Customer support made easier and more efficient while respecting established workflows and systems.

An additional software layer on top of any ticketing or communication software for customer support.

It finds the best replies to a customer query, increasing the support staff efficiency or even automating part of the responses.

The system consists of few interconnected parts:

• Knowledge base that contains all approved / appropriate answers to customer queries
• Machine learning based matching engine that finds the best response from knowledge base
• The historical data training process that teaches the matching engine on what the right answers are for the specific business
• Integration layer with helpdesk / customer support platform used by your business

At the basic level, the system is simple to understand. You provide us with potential answers to customer queries that we put into knowledge base, we transform these vector representations of these answers, plug it into a Savvier matching engine, and whenever you have a query, we return back a sorted list of best answers to it - somewhat similar to a web search engine.

However, to fully utilize the system, using a generic matching engine is still missing something- it doesn’t know the context of your business specifics.

If, for example, you have a product called “Socrates” (that is used to generate reports on your platform), the generic matching engine will think customers are referring to the philosopher and return slightly confused responses.

A much better approach is to train the engine according to your data - all the products, lingo, and subtle context can be transferred to Savvier by fine-tuning the matching engine to your business domain.

By using historical data - that is, previous customer and support staff interactions - we can teach a matching engine to understand your business and provide more accurate answers. This not only includes responses that existed in historical data - if a new response or piece of knowledge is added to knowledge base, the matching engine already has the context to understand how it relates to other queries, and can use this response right away if it is a match.

To take a step further, you can then close the loop - all new responses through integrated interface automatically become part of the historical data.

That's surprisingly common! Building a knowledge base is often a daunting task - someone has to sit down and try and come up with usual questions and answers you receive, often slogging away reading through past emails and notes to come up with something that "feels" right.

Savvier can make it easy and turn qualitative answers into quantitative data. We can parse any semi-structured historical communication (for example, email archives) into actual representative customer queries and responses for you to approve.

Since the system is custom tailored to match your business needs, we can tailor and expand the system to support the processes you have. It’s hard to list the applications without understanding the specifics, but some past examples to get the idea are as follows:

• Routing messages to specific departments - as an implementation for higher education institution, Savvier tags the messages to a “general”, “technical support”, “undergraduate” or “postgraduate” categories and routes to specific departments in the institution.
• Generating dynamic responses from API calls - given a customer provided a tracking number and verified their identity, system queries and generates a message about a status of a parcel delivery - the support staff don’t need to switch between systems to check it.
• Extract and build response from documentation - for a client that provides a technology solution, part of the knowledge base is the documentation. For questions in technical nature, Savvier parses the documentation and provides a snippet with a link to the documented example of the technical implementation.
• Fully automated response system - although we are very hesitant to let Savvier do customer support without supervision, we can enable it to automatically respond to well documented questions that match historical records with an answer that have been approved by business staff before - used sparingly, it can solve the most annoying and most repetitive questions without a human intervention.
• Self-serve knowledge base search - instead of building a chatbot, why not offer a smart search system for a public knowledge base. Savvier understands the meaning of query, so even if the customer doesn't know the exact words, it can intuitively provide the answer they are looking for.