Customer Engagement Using AI

It could be said that in the near future, customer engagement which encompasses product marketing, query handling, customer support can be handled by machines without much human intervention.

Customer Marketing, Sales, and Support, which is mostly built on communication, is always a dialogue between a particular customer and the brand. This human dialogue (interactions) is mostly about seeking information about your brand, or negotiations before purchase or after sales queries or issues, managed by support teams.

These ongoing dialogues which exhibit a sense of context and understanding of sentiment are quickly managed by humans demonstrating intelligence(complex understanding). Machines have not really scaled up to manage ongoing dialogues, like the way humans manage it, simply because machine data is in silos and not integrated like the human brain to find context quickly and react to the dialogue.

A machine that can constantly learn from customer behavior based on supervision rules can be in a position to understand customer types (whether he is a new visitor or a prospect or a loyal customer); understand customer’s intent to bring relevance to the engagement, and understand the intensity of their behavior to fine-tune its engagements, mostly mimicking human behavior.

Recurrent Neural Networks(RNN), which is a class of artificial neural network, can be used to mimic human behavior. The connections between units form a directed cycle creating an internal state of the network which allows it to exhibit dynamic temporal behavior. RNNs can use their internal memory to process arbitrary sequences of inputs.

Plumb5 Neural Network can be compared to RNN but is more dynamic where states are changing continuously until they reach an equilibrium point and remain until the input changes. The activation of states is derived based on the underlying state changes within its hierarchy. The final state is achieved through a staged manner based on a feedback received from previous iteration’s output.

This allows Plumb5 to exhibit AI in real-time and mimic human behavior by incorporating context and depth-of-sentiment into ongoing dialogues with customers.

How does it work?

Before we implement the network, we need to make sure that data needs to be unified(in one place) so that we don’t miss out data from certain channels, which could cause errors leading to wrong contexts. It is essential that all the customer communication channel data is integrated and this unified set is used for learning.

A semantic data model needs to be implemented to build associations between data coming from your omnichannel solution. The hierarchical model of Plumb5 can be used to quickly deploy the data model for integrating and organizing data, which will serve as the classification rules for the learning network. With classification in place, the network has to just allocate weights based on occurrences to manage states, allowing the network to learn in real-time.

These states are used to deliver engagements with custom content, incorporated with relevant messages, which can be supervised by humans.

States: In order to automate response selection and engage in conversations, it is important to integrate responses back into the network and arrive at new states, to be able to intelligently converse with customers.

Plumb5 has default states, custom states and derived states in order to manage customer conversations with minimal human intervention.

Default States These states act as markers to understand user types like anonymous, prospects, customers etc. Marketers can add more states to identify a particular audience profile. For example, the default state “Prospect” can quickly help in identifying that the user has not made a purchase.
Derived States


These are sub-states within each default states that allow you to identify how far the user is from the next milestone. This is deduced from weights assigned to generic behavioral attributes. For example, Sales funnel which is classified as Cold/Warm/Hot can be considered as sub-states. Another set of states (super sub-state) within these sub-states to understand the dynamic intent is enabled to gauge the probability of success of achieving the next milestone. These weights are primarily deduced from engagement attributes.
Custom States If the marketer anticipates certain specific states which are specific to their domain, they can configure new sub-states. For example, if you were to identify a specific segment and converse with them with a new strategy, then they can create a new sub-state. If users have not bought a specific product during a timeline and the marketer wishes that the machine engages with these kinds of customers, with a new engagement, then they can set custom states and automate the conversion campaign.

How does Plumb5 handle States?

A Unique Tag is assigned to understand the state of the customer

As Plumb5 maintains a unified stack for each individual customer, it is easier to identify a user. Based on associated parameters, they get tagged as

Anonymous If the user is identifiable by only IP
Prospect If the user is identifiable by email/mobile phone
Customer If the user transaction field is activated
Repeat Customer If the user transaction has more than one transactional record
Profitable Customer If the NPV(Net Profit Value) of the user is positive

How does Plumb5 handle Sub-States?

From the weights assigned to input variables, the marketer can deduce sub-states either by supervising weights to identify sub-states or resort to unsupervised learning.

The Unsupervised algorithm deduces the sub-states over three layers. The first layer extracts the depth of engagement, which is reduced to nearest in the second layer using relevance, which is further trimmed down by recency.

The marketer can supervise this flow by setting up weights based on interactions, where the engagement parameters can be prioritized, enhancing extraction outputs in the first layer.

The marketer can also hard wire these states and create a completely supervised environment.

How does Plumb5 handle Super Sub-States?

Using the same input weights, the layer prioritizes scores based on engagement(responses), page events, recency of the event and sentiment, to understand the finer intent of the customer. These states will allow the machine to understand whether the customer’s intent has increased or reduced, in order to fine-tune the communication. For example, a negative weight for low recency might trigger follow-up communication whereas a negative sentiment weight might trigger an advocacy response, with an objective to win back customer confidence (positive sentiment weight)

How does it manage relevant content?

The interrelationship between states and encoded weights constitutes a pattern. Each pattern can be compared to a memory. These patterns are represented by alphanumerical strings that contain an encoded weight for each unit level input tags and derived tags (hidden). These tags are used to select relevant content to populate into the dynamic templates.

