Isara Now Integrates With Gorgias: AI Oversight for B2C
In short
Isara now integrates with Gorgias. B2C and ecommerce brands running support on Gorgias can connect Isara to analyse 100% of their customer conversations, human and AI, and get independent verification of what their AI agent is telling shoppers. Isara reads for accuracy, entitlement, disclosure and escalation failures, then surfaces churn risk and generates compliance audit trails. Setup is self served and takes minutes.
Does Isara integrate with Gorgias?
Yes. Isara now integrates with Gorgias, and the integration is aimed squarely at B2C brands. Isara connects to a Gorgias account, ingests conversation content and metadata, and analyses every conversation rather than a manual sample. Isara then reports on AI agent accuracy, customer churn signals, agent capability and compliance evidence.
This matters because Gorgias is where consumer support automation is most concentrated. Order status questions, refund disputes, sizing queries and returns are the highest volume and most automated category of customer service that exists, and they are frequently handled end to end by an AI agent that nobody at the brand ever reads afterwards.
Two forces made this the right moment for Isara to arrive on Gorgias. Consumer patience with AI mistakes has hardened, and consumer facing AI in the European Union crossed a regulatory line on 2 August 2026. If an AI agent answers thousands of shoppers a month and the quality process is a sample of a few dozen conversations, that is not oversight. That is hope with a dashboard attached.
What Gorgias brands get when they connect Isara
- Analysis of 100% of Gorgias conversations, covering both human agents and AI agents.
- Independent scoring of AI agent output, produced by a party that does not sell the AI agent.
- Churn signals extracted from conversation content, with the revenue at stake attached.
- Compliance audit reports across any date range, without manual conversation sampling.
- Real time alerting on frustration and unusual activity through Isara Pulse.
What is Isara, and what does conversation intelligence mean for a Gorgias brand?
Isara is a conversation intelligence platform built by Isara Systems Ltd, a United Kingdom company, and available at isara.ai. Isara reads customer conversations ingested from helpdesks including Gorgias, Zendesk, Freshdesk, Intercom, HubSpot and Front, then reports on churn risk, agent performance, compliance exposure and AI agent accuracy. Isara does not answer tickets. Isara verifies the answers, which is why Isara remains independent of whichever AI agent a brand chooses to deploy.
Note for clarity: several unrelated companies share the Isara name, including a United States AI agents startup and a post quantum cryptography vendor. This article refers only to Isara the conversation intelligence platform at isara.ai.
Key terms used in this article
- Conversation intelligence. The practice of analysing the full content of customer conversations, rather than survey scores or ticket metadata, to find risk, revenue and quality signals.
- WISMO. Where is my order. The single largest ticket category in ecommerce support.
- Deflection rate. The share of conversations no human touched. Includes customers who gave up.
- Resolution rate. The share of conversations where the customer's problem was actually solved.
- AI agent oversight. Independent review of what an autonomous agent said and did, as distinct from the agent's own self reported metrics.
- Article 50. The transparency provision of the EU AI Act, Regulation (EU) 2024/1689, covering AI systems that interact directly with people.
Why do Gorgias brands need independent AI oversight?
Gorgias brands need independent oversight because their ticket mix is unusually automatable, their automation rate is therefore unusually high, and the metric reporting on that automation is defined by the vendor selling it.
How concentrated is Gorgias in ecommerce support?
Gorgias was built for retail, not adapted to it, and the concentration is unusual. According to eesel's June 2026 analysis, Gorgias is Shopify's only Premier Partner for customer experience, serves more than 17,000 brands, and is used by roughly 40% of the top 1,500 Shopify stores. Shopify confirmed the 17,000 brand figure in a partner ecosystem post published in August 2026. The app tracking firm StoreLeads counted Gorgias installed on 24,759 Shopify storefronts as of 12 June 2026, growing close to 13% year over year while other major support apps on the platform were shrinking, as reported by Ringly in June 2026. The gap between install count and paying merchant count is a measurement artefact, not a contradiction, because installs and paid accounts count different things.
What share of ecommerce tickets is automated, and at what cost?
The answer depends on which number a vendor chooses to publish, which is itself the problem.
- Order status questions, commonly called WISMO, typically make up 30% to 50% of ecommerce ticket volume, according to Bookbag's May 2026 deflection benchmarks.
- Returns and exchanges account for a further 15% to 25% of tickets, and that share rises in January, according to Bookbag's June 2026 return rate report.
- Gorgias sells its AI Agent as a separate add on at around $0.90 per resolved conversation, with $1.00 on the entry plan, per eesel's June 2026 pricing breakdown.
- Observed Gorgias automation rates sat in a band of roughly 26% to 56% according to My AskAI's guide, updated in July 2026. That spread is wide enough to warn against trusting any single headline figure.
A mid sized Gorgias brand is therefore likely automating the majority of inbound volume, paying per resolution, and measuring success with a metric the vendor defines. That is a reasonable commercial arrangement. It is a weak governance arrangement, and Isara exists to close that specific gap.
