video1.0<iframe src="https://www.loom.com/embed/9b674f3cae38462597a53769d4dea911" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/os/loom-media-content/loom/c9d31a95-3433-4197-bce1-71e74cff6782/939c408a-ad61-4b6f-b7ab-fc13245343dc/sessions-thumbnails-9b674f3cae38462597a53769d4dea911-18a1dc9b810c3d7d.gif/binary?authToken=eyJraWQiOiJtaWNyb3MtZ3JvdXAvdGRwLW9zLWFyY2hldHlwZS9iOW5jNGxsMTMwYXUzYjduIiwiYWxnIjoiUlMyNTYifQ.eyJicyI6Imxvb20tbWVkaWEtY29udGVudCIsImF1ZCI6InRkcC1vcy1hcmNoZXR5cGUiLCJuYmYiOjE3OTIzMjA3MjcsInRpIjoiODU3ODEwYTctNjA5ZS00YmU2LWFlMzctZDkyM2FlNDU0MjlkIiwiYWRpIjoibG9vbS9sb29tIiwiaXNzIjoibWljcm9zLWdyb3VwL3RkcC1vcy1hcmNoZXR5cGUiLCJwaWQiOiJhZjg0OWY2NS0wZDc1LTRlOTgtYmE2Ny05YTI2M2M4MTZkYjMiLCJ3YSI6ImZhbHNlIiwib2lkIjoic2Vzc2lvbnMtdGh1bWJuYWlscy05YjY3NGYzY2FlMzg0NjI1OTdhNTM3NjlkNGRlYTkxMS0xOGExZGM5YjgxMGMzZDdkLmdpZiIsImV4cCI6MTc5MjMyMTMyNywiaWF0IjoxNzkyMzIwNzI3LCJqdGkiOiI1NmVhNmE4OC1iNjg4LTQzNjQtYWM3OC0xNjRmMDdlMmIwMjUifQ.okMzwdf7qE0WAIn1F2t6yz5AE1S4MM6VeSOy3Kslka5PTWnt2mszJtPvPGYmW_Vf0W4fTN-5pMh5bypcve4FmEEesd32X5JlToQiu3CEboMdTe7pRizr0yH4d3iAuDss8IIWRBBWZzXcdcE_4h4U7C3LTQmfQ3NStmKTfxfCIcnAq1AC4kjOaDbSDCijO0Mkd_09yNHbyXGaSgenBKvXLJrUdr40Db92eYL6OOun72sXySbKFjrLgiCn9BguQCfKTIScs7ozlqkn2UT_-9FEQxqae-1PwU-kt5zevhfnU9mDvOrwVUF8DF0sk97io3easXg3CvwVo7tp19c5UZw6vQ&Expires=1792321327&Signature=kfr7QlDenKo5gW963lpuSsh0m4-d1bw7ElhMaSnbOV9VXywjeutbjatQPT4hDgmEHtMn%7E%7E1VnntDZPx7e8vVokR8rHitMrsTuEywGt290TlfwFXAI8CP-MFWsfzXpNLSTBsCfVVeSZeVm2y9kLRuJfCkrUdi7a1UYZ91WZGf%7ELmpXXwE2onHoLbRoVX3JaxlASO7JvX79zbFl45S5FZzMWQUnuI%7EIP%7EOn%7EoCqNOl4mgnJQajRW8NW4rnc6TCnwAlVB2XC8SlJgiyGUAoWXH6VxukJ2LLqMytU3goTSr-T1YPWo1ZPiLaGqUfAy3nA5Rkt3kvKmOJYfyxaCyeYAggvA__&Key-Pair-Id=KQOSYIR44AIC0259.833333This Loom demonstrates an AI customer support automation workflow built with Anything Google Gemini, Airtable, GNL, and webhooks. It starts with a webhook that receives messages, checks Airtable for duplicate message IDs, and if none exist, has Gemini classify the request into one of six categories while returning a confidence score, refund indicator, and a human review flag. The workflow parses Gemini JSON, creates a support ticket in Airtable, drafts a response with Gemini, and then either closes the ticket or escalates to humans when a refund is requested, confidence is below 80%, or the AI indicates a review. It also includes an error-handling path that logs failures to an error table and notifies an internal Gmail address.