Content is managed using dynamic templates, which is set to each state. Whenever the customer is true for a state, the dynamic template picks up the relevant headline, associated product titles, images, descriptions, price and other relevant information.

Using tags, it can be easily connected to an external content management system to populate content into the specified template.

How does it manage dialogue?

The tags and state information play an important part in real-time customer conversations. The NLP Layer can be integrated to utilize context and depth during the conversation. The Reassembly algorithm identifies the state and key tags of the user in conversation, to incorporate into the response assembly, mimicking human-like behavior.

For instance, If there is a particular customer on an apparel store, who has browsed through 50 products in the past, is back on the website. It is important for the machine to get quick access to the relevant attributes like product type, color, patterns and more. These tags are incorporated into the response assembly, which will allow the machine to keep conversations meaningful. You can click here to view a post on how it’s done.

With this implementation, marketers can just lean back and monitor how each of their customers is continuously engaged and step in with strategies that can speed up customer satisfaction and revenue conversions.

Plumb5 readymade framework can quickly allow businesses to plug their data into the unified data model, from which the inbuilt network can process decisions and provide a real-time AI environment for businesses to supervise the machine.

Prism of Possibilities

Plumb5 integrated learning engine can provide businesses with a range of growth possibilities influencing top-line and bottom-line revenue growth. The AI engine which learns from the unified customer stack using segment-of-one customer data is able to address individual patterns and deliver personalization based on individual patterns. For a more global view, these similar individual patterns are grouped for users to view at a macro-segment level.

Listed are the aspects that are solved or optimized in order to ensure business profitability through AI and Automation. Continue reading →

Plumb5 v6

A look into the V6 version of Plumb5 with built-in real-time decision models for state automation overlayed with NLP algorithms. It is capable of intelligent machine-driven automation by learning and firing next actions.

The highlight of this version is the NLP-driven data search, which enables users to get real-time metrics across any data combinations taking away the cost of time to create custom reports and to manage this pipeline.

Contact us to try a proof of concept over your existing data.

Overlay NLP Based Search over any data infrastructure



The Plumb5 Data Search allows users to simply search for answers and reports by simply integrating it into the existing data environment.

The integration can be done in 3 easy steps

  1. Data Tagging: If the data is lying in silos and not unified, then we plug all data sources to tag relevant data entities using the tag manager. If data is unified, we can jump to the next step.
  2. Based on the relationship, we generate a structure and map it with the NLP keyword list, created by the data prep team.
  3. Set permissions for users allowing them to get answers within the permissible limit

The business can now allow any user to ask and find answers or reports over the connected data. This saves the humungous amount of time required for techies and analysts to create custom reports over multiple silos.

As data grows, all we need to do is to add the new data entities to the relationship structure and update the keyword list.

The Plumb5 Data relationship model is scalable. Click here to read about the data model and how it helps in real-time data synthesis.


Data Security and Governance

As the volume and proliferation of data—from cloud to mobile to social to machine—continues to grow, serious measures need to be taken to deliver safe harbor for sensitive data. Data security governs protection across data states throughout its lifecycle, whereas data privacy defines that access based on privacy policies and laws.

The focus of this article is to explain how Plumb5 users can easily govern their data by controlling data access to its unit parameter and be assured of protection across data states (in motion, at rest, creation and disposal)

Data Security: Businesses can address each of their security needs by data masking and data encryption techniques across the 4 data states. Be it data for customer service, analysis, learning or reporting or even to share it with outsourced vendors, sensitive data can be protected from any kind of attacks.

dataprotRobust network security will help protect data in transit. Firewalls and network access control will help secure the networks from attacks and intrusion.Dynamic data masking limits sensitive data exposure by masking it to non-privileged users.

Policies for systematically categorizing and classifying all company data, no matter where it resides, in order to ensure that the appropriate data protection measures are applied while data remains at rest and triggered when data classified as at-risk is accessed, used, or transferred.

By defining, analyzing, and protecting sensitive data at its source, Plumb5 helps in keeping your data secure before it proliferates throughout your organization.

Data Access and Permissions

The platform provides a flexible data sharing model that makes it easy to assign various data sets to different sets of users. This ensures you can balance security and convenience, minimizing the risk of stolen or misused data while making sure all users can easily access the data they need.

The platform includes simple-to-configure security controls that make it easy to specify which users can view, create, edit, or delete any record or field in the app. You can configure access at the level of the organization, objects, fields, or individual records.


By combining security controls at different levels, you can provide just the right level of data access to thousands of users without having to specify permissions for each user individually.

Users can create a user group and map data columns individually with further permissions to read/write data.

With the unified data architecture, the business can improve privacy through individualization where privacy rules can be designed based on traits and preferences, which can help in individual compliance with data privacy.


Smarter and Effective Data Infrastructure for Businesses

A smart data infrastructure which enables self-service model, not only solves cost overheads but also ensures effective and timely data dissemination paving way for quicker business growth.

IT HyperConvergence just around the corner – Gartner

With Hyperconvergence happening in the Systems area, Silos will be completely eliminated.Secondly, “One throat to Choke” will be approach taken by all IT infra teams as it is easy to manage security with an converged IT,….in the near future,we are going to see 100s of end points, that does not mean 100s of systems & operations. This is leading to a growing demand in Integrated Systems. Hyperconverged operational & BI systems is the way forward

Continue reading →