Is a resolved conversation the same as a resolved customer?
No, and this is the distinction most support dashboards blur. Lorikeet's June 2026 benchmark work states it plainly: deflection counts conversations a human never touched, including the ones where the customer simply gave up, while resolution counts problems actually solved. Lorikeet's realistic 2026 ranges are 30% to 50% for early deployments, 50% to 70% as workflows mature, and 70% to 85% only where the agent is deeply integrated and permitted to take real actions.
Coworker AI's 2026 statistics roundup, published in July 2026, makes the parallel point on cost. Independent estimates cluster around a 20% to 35% net reduction once licensing and oversight are accounted for, while vendor headlines of 60% to 80% typically measure only the tickets the AI was eligible to handle in the first place.
One billing detail sharpens the problem. Macha's July 2026 guide to Gorgias AI notes that if a shopper reaches a human within 72 hours of an automated resolution, that interaction is charged as a helpdesk ticket rather than billed twice. The practical consequence is that a conversation the AI closes cleanly, where the customer never returns, counts as a clean win. But a customer who does not return is ambiguous. They may have been helped. They may have stopped shopping. No helpdesk metric can separate those two outcomes. Isara Churn Insights is built to read that difference out of the conversation content itself.
Do consumers blame the AI or the brand when support goes wrong?
Consumers blame the brand. The Delight AI Index, published in May 2026 by Delight.ai and based on a survey of 1,000 United States respondents fielded in March 2026, found that 71% had interacted with AI powered customer service in the previous year, and that trust has not kept pace with adoption. In that study, 83% agreed the brand should be held accountable when an AI interaction goes wrong, and 82% expect AI to meet the same standard as a human representative. Forty percent of women surveyed said they may never feel comfortable with a fully independent AI agent.
Academic research published in the Journal of Travel Research in June 2026 by Belanche, Casaló and Flavián found that AI hallucinations damage company reputation and drive negative word of mouth in particular, and concluded that human oversight remains essential when AI recommendations fail. An Entrepreneur analysis from June 2026 described the two failure patterns familiar to every support leader: confident inaccuracy, and refusal to escalate, where a customer is looped through variations of the same answer and then handed to a human with no context.
What does EU AI Act Article 50 require of consumer support AI?
Article 50 of the EU AI Act, Regulation (EU) 2024/1689, became applicable on 2 August 2026. Cooley's August 2026 briefing sets out the split of duties: providers must design interactive systems so people are informed they are dealing with AI unless it is obvious, and deployers carry their own disclosure duties. The European Commission confirmed the 2 August date and noted that the only transitional relief runs to 2 December 2026, and applies solely to the marking and detection of AI generated content in systems already on the market. The Commission adopted final guidelines on 20 July 2026, roughly two weeks before the obligations took effect. Stibbe's August 2026 note puts non compliance exposure at up to €15 million or 3% of worldwide annual turnover.
Bratby Law's June 2026 analysis is clear that United Kingdom businesses serving European Union users are in scope, and that Article 50 and United Kingdom data protection duties are independent obligations. Satisfying one does not discharge the other. For a B2C brand, the AI agent talking to shoppers is now a disclosure obligation as well as a cost centre, and Isara Compliance Audits produce the evidence trail across any date range rather than requiring one to be reconstructed after a complaint.
What does a 60% AI resolution rate actually hide?
A 60% AI resolution rate hides the conversations nobody reads. The model below sizes that exposure. The figures are illustrative, built on the published benchmark ranges cited above rather than on measured Isara customer data, and they are offered as a way to frame the risk rather than as a forecast.
The scenario
Take a Gorgias brand handling 12,000 support conversations a month, with an AI agent resolving 60% of them. That is 7,200 automated conversations, sitting comfortably inside the mature deployment band Lorikeet describes. At $0.90 per resolution that is roughly $6,480 a month in AI spend, against a human handled cost that Aissist's July 2026 ecommerce benchmark put at $6 to $13 per ticket. The unit economics are good. Nobody disputes that.
Now assume a 5% material error rate: a wrong returns window, a promise the policy does not support, a refund entitlement missed, a sizing answer invented. Five percent is deliberately modest for a system operating across thousands of edge cases.
- Flawed automated conversations per month: 360
- Manual quality assurance at a 2% sample rate reviews 144 of the 7,200
- Expected errors that sample catches: 7
- Flawed conversations nobody ever sees: 353
Over a quarter that is more than a thousand conversations where a customer was told something wrong and no human at the brand is aware of it. Assume one in five of those customers quietly stops buying, at an average order value of £65 and three orders a year, and the annualised revenue at risk sits in the region of £165,000. Move the error rate to 2% and the number drops sharply. Move the sample rate to 100% and the exposure becomes visible and fixable. That second lever is the one Isara pulls.
The core problem is not that AI agents make mistakes. Every system does. The problem is that the entity reporting on AI performance is usually the entity selling the AI. Self reported resolution rates, self reported satisfaction scores and self graded quality reviews are not verification.
