Tymely CEO Ohad Rozen on the Future of Customer Service via AI-Human Hybrid Technology

Market Analysis

Source: Tymely, approved to use

The universal flaws of customer service in multiple industries are often swept under the rug. From long waiting times, complex service, to the lack of out-of-hours service, the gaps that need to be bridged are glaring yet these have been the norm that most companies and customers simply need to navigate their way through. But as the age of digitalization advances, the opportunities to revamp the inefficiencies of CS agents and chatbots are rising, and the challenges that these inadequacies are becoming more difficult to ignore.

We had the chance to interview Ohad Rozen, Co-Founder & CEO at Tymely, Ph.D. Candidate for NLP & Deep Learning. Tymely is the AI for eCommerce customer support emails that works with human-level accuracy. Their cutting-edge AI-Human hybrid technology handles support emails in minutes, provides empathetic and accurate service, and turns each customer into a promoter.

In your recent press release, you mentioned that the motivation behind the idea of “an AI for customer service” is because “customer service is broken.” What are the most glaring industry gaps in eCommerce that you’ve noticed that led you to the inception of Tymely?

Why do we claim that CS is broken? Many reasons…

  • Customer service is one of the most effective ways for brands to communicate with their customers. However, customers sometimes feel like brands make it difficult to get in touch with their service team, as they ‘hide’ their contact details (email/phone) behind a pile of Q&A pages and self-help articles. This is because good customer service is hard and expensive, and they can’t handle too many customer service inquiries, so they constantly try to let their customers help themselves. But guess what, customers don’t want to read articles or try to figure out how to help themselves. Customers want easy, fast, and empathetic service, which today is a very rare commodity.
  • In most cases, customer service today is poor:
    • Long wait times - sometimes days just to get an email response or a callback. If you want to handle a case via chat, the process can take 40 or 60 minutes, during which you are grounded to the screen, or they shut the session down
    • Complicated service - agents direct you from one to another, a lot of back-and-forth communication, and they often throw some articles at you for you to figure out by yourself.
    • Unempathetic - customers feel like the agent doesn’t care about them
    • No out-of-hours service - If a customer wants to cancel an order on Saturday, why do they have to wait until Mon?
  • Brand policy is often not defined (e.g. Should I compensate a customer that got a seriously delayed shipment for the second time during the past 3 months? How much compensation should I give?) and in practice set according to the agents rather than the management. Not enough direct and measurable metrics, no A/B testing for the policy, and no best practices by measuring it.

AI-powered customer service is a big undertaking. What were the challenges you’ve encountered along the way in cracking the code for Tymely?

Yes, it is hard. It’s hard because customers’ language is complex, and it’s difficult to understand a customer’s exact intent with all of its nuances. In fact, despite all the technological progress, there is no NLP model today that can perform this task with reasonable accuracy. This is a real challenge, and we solve it by using a hybrid human-AI system that combines human agents in the loop.

Another major challenge is that brands have a complex and often undefined policy, and to automate it we have to translate the policy from the agent’s mind into code and thus digitize it.

And finally, another major challenge we had to overcome is building this digital policy. We have developed tools that allow translating a brand’s policy into code in a fast and accurate way without any effort from the brand at all. Brands policy encapsulates hundreds of different parameters and variables (e.g. how long was the shipping delayed; is the customer address closer to the Houston warehouse or the Jersey one? Is the customer angry? etc.), and we have to understand what the relevant pieces of information are, extract them from the brand’s systems and the customer’s text, take them all into account, and then provide a hyper-personalize answer as well as to handle the ticket (e.g. provide compensation or cancel the order). This includes supporting tens of thousands of different potential answers to customers, which is very complicated to do, and this is exactly why we do it for our customers. We have built internal tools that enable us to build such complex workflows that would be virtually impossible for any brand to build for themselves without a massive undertaking.

What are the inadequacies and limitations of bot language processors in solving customer problems? Why do you think organizations with the most advanced NLP models still struggle with handling customer inquiries?

There are several reasons why organizations still struggle to transform their CS despite advanced NLP being available, mainly -

* Even state-of-the-art NLP models still don't understand complex language intricacies. In fact, 91% of all customer service inquiries are being redirected to human agents (see Gartner report, page 10). Examples of confusing inquiries that NLP struggles with:

Does the customer complain about not receiving the shipping, or the shipping information? Do they want to return the item, or are just asking about their return policy?

* Bots are geared towards making conversations - they're even called "conversational AI" - they're not optimized at taking actions to solve problems.

Walk me through your human-AI hybrid technology. How does it work?

For obvious reasons I can’t get into details, but I can say that we have our human agents that verify every AI decision on each customer inquiry, and the result is super-human-level accuracy handling of the tickets at AI scale and response times.

What is the difference between CS agents handling tickets vs. Tymely’s CS experts' verification process?

Everything is automated and digital, so you get all the benefits of a digital service, just without the annoying AI mistakes

  • The service is very fast, customers receive responses in minutes rather than hours or days.
  • Service is always accurate and according to the brand’s policy as the policy is digital
  • No ‘tired’ agent that forgot to check all the relevant information of the inquiry
  • No human mistakes with customer systems, everything is done automatically

How does Tymely optimize resolution time via a human-AI hybrid model vs. CS agents handling inquiries from end to end?

The AI does all the heavy lifting. The human agents are just verifying it. And AI is of course very fast. No agents have to think and make decisions, no agents to write a detailed response, and no agents to take complex actions in the systems. Everything is automated.

Just like every other cutting-edge technology, integration is crucial. Can you talk about the integration process of Tymely to existing workflows? What do companies need to provide?

Yes. A good integration process is key to a successful partnership. We integrate with the brand’s ticketing system (e.g. Zendesk), order management system (e.g. Shopify), and any other system the brand’s agents are working with. Some integrations can take a few days, and some a bit more, depending on the API types and how API-friendly the systems are. Customers of course need to provide us with API access to any such system.

This is original content from NewsBreak’s Creator Program. Join today to publish and share your own content.

Comments / 0

Published by

Tech Journalism, Product Reviews, Startups, Investing, Fintech

New York, NY

More from Market Analysis

Comments / 0