Covidence helps researchers around the world run complex systematic reviews. As the platform grew, monthly active users climbed and ticket volume followed. A small six-person team was responsible for support, training, webinars, and consulting, with support accounting for only a fraction of their workload. They wanted to keep their median response time under two hours without sacrificing the research-quality answers their users relied on.
Helply became the way to scale support without compromising accuracy. By using Helply’s Gap Finder feature, Covidence built an AI agent that could solve customer problems reliably while researchers focused on high-stakes work.
Key Performance Highlights
30% reduction in tickets
Helply now resolves roughly a third of inquiries that would otherwise go to human agents, even though it only runs on the knowledge base today.
70% resolution rate
Internal review shows that Helply answered correctly and completely the vast majority of the time.
Reduced “Thumbs down” rate to under 3% with Helply answers
Researchers and users consistently rate the vast majority of responses as highly satisfactory.
When growth turned support into a constraint
As Covidence became more widely adopted, monthly active users climbed and ticket volume rose to roughly one thousand to fifteen hundred relevant tickets per month. Each ticket often included multiple back-and-forth messages, turning those one thousand tickets into three thousand or more individual interactions.
The support team was small and highly skilled. Every support agent was a researcher first and a support professional second. Support made up only about twenty percent of their workload. They also delivered trainings, webinars, and high-touch consulting.
The team had a clear target:
- Keep median response time under two hours.
- Maintain quality that matched the expectations of evidence-based research.
- Do all of this without adding headcount.
It worked, but it was fragile. Any surge in demand risked pushing response times up or pulling researchers away from higher-value work.
Why Covidence started looking at AI support
The goal was simple:
- keep up with growing demand
- protect the reputation of Covidence support
- let researchers focus on critical, complex questions
They did not want a generic chatbot. They needed a system that could safely answer customer questions without hallucinations or fabricated answers. In a research environment, a wrong answer is worse than no answer.
Covidence needed an AI agent that understood this.
Why Helply stood out
Covidence evaluated multiple tools. Most failed on integration and safety. Helply passed with flying colors.
Helply is SOC 2 and GDPR compliant, and integrates directly with help desks. That integration allowed Helply to:
- pull historic tickets
- ingest the existing knowledge base
- map real questions to real answers from the support team
- identify “gaps” in knowledge (and automatically answer and fill those gaps)
"Helply is integrated with Groove, and that is extremely useful. It has been trained on real conversations, and the way it connects tickets, knowledge base articles, and training data gave Helply a clear advantage over other tools."
Onboarding that required no engineering
Onboarding was intentionally light. Because Helply handled the heavy lifting, Covidence did not need engineering resources to go live.
Here’s how it worked:
- the engagement started with an onboarding call
- Helply pulled in knowledge base content and analyzed historical Groove tickets
- Covidence received a recorded walkthrough of the dashboard and training workflow
- Covidence gained access to a dedicated Slack channel for ongoing support
Helply prepared the initial training data, wired up the AI agent, and delivered a ready-to-use system. Whenever questions came up, Helply recorded screen shares, answered in Slack, and iterated based on feedback.
For a research team with limited time, this level of guidance mattered.
"The Helply team was genuinely invested in making the AI agent work for our users. Feedback was not ignored. Most of it turned into real improvements."
How Covidence uses Helply today
Covidence currently runs Helply on their knowledge base. Most users work inside the primary application and click through to the knowledge base when they have questions. There, the Helply agent sits ready to help.
Even with that limited surface area, Helply now handles about 30% of the ticket volume when controlling for constant monthly active users. That is a significant load off a small, specialized team.
The typical user workflow:
- user encounters an issue in the app
- user visits the knowledge base
- Helply answers the question directly where possible
- Helply hands over unclear cases to human support
The team reviews AI agent conversations to optimize Helply. They answer as a community manager would, save that answer, and then retest. Over time, Helply becomes better at the questions that matter most.
What the numbers show
From Covidence’s internal review:
- 30% reduction in tickets at constant MAU
- 70% resolution rate
- Reduced “Thumbs down” rate to under 3%
- 80%+ of Helply responses rated satisfactory by community managers (for the remaining 20%, when Helply is unsure, it hands off instead of guessing)
"What we like is that it has been programmed to hand over when it is not sure. That is exactly what we want. Handovers are better than hallucinations."
Protecting research quality while scaling support
Covidence does not judge Helply against generic customer support. They judge it against experienced researchers who have deep domain knowledge. Even under that standard, Helply clears the bar.
It also changes how the team spends time:
- Helply handles repetitive customer questions
- community managers focus on complex research issues
- researchers spend more time on webinars, trainings, and one-to-one help
Helply has not eliminated the need for human support. It has made that support focus on more meaningful, valuable, and higher-leverage activities.
A clear plan for the future
The next step for Covidence is to bring Helply into the main app. This is a priority for the upcoming quarter. Once the AI agent sits inside the core workflow, Covidence expects a much larger share of questions to be deflected before they ever become tickets.
They see this as the biggest opportunity for impact:
- more in-app deflection
- fewer tickets for simple issues
- more time for high-impact education and outreach
Final word
"I would say definitely give Helply a go. The team is amazing, the support is amazing, and the AI agent is amazing. It will hand over instead of hallucinating, and if your goal is to give users the right answer, Helply is a very strong solution."
