Groove is now Helply.

Customers/Research / Healthtech

How Covidence automated researcher support without sacrificing accuracy with Helply

Automated routine researcher support with accuracy-first, human-in-the-loop escalation.

Zerodrop in accuracy while resolving 64% of support end to end (72% at peak)
Share article
Covidence team

Challenge

Covidence’s most valuable support came from experts, and those experts were drowning in routine questions.

The support team is small and deeply technical, staffed by people who understand systematic review methodology. But a growing share of their day went to the same repeatable questions about configuring a review, plan limits, and importing references, routine work that squeezed out the complex methodological problems only they could solve. Their hard-won expertise lived in individual inboxes, with no system turning it into leverage.

The obvious fix, a generic chatbot, was the one they most distrusted. Their users are researchers on strict deadlines and evidentiary standards, and a confidently wrong answer could corrupt real research. Automation that guessed wasn’t a convenience, it was a liability.

“Our whole value is that we get the hard questions right. A support bot that hallucinates a workflow step isn’t helpful, it’s a threat to a researcher’s review. And everything we’d learned was trapped in people’s heads. That bar is why we’d held off for years.”
Razia Aliani, Senior Systematic Reviewer

Covidence needed a complete platform that took the routine load off the team, captured their expertise, and never guessed on the things that mattered.

Solution

Covidence deployed the full Helply platform, with the agents tuned for accuracy above all.

The Support Agent answered strictly from Covidence’s verified support history and documentation, resolving routine workflow, plan, and import questions end to end, and configured to escalate rather than improvise whenever confidence dropped, handing off to a human with the whole investigation attached. Alongside it, the Assist Agent worked inside the experts’ own conversations, investigating across connected systems in the background and surfacing evidence-backed answers so the team moved faster on the genuinely complex cases.

Every resolution fed the platform’s institutional learning: an expert’s approach to a tricky methodology question became company memory and, where it recurred, a reusable Skill the agents applied consistently. Living documentation turned repeated questions into drafted articles for the team to approve, and native connections to Confluence, Notion, and Jira let the agents pull from living docs and file engineering issues without a human relaying them.

“What convinced us was watching it decline to answer, it hit a genuinely ambiguous question and escalated instead of bluffing. And now when one of us solves something hard, the platform remembers it for everyone. Our expertise finally compounds instead of evaporating.”
Razia Aliani, Senior Systematic Reviewer

Onboarding centered on tuning escalation thresholds until the team trusted the agents to know the limits of what they knew.

Results

Covidence offloaded the repetitive majority of its queue while holding its accuracy bar exactly where it was, and turned expert knowledge into a shared asset.

The Support Agent resolved 64% of inbound conversations end to end at steady state, climbing to 72% at peak review seasons, with no measurable rise in inaccurate answers over the human-only baseline because it escalated rather than guessed. The Assist Agent accelerated the human team on everything else.

Platform highlights

  • Full platform adopted, help desk plus agents on one learning layer
  • 64% autonomous resolution at steady-state, 72% at peak, with no drop in accuracy
  • Assist Agent investigates complex cases in the background for the expert team
  • 22 review-workflow Skills built from verified expert resolutions
  • 20+ new knowledge base articles auto-drafted by living documentation

With routine volume handled and their expertise captured as reusable Skills, the team redirected recovered hours to complex review configurations, difficult imports, and onboarding new institutions. Researchers got instant answers on the basics and a human specialist, now faster, thanks to Assist, on anything that genuinely required one.

“We stopped choosing between fast and accurate. The platform takes the routine volume, knows exactly when to hand off, and remembers how we solve the hard problems. Our experts finally spend their time on the work that actually needs experts.”
Razia Aliani, Senior Systematic Reviewer
Share article

More customer stories

Get the latest from Helply.

Ideas on AI agents, customer support, and building AI-native customer experiences.