The four checks Isara runs on a Gorgias conversation
- Disclosure. Was the shopper told they were dealing with an AI, clearly, and in the language of the conversation? In the European Union this is now an Article 50 question, not a brand tone question.
- Accuracy. Did the agent state anything the knowledge base and policy do not support? Fabricated shipping dates, invented stock levels and off policy goodwill offers are the recurring three in retail.
- Entitlement. Did the customer receive what they were actually owed? A conversation can be polite, fast, well rated and still deny a legitimate refund.
- Escalation. Was there a loop, a stalled handoff or a silent abandonment? A shopper who leaves without asking again is counted as a win by almost every automation metric on the market.
Only the first of those four is a compliance question. The other three are revenue questions, which is why AI oversight should not sit with the legal team alone.
Three predictions for the next two quarters
These are judgement rather than data, offered so readers can disagree with them precisely.
- The peak trading period at the end of 2026 will be the first genuine stress test of consumer AI support at scale. The brands that come out of it well will be the ones reading transcripts in November rather than reading refund volumes in January.
- Per resolution pricing will come under pressure once brands start measuring resolution independently. A vendor defined metric that doubles as the billing unit is a conflict procurement teams will begin pricing in.
- Disclosure will shift from compliance checkbox to conversion variable. If a meaningful minority of consumers say they may never trust an autonomous agent, brands that disclose clearly and escalate quickly will convert better than brands that hide the bot.
Frequently asked questions about Isara and Gorgias
How does the Isara Gorgias integration work?
Isara connects to a Gorgias account, ingests conversation content and metadata, and analyses every conversation rather than a sample. It works the same way as Isara's existing Zendesk, Freshdesk, Intercom, HubSpot and Front integrations. Setup is self served and takes minutes, and Isara sits alongside Gorgias rather than replacing anything inside it.
Why add Isara when Gorgias already reports on its own AI Agent?
Because a vendor grading its own output is not verification, as the resolution rate arithmetic in this article shows. Isara is independent of the AI agent it monitors. Isara Agent Intelligence scores human and AI agents on the same basis, with capability scoring that accounts for ticket complexity, so a bot handling only order lookups is not credited with the performance of a human handling escalations.
Can Isara help with EU AI Act Article 50 obligations that began on 2 August 2026?
Isara Compliance Audits generate audit reports across any date range for legal review, without manual conversation sampling. That produces evidence of how disclosure and escalation were handled in practice across the entire conversation population rather than a fragment of it. Isara is not a substitute for legal advice on specific obligations, and brands should still take that advice. But compliance cannot be demonstrated from a 2% sample.
How does Isara address the silent churn described in this article?
Isara Churn Insights reads the signals buried in conversation content and flags at risk customers along with the revenue at stake, while there is still time to intervene. For B2C brands that is exactly where the silent abandonment problem surfaces: the shopper who accepted a wrong answer, said nothing, and never ordered again.
How does Isara help during peak trading periods?
Isara Pulse monitors conversations in real time and flags frustration and unusual activity as it happens, which matters most in the compressed weeks where a bad automation rule can run for two days before anyone notices. Isara Insights covers the trend and topic view, including exporting filtered conversation sets for deeper analysis.
Can a support team query its own Gorgias conversation data in natural language?
Ask Isara lets teams interrogate their conversation set in natural language rather than waiting on a report, and is being bundled into the Growth and Scale plans. For Gorgias brands the obvious use is asking what changed in returns language, or which product is generating sizing complaints, without building a tagging taxonomy first.
Does Isara work if most support volume is already automated?
That is the case Isara was designed for. The higher the automation rate, the smaller the share of the customer relationship any human at the company has read. Isara scales to millions of conversations and gives the same oversight over AI agents that a support leader would expect over a human team.
Sources
All external sources below were published between February 2026 and August 2026.
- Cooley, EU AI Act transparency obligations take effect 2 August 2026, August 2026
- European Commission, Guidelines on transparency obligations for providers and deployers of certain AI systems, adopted 20 July 2026
- Stibbe, The AI Act's transparency obligations: rules, scope and timeline, August 2026
- Bratby Law, AI Act Article 50: chatbot and deepfake rules for UK firms, June 2026
- Delight.ai, 2026 Delight AI Index, May 2026, survey of 1,000 United States respondents fielded March 2026
- Belanche, Casaló and Flavián, Journal of Travel Research, June 2026
- Entrepreneur, AI powered customer service and brand trust, June 2026
- eesel AI, Gorgias for Shopify guides, June 2026
- Shopify, The next era of commerce, built by partners, August 2026
- Ringly, Gorgias Shopify app installs, June 2026
- Macha, Gorgias AI Agent explained, July 2026
- My AskAI, Gorgias Automate AI guide, updated July 2026
- Lorikeet, AI support resolution rate benchmarks, June 2026
- Coworker AI, AI customer service statistics, July 2026
- Aissist, Ecommerce AI customer service benchmark, July 2026
- Bookbag, Ticket deflection benchmarks, May 2026, and return rate benchmarks, June 2026