The World’s Work Marketplace
Upwork is the world’s work marketplace. It serves everyone from one-person startups to over 30% of the Fortune 100 with a powerful, trust-driven platform that enables companies and talent to work together in new ways. The Upwork talent community earned over $3.3 billion on Upwork in 2021 across more than 10,000 skills in categories including website & app development, creative & design, customer support, finance & accounting, consulting, and operations.
Brandon Savage, VP2 of Customer Experience & Trust at Upwork, prioritizes predictably great outcomes at every step of the support process. He strives to create a support environment where every time a customer has a problem that needs to be solved, agents respond quickly and effectively.
Part of delivering great support experiences is providing self-serve options so users can get answers to their questions quickly. Alix Perez, a Project Manager for Chatbot & Help Center Products on Brandon’s team, handles much of the self-serve customer support tools. Her role includes building workflows for chatbots, handling support data, and managing various projects for the customer support team.
Challenges– Alix Perez, Project Manager for Chatbot & Help Center Products
I like working with Forethought because they never say no to my requests. The product is straightforward and easy to use, and when I do have suggestions or need additional features, the team is always there to help.
Consistency Across a Distributed Support Organization
As Upwork scales to onboard more freelancers to their talent community, providing accurate responses without an agent touchpoint became increasingly important.
“We came to Forethought to solve a problem of consistency across a broad number of support agents. Our teams work remotely all over the world and are loosely connected. Ensuring they have the same access to the same answers, are trained consistently, and deliver accurate responses is so important,” says Brandon Savage.
Prior to using Forethought, Upwork used a competitive product for their chat widget on their website. On the frontend, users were displeased with the chatbot because they were given inaccurate responses. Oftentimes they just wanted to connect with an agent instead of getting wrong answers from the chatbot.
On the backend, the chatbot had to be trained manually. Perez went through every user interaction to find key words and phrases that were frequently used to manually enter them into a query. This took at least 1 hour per workflow, not counting the workflows she had to re-train because they were not trained correctly initially. With thousands of workflows, training models was a full time job.
– Alix Perez, Project Manager for Chatbot & Help Center Products
We were working with a different company before. They don’t brand themselves as an AI product because they work differently–I had to train every workflow myself. We had thousands of workflows built that were often duplicated that answered questions incorrectly. It became a big monster that was too complicated to manage.
From an agent perspective, it was difficult to search the vast array of support resources quickly enough to provide a timely response to customers’ inquiries. Before Forethought, it was common that an agent would have a customer chat open in one tab, with 20 additional tabs with different help center articles. While the user would finish their question, agents were rapidly searching through their tabs to extract the perfect answer.
With a remote, distributed workforce of independent professionals, connectivity issues occur often. With 20+ tabs open at once, agents’ computers were often slow to load information–leaving customers’ inquiries unanswered while the various tabs loaded.Solution
Creating Predictable Outcomes
Upwork uses Forethought’s Solve product for both email and their chat widget, as well as Forethought’s Assist and Triage products.
Accurately Solving with Real AI
Upwork uses Solve’s chat widget within their support community. Rather than relying on keywords and decision-trees to get to the root of the support inquiry, Solve uses real AI to understand the meaning of the inquiry. The widget is able to quickly decipher the meaning of complex chat inquiries and surface up the perfect answer from support documents.
With Workflow Builder, Upwork can build their own guided experiences within the chat widget. Workflow Builder allows the support team to build short, simple workflows that replace the need for users to read lengthy support articles.
“Now that I have Workflow Builder, I’m able to build from scratch and add the functionality of true AI. What I like the most is that you can’t create a workflow without first having an intent trained.”, says Alix Perez.
With Solve Email, freelancers can receive accurate answers to their lengthy support inquiries via email. By comprehending sentence structure, meaning, synonyms, tone, and nuance, Solve Email deciphers the intent of long-form text within emails. Solve responds with quick answers and helpful articles.
Surfacing Relevant Information to Agents
Upwork support agents use Assist to surface relevant knowledge articles, past cases, macros, and notes within their help desk. Agents particularly love that they can use the Notes feature to create their own templates, craft how-to guides, and keep personal notes that will be automatically suggested for future related tickets – all in one location.
Instead of relying on multiple open tabs with different support articles, agents open up Assist to find answers. Once they open Assist, which has already read and understood the contents of the support ticket, relevant knowledge articles and information is readily available to them. This alleviates manual effort for agents by reducing the need to toggle between many different tabs or screens to find relevant information.
– Alix Perez, Project Manager for Chatbot & Help Center Products
Forethought Assist is one of the best and most robust products out there. It is always there and always working for us. The resources are optimized so that all knowledge bases are at our agents’ fingertips.
Understanding Sentiment to Triage Quickly
With Triage, Upwork uses historical data to proactively predict and classify new support tickets. Using existing routing capabilities, agents prioritize and handle inquiries without manually triaging.
Triage also helps the team instantly understand the sentiment of each ticket. Triage detects patterns in text to automatically classify the overall emotion of the inquiry as positive, negative, or neutral. By understanding which types of tickets have a negative sentiment most often, the team redirects resources to create better content and workflows to address those topics.Results
Self-Serve Options Provide Better Experiences
Upwork provides seamless experiences for both users and agents with Forethought.
With Solve Email, users who send in emails to customer support get responses with 99% accuracy. With Upwork’s previous chatbot provider, they achieved a 25-30% containment rate, which the team has converted into a 45% self-serve rate. Now with Solve, Upwork has achieved a 52-65% average self-serve rate.
- 52-65% average self-serve rate achieved via chat widget, compared to 45% with previous chatbot provider
- 575K inquiries resolved within chat widget
- 50% self-serve rate for emails
- 800 emails deflected per week
With Assist, agents who are onboarded with the product are 100% more efficient than agents who have not yet adopted the tool. For agents who are not yet using Assist, cases are resolved in 8 minutes on average, compared to 4 minutes on average for agents who are using Assist. Similarly, agents who use Assist complete 100% of tickets in their queue, compared to 90-92% completion rates for agents who are not yet using Assist.
- 50% reduction in time to close tickets
- 100% completion of tickets in queue
The main benefit of Triage has been reducing the number of clicks needed for agents to resolve cases. The need for manual processes, such as tagging and triaging cases, is now automated.
- 500K auto-predictions on incoming tickets applied
- 90% accuracy and coverage in classifying new tickets
Driving Product Decisions with Support Data
As a community-first company, Upwork aims to reduce friction in the support process for both external customers (freelancers) and internal users (their agents). The support organization plans to prioritize the client experience by offering self-serve options to solve all of their issues.
In the future, Upwork hopes to make the chat widget their only channel to access customer support. They plan to install Solve in more places on the Upwork website, including the resource center and Upwork’s new Academy product. As a data-driven company, the support organization plans to leverage Forethought’s data analysis capabilities to understand how they can improve upon products that receive the most customer inquiries with negative sentiment.
– Brandon Savage, VP2 of Customer Experience & Trust
AI is a key value we’ve looked for in our relationship. As Forethought’s analytics capabilities keep becoming more robust, the products are helping us see outcomes of our AI investments in very tangible ways.