Subscribe to DSC Newsletter

Hello all,
I'm looking for a software tool that can deliver real time modeling of inbound calls.
The software should be able to analyze costumer’s data during their conversation with company representative (Real-Time) and their probability to buy new or another product.
Do you have any recommendations?

Views: 130

Reply to This

Replies to This Discussion

I would suggest to break your question into 2 solutions to solve the problem.

- modeling
Analyze the data you have from past transactions and build a model that captures customer preferences and predicts cross-sell opportunities.
Pretty much any of the mainstream tools should be able to do the job. Open source R is an excellent choice, as are SPSS or SAS. Once you have a model, export it using the Predictive Markup Modeling Language (PMML) standard which these statistical tools support.

PMML will enable you to then deploy it in your production system for real-time scoring as well as make the updates.

- deployment / real-time execution
This part needs a scalable scoring engine that you can integrate with your existing application, e.g., though web services, to execute your predictive model(s) in real-time and on-demand whenever you have the customer on the phone with an agent.

Please take a look at the ADAPA scoring engine we developed here at Zementis.

We focus on SOA principles for integration, open standards like PMML, and a SaaS option to deliver a cost-effective and scalable solution. ADAPA is available as a fully hosted solution on Amazon's Cloud Computing infrastructure or as an onsite license.
Thanks Michael,
It's seems very promising solution but i need a solution that do real-time modeling in one stream and not two ( which can not be a real-time in any way). Can your product be installed as stand alone or only as Saas?
Hi Ron,

Yes, ADAPA is also available standalone as a regular software license.

I would like to learn more about your requirement for real-time modeling. What features would need such a rapid model adjustment?

Not sure, if it is feasible to build a model in real-time and guarantee that it is stable and that the results are acceptable. How about refreshing the model on a nightly/weekly/monthly in a batch process and then deploy the new model for scoring whenever needed?
Hello Michael,
I need the software for online interaction. The idea is that the software is supposed to hold a pre model that can be change during the interaction, For example: a customer has a risk of 60% to churn. During the conversation the risk get higher because he has mentioned a competitor so now the representative should have new offer for him based on real time calc.
Softwares like : ThinkAnalytics,Oracle,Unica Interaction have this kind of software but I would like to receive an opinion from someone who actually deployed such a solution.
Hi Ron,

Got it, thanks for the clarification!

We have experience with similar requirements and what you describe does not require you to build a new model on they fly or update the model itself. You only need to (re-)score the active customer with the updated customer data as you collect new bits of information.

The scoring may be done with the same model, e.g., by feeding new input to the model as you collect new information. As such, in your example the model would expect "mentions competitor" as input, just that it was "0" initially, and then it is "1" as the agent talk to the customer. As soon as new relevant information is collected, a new request is sent to the scoring engine which returns and updated score ("risk to churn").

Alternatively, your scoring solution may invoke different models that are geared towards certain data input that is only available at certain stages of you business process (i.e., customer interaction). You could, for example, segment a collection of predictive models with business rules, which is a feature that our ADAPA Enterprise Edition (mixing Rules + Predictive Analytics) supports. Please see for more details.

Key issue is, any model that you are planning to use in a fully automated, mission-critical, operational environment must be tested and absolutely reliable. Otherwise, the model's decisions could adversely impact the business bottom line ($$$) and you would not even notice it until it may be too late.

I'd assume that the solutions you mention work in a similar fashion. If not, I'd certainly like to learn more about it.

Hope this helps!

Kind Regards,
Hello Mike,
Thanks for your replay, I'll enter your site for more details.
Wondered if you had a chance to visit our site.


On Data Science Central

© 2021   TechTarget, Inc.   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service