We read public reviews of each AI product, quote them word for word, and code each one on five points: does it do the job, is it accurate, does it hand off well, is it worth the money, is it easy to run. Every share comes with its 95% interval and the reviews behind it.
AI agents: reviewers of Gorgias AI Agent most often say it resolves customer questions on its own (100% of 21 reviews that mention it), but its interval overlaps 4 other products, so first place is not settled. 5 of 10 AI agents have too few public reviews to rate.
Ranked by the share of reviews that say the agent resolves customer questions on its own, among reviews that mention it either way. Rank ranges show where intervals overlap.
| Rankrange | AI agent | Does the jobpositive share, 95% interval | Accuracypositive share, 95% interval | Handoff to a personpositive share, 95% interval | Costpositive share, 95% interval | Setup and upkeeppositive share, 95% interval | Reviews |
|---|---|---|---|---|---|---|---|
| 11–5 | Gorgias AI AgentGorgias | 100%85%–100% 21 of 21 | 2 of 5too few to rate | –too few to rate | 1 of 3too few to rate | 4 of 6too few to rate | 25 |
| 21–5 | DecagonDecagon | 100%72%–100% 10 of 10 | 0 of 5too few to rate | 1 of 3too few to rate | –too few to rate | 1 of 5too few to rate | 10 |
| 31–5 | Lyro AI AgentTidio | 95%77%–99% 20 of 21 | 3 of 4too few to rate | 2 of 2too few to rate | 1 of 2too few to rate | 8 of 9too few to rate | 25 |
| 41–5 | Freddy AI AgentFreshworks | 94%73%–99% 16 of 17 | 2 of 3too few to rate | 1 of 1too few to rate | 3 of 4too few to rate | 2 of 4too few to rate | 20 |
| 51–5 | Fin AI AgentFin (formerly Intercom), part of Salesforce | 71%45%–88% 10 of 14 | 0 of 9too few to rate | 2 of 5too few to rate | 0 of 1too few to rate | 0 of 9too few to rate | 25 |
| – | AdaAda | Not rated: 9 reviews found, 8 say whether it does the job (a rank needs 10). | 9 | ||||
| – | Agentforce ServiceSalesforce | Not rated: 5 reviews found, 4 say whether it does the job (a rank needs 10). | 5 | ||||
| – | Forethought AI AgentsZendesk (Forethought) | Not rated: 6 reviews found, 6 say whether it does the job (a rank needs 10). | 6 | ||||
| – | SierraSierra | Not rated: 0 reviews found, 0 say whether it does the job (a rank needs 10). | 0 | ||||
| – | Zendesk AI agentsZendesk | Not rated: 13 reviews found, 4 say whether it does the job (a rank needs 10). | 13 | ||||
Each category has its own definition of doing the job. Products are only compared inside a category.
Draft replies, summarize tickets and suggest articles for human agents.
Drafts and suggestions are usable
Score every conversation against your QA scorecard.
Scores match human judgment and save review time
Answer phone calls end to end.
Handles calls on its own
Word for word, linked to the original.
“Pricing could be more competitive for the services provided.”
“Our copilot constantly misses on internal docs, and simply says it could not find anything or provides irrelevant recommendations, how can this be improved?”
“The AI assistance is particularly helpful for resolving issues more efficiently and maintaining a consistent service experience.”
“The AI-powered conversation analysis reduces the need to manually review calls, making it much easier to identify trends, coaching opportunities and areas for improvement.”
“The gorgias AI agent is amazing and has really relieved pressure from our customer services department.”
“Our AI resolution rate is around 90%, which has helped us significantly reduce the number of customer service inquiries our team needs to handle manually.”
Twenty reviews can swing a share by twenty points. Try it.
Two made-up products. Change the number of reviews and the shares to see when one product is really ahead.
A: 70.0% (48.1% to 85.5%). B: 65.0% (43.3% to 81.9%). The intervals overlap, so A and B would share a rank range.
| Reviews | Interval | Width |
|---|---|---|
| 10 | 39.7% to 89.2% | ±24.8 pts |
| 20 | 48.1% to 85.5% | ±18.7 pts |
| 50 | 56.2% to 80.9% | ±12.3 pts |
| 100 | 60.4% to 78.1% | ±8.8 pts |
| 300 | 64.6% to 74.9% | ±5.2 pts |
A public review by someone who used the AI product, dated Oct 1, 2024 or later, that talks about the AI itself. Vendor case studies and staff posts are left out.
Each of the 22 products also has a sourced profile: how it hands off, what controls it has and how it is billed.