Better Customer Experience: Trendyol Assistant Journey
In the rapidly changing e-commerce industry, all stakeholders in this ecosystem require support for various processes to keep pace with the changes.
In response to the ever-evolving needs of our stakeholders, we are committed to providing tailored solutions through our chatbot product, also known as Trendyol Assistant.
Since 2020, with the daily increase in the number of orders, we’ve been working hard to answer customer questions about delivery processes and resolve any issues with their orders.
In this article, I will outline the key aspects of the journey and technologies employed during the period in which we elevated our Trendyol Assistant success ratio from approximately 55% in 2020 to 77% by October 2023.
The Objectives of Trendyol Assistant
Trendyol Assistant is a user experience product designed to enable users to resolve their issues within the chatbot through a self-service method, without the need to connect with live support agents.
In other words, our goal is to reduce reliance on agents through Trendyol Assistant, thereby decreasing agent costs and simultaneously providing a better post-order user experience.
Let me delve into the feature sets we employ in the Trendyol Assistant to provide a better post-order experience to users.
Local Language Support
Currently, within Trendyol’s ever-growing ecosystem, we have 17 different chatbots that we continuously develop to address the evolving needs of various stakeholders, including customers (core, go, dolap, international, MENA, Azerbaijan etc), sellers (international, domestic, go), influencers, and more.
We have also added local language support (Arabic, Azerbaijani, German, English) to these bots in line with the evolving customer needs. In our adventure of adapting to local languages, in the upcoming days, we will launch 7 new local language-supported bots for 7 different EU countries.
Announcement Feature
Upon entering Trendyol Assistant, users are greeted with announcements, which can be displayed as pop-ups if desired. Business teams frequently use this announcement feature, particularly when introducing new features on Trendyol or during incidents to inform our users. This proactive announcement approach informs the end users beforehand and prevents future contacts.
Service Integrations
Through service integrations with other teams, users can accomplish various needs within Trendyol Assistant.
For instance, they can track the status of their orders or return requests and check the progress of the support tickets they’ve submitted.
Additionally, users can directly communicate their order-related complaints to the seller or, in specific scenarios, submit requests to relevant courier companies through the ticket workflows we provide them.
In addition to these, users can also fulfill their needs such as order cancellation, order address change, generating return codes, and accessing their invoices through Trendyol Assistant.
Various Message Types
Trendyol Assistant supports various message types, including emojis, bold text, italic text, colored text, bullet points, and videos. With these diverse features, we provide users with a more readable experience.
Conversational Answers
With our conversational model project conducted in collaboration with the Data Science Team, German and Turkish customers can receive responses from Trendyol Assistant not only by selecting from predefined menus but also by writing their issues themselves.
First, the data science team categorizes and understands the questions users write, and then we as a Trendyol Assistant team prepare specific answers to users for their questions. This allows us to respond to customers’ inquiries on specific topics beyond the predefined menus, covering over 300 different subjects. Through our ongoing analyses, we continuously enhance the model.
Designing Customer-Centered Solutions with Assistant: Behind the Scenes
To develop a journey for Trendyol Assistant, we regularly meet with the relevant customer/seller experience and business teams. In these meetings, we thoroughly examine the detailed analysis of end users’ contact data and NPS reports. In addition to these, we also use Trendyol Assistant survey comments and the records of livechat data to understand customer needs. Alongside our UX team, we conduct Trendyol Assistant end-user interviews to obtain feedback and regularly make benchmarking with other global chatbots.
After identifying the problems in the end users’ journeys, we make kickoff meetings with the relevant experience teams. During these meetings, we strive to understand the root causes of the end user problems and discuss possible solutions for them. While discussing the journeys, we occasionally address solutions not only to Trendyol Assistant screens but also to different touchpoints of the customers, such as the order detail page or post-order transactional communications. So, another primary goal of these meetings is to minimize the need for users to reach the chatbot or contact center in the first place with the right project ideas through improvements made by other product teams.
Considering the impact of the solution on overall contact and the end user’s experience, we track these project ideas, along with their priorities, in our annual backlog document. In alignment with these yearly goals, we conduct monthly backlog planning sessions with related business teams. Sometimes we make integrations to design the Trendyol Assistant journey, and we reach out to the related Product Managers of those teams, mostly to; Order, Delivery, Storefront, and ticketing tool domains. Every week, we develop the planned tasks with our Trendyol Assistant development team during our weekly sprints. Also, we gather with the team at our weekly retro meeting to analyze our sprint velocity, discuss what went well, identify areas for improvement, and plan actionable items for the next sprints.
Trendyol Assistant Success Meetings
Every month, we implement numerous improvements in Trendyol Assistants. To ensure a user-centric roadmap, beyond just making these improvements, we closely monitor their impact on the user journey and operational outcomes once they go live. Therefore, each month, we collaborate with the relevant business teams to review the metrics of the respective assistant at our Trendyol Assistant Success meetings.
During these meetings, we generate action plans by brainstorming ideas derived from the click reports prepared specifically for each assistant.
The two key metrics we primarily focus on in the click reports are:
- Adjusted Success Ratio: If a user does not connect to the voice or live chat contact center within 15 minutes after creating a session first in Trendyol Assistant, we consider this a success. We closely track the success rate of each assistant on a daily, weekly, and monthly basis.
- Trendyol Assistant Survey Score: At the end of each Trendyol Assistant session, we ask the user to rate our bot service on a 5-point scale. It allows us to measure the user satisfaction score of each Trendyol Assistant.
I’ll share the direct positive effects of a few sample improvements we’ve made in the past few weeks on the mentioned metrics.
- Meal Customer Assistant: We adopted a proactive screen approach for meal customers. Customers previously felt the need to inquire about the status of their active orders or active return requests for meal orders. With a proactive approach, as soon as customers enter the assistant, we now detect and provide upfront delivery/refund information if there’s an active order or return request. We show estimated delivery times and return request status. With this enhancement, we achieved a 10% increase in the success ratio.
- International Customer Assistant: For defective products customer complaints, we introduced the feature for users to create self-service tickets through the assistant. This change resulted in a 4% increase in success ratio.
- Domestic Customer Assistant: In cases of orders being in transit for a long time, we enabled customers to create self-service tickets through the assistant. With this feature, we saved more than 64 thousand sessions in the Assistant and prevented agent necessity for these chats.
Which technologies do we use to conduct multi-stakeholder management?
Trendyol Assistant is fundamentally built upon the State Machine architecture. To implement the state machine concept, we utilized the Spring State Machine framework alongside SpringBoot in our core projects on the backend. Additionally, we leverage a mix of languages including Kotlin, Java, Golang, and React, and utilize databases such as Couchbase and PostgreSQL. For cache management, Redis is employed, while monitoring is facilitated through Newrelic, Kibana, and Grafana. In terms of deployment and management, we rely on Docker, Kubernetes, Istio, Argocd, and Consul technologies.
Additionally, we’ve developed a dynamic Auto State Management structure in Trendyol Assistant. It allows us to add and update menus and contents dynamically. With this framework, we can swiftly add/remove menus and update content without the need for extensive development efforts.
Conclusion:
So far, by embracing technology with Trendyol Assistant, we have developed user-centric solutions and contributed to the experiences of end-users. Moving forward, we will continue making impactful improvements that touch the lives of users while concurrently reducing our operational costs behind the scenes :)
