First-party benchmark

The Outlearn AI Support Benchmark

Across 29,003 customer conversations handled from Oct 2025 to Aug 2026, 45.9% were resolved automatically with no human involved (13,299 / 29,003). That means more than half of conversations ended some other way. We publish that part too.

Last updated 14 August 2026

This is a first-party benchmark across Outlearn workspaces that publishes resolutions, handoffs, and conversations that ended without either.

Outlearn is an AI support agent that reads the documentation and history your company already keeps, then keeps rereading what you connect. It answers in the website widget, over email, in Slack, Microsoft Teams, and Google Chat, and takes real actions mid-conversation. It hands hard conversations to your team with the full transcript attached and charges credits for the AI's work, never per seat.

Automatic resolution

From Jun 1 to Aug 14, 2026, 50.8% of 10,694 customer conversations were resolved automatically with no human involved (5,433 / 10,694), while 50.4% across customers to date is a per-customer average.

Both are measured averages, not a promise for any individual team.

The recent number is conversation-weighted. The standing number is a per-customer average. Neither replaces the other.

What your number looks like depends on what customers ask, what you connect, and whether the information is current enough to answer from.

Human handoffs

Across 29,003 customer conversations handled from Oct 2025 to Aug 2026, 5.7% were handed to a human (1,662 / 29,003), compared with 2.4% of 10,694 conversations in the last 90 days (259 / 10,694).

These are the two measured handoff rates for the periods shown.

Resolved automatically and handed off don't add up to the full total. Some conversations ended without either outcome.

Connected knowledge

Across 232 Outlearn workspaces with at least one connected knowledge source, the average workspace connected 2.7 knowledge sources.

It reflects workspaces connecting more than a single polished help center, pulling in the docs, internal pages, and product information they already maintain.

Production volume

In July 2026, Outlearn handled 4,571 customer conversations, or about 148 conversations per day (4,571 / 31).

July was the last full month in this snapshot. This is production usage.

Methodology

This benchmark covers conversations in Outlearn workspaces. A conversation is an interaction handled by Outlearn. “Resolved automatically” means the agent closed the conversation with no human involved. “Handed off” means the agent escalated the conversation to a human. The remaining conversations ended without an explicit resolution or handoff, such as when a visitor left or completed a low-stakes lookup.

In March 2026, 9,064 of 29,003 conversations, or 31% of the corpus, occurred in a single month.

Volume isn't uniform. We report both the all-time rate and the recent rate, and we publish both automatic resolutions and handoffs.

These figures come from Outlearn's own production aggregates, available through public Metabase cards. They cover Outlearn workspaces, not the wider industry.

No customer is named or identifiable in this benchmark. The figures are aggregates, and published per-workspace statistics use a minimum cell floor.

The cards refresh when the underlying data changes.

FAQ

What does “resolved automatically” mean?

It means the agent closed the conversation without a human joining it. The customer got an answer and the conversation reached an explicit close.

What does “handed off” mean?

It means the agent escalated the conversation to a person. Outlearn includes the full transcript, so your team can pick up without making the customer repeat themselves.

Why don't the outcomes add up to the full total?

Some conversations end without an explicit resolution or handoff. A visitor can leave, get a quick answer, or stop after finding what they needed.

Can I trust vendor-published benchmark data?

Look for a raw denominator, a time window, and clear outcome definitions. We publish each, including conversations that ended without either outcome.

Compare your own resolution and handoff mix against these definitions.

Outlearn has a free tier. Connect your docs and see your own number.

AI Support Benchmark 2026: First-Party Outlearn Data