly containing several thousands of columns (variables). Such databases are really easy to produce for banks and telecom because these type of company have many information about their customers.

For example, in 2009, the Orange Telecom company gave an extract of their customer database for a datamining competition (KDD2009). This extract of their database had 15.000 columns. I personally analysed a dataset for a B2B banking application that had more than 20.000 columns.

Anatella is the only ETL tool that is able to create such large datasets in a reasonable amount of time (There are some tools but they run for days instead of a few minutes for Anatella).

Outside TIMi, there exists no predictive analytic tool that is able to analyse such large datasets in a reasonable amount of time to produce the required predictive models (There are some tools but they run for weeks instead of a few minutes for TIMi and they produce low accuracy models).

The unique ability to process vast quantities of data grants to the “TIMi Suite” unmatched predictive model accuracy (and thus unmatched ROI).

3. Reduced data preparation time for predictive analytics

The TIMi suite offers another valuable unique advantage over other predictive analytic tools: Short Data preparation Time.

Usually datamining projects are composed of three steps.

When using classical datamining tools, each time that you want to produce a new predictive model, you have the following:

  • Step 1: Data Preparation: Usually 50 % of the time needed to produce one model
    • Lengthy and of moderate complexity.
    • Does not necessitate any statistical knowledge. For IT staff only.
  • Step 2: Modelization: Usually 30% of the time needed to produce one model.
    • Lengthy and difficult.
    • Specialized staffs required with deep knowledge in intricate statistical and datamining algorithms.
  • Step 3: Industrialization: (creation of reports & automated scoring): Usually 20% of the time needed to produce one model.
    • Short and very difficult.
    • The IT staff must inter-act with the statistician staff to automate the prediction process inside the production environment. Many algorithmic problems appear when new sets of data are submitted to the system. Difficult to debug due to the complexity of the algorithms.

Furthermore with classical predictive tool, each time you create a new predictive model, you usually need to create a new dataset that is tailored for the new predictive task to accomplish. For example, the time required to produce 5 models is: 5x(10+6+4)=100 mandays (5x10 days for “Data Preparation”, 5x6 days for “Modelization” and 5x4 days for “Industrialization”).

When using the “TIMi Suite”, you have the following:

  • Step 1: Data Preparation: Usually 90 % of the time needed to produce one model.
    • Lengthy and of moderate complexity.
    • Does not necessitate any statistical knowledge. For IT staff only.
  • Step 2: Modelization: Usually 5 % of the time needed to produce one model.
    • Short and easy.
    • No specialized staff is required.
  • Step 3: Industrialization: (creation of reports & automated scoring) : usually 5 % of the time needed to produce one model
    • Short and very easy.
    • The IT staff is using the fully automated procedure of TIM to automatize completely the production environment.

With TIMi, you only need to create one giant dataset that can be used to create ALL the models. For example, the time required to produce 5 models is: 10+5*(1+1)=20 mandays (10 days for “Data Preparation”, 5x1 days for “Modelization” and 5x1 days for “Industrialization”).

To summarize: Compared to a standard predictive analytic suite, the TIMi approach typically divides the whole project time by 5 and only use IT staff with a “basic” knowledge or no knowledge at all in datamining.

With TIMi, No more Lengthy Data Preparation but, to really gain time, TIMi requires a different way of working: You should federate all your data in a single very large Dataset, typically built using Anatella (because it’s the only ETL tool unlimited in the number of columns).

This is only possible because, unlike all other predictive tools, TIMi offers you the following functionnalities:

  • TIMi is unlimited in the number of columns (unique feature).
  • TIMi is robust against dirty data: (Nearly) no need to clean the data! (unique feature)
    The rule "Garbage IN = Garbage OUT" does not apply anymore!
  • The accuracy of TIMi does not degrade with the number of columns. (unique feature)
  • TIMi can model very small targets (less than 0.003% of the dataset is ok.) (unique feature)

Thanks to this unique set of features of TIMi, you can deploy with the TIMi Suite predictive analytics everywhere without any fuss or wuss!

To summarize, we claim that TIMi is the only predictive analytic tool that allows you to really industrialize predictive analytics on full-size industrial databases. No other tool offers this unique and critical functionality.

4. Comprehensible models

The Rexer 2011 independent survey reports that the biggest, most difficult challenge against the general adoption of predictive analytics is “to explain to others the results of the (predictive) datamining process”.

…and we agree: This is extremely difficult to explain your predictive models or your results when you are using a “classical predictive analytic tools”.

TIMi changes the rules of the game again!

Although TIMi is able to analyse in a few minutes datasets with tens of thousands of columns, TIMi still gives you, at the end, very simple and intuitive models. Most predictive models are composed of a very reduced set of variable (typically between 10 and 20 variables). These models still have a very high accuracy (as demonstrated by our outstanding results in various datamining competitions) and are easily comprehensible by any user, including any non-trained business user. The models are explained through a set of simple charts available inside self-contained Word and Excel reports (containing many intruitive charts).

Since everybody is able understand “how the TIMi model works”, the adoption of the predictive model by the management or by your sales team is usually easy and immediate.