OTHER
Return photo-vs-claim mismatch detector
Cross-checks a customer's stated return reason against what the uploaded photos actually show using a vision model.
How it runs
The automated pipeline, trigger to output.
- TriggerNew return conversation in FrontFront
- ActionExtract photos and stated return reasonFront
- ActionDescribe actual photo condition with Hugging FaceHugging Face
- LogicCompare visible condition against claimed reason
- ActionLog cleared or flagged result to AirtableAirtable
- OutputAlert fraud-review channel on mismatch in SlackSlack
What it does
This workflow compares what a customer says is wrong with a returned product against what their photos actually depict. When a return arrives in Front with a stated reason ("arrived broken", "never opened", "wrong item"), it sends the photos to a Hugging Face vision model and checks whether the visible condition is consistent with the claim. If the photos contradict the reason — an item described as unopened but clearly used, or claimed broken with no visible damage — it flags the RMA for fraud review instead of letting it auto-approve.
When to use it
Use it when return fraud or claim-padding is a real cost and you want a second check on the stated reason before a refund clears.
How it works
- 1A new return conversation in Front triggers the flow.
- 2Photos and the stated return reason are extracted.
- 3A Hugging Face model describes the actual visible condition of each photo.
- 4A logic branch compares the described condition against the claimed reason.
- 5Consistent claims are logged to Airtable as cleared; mismatches are marked for review.
- 6Flagged mismatches post an alert with photos and reasoning to the fraud-review Slack channel.
Set it up
What you configure once, before turning it on.
- 1Connect FrontShared inbox, conversations.
- 2Connect Hugging FaceModels, datasets, spaces — the open-source hub.
- 3Connect AirtableBases, tables, views, automations.
- 4Connect SlackChannels, DMs, threads, mentions.
- 5Set each agent's modelWe leave models unset so you pick the tier — fast + cheap, or top-quality.
- 6Tune it to your dataEdit the prompts, filters, and field mappings so it matches how your team works.
- 7Test, then turn it onRun once against a sample, confirm the output, then enable the trigger.
More Other workflows
Sync IVR Prompt Registry in Airtable to Fresh ElevenLabs Audio
Runs nightly against an Airtable IVR prompt registry, finds rows whose script text changed since last synthesis, regenerates only those ElevenLabs clips.
Agent-Driven Full IVR Re-Voicing for a Rebrand
An agent takes a rebrand brief from Notion, audits every IVR prompt for old naming, rewrites and re-synthesizes the affected ones with ElevenLabs, archives them to Dropbox.
Regenerate IVR Voice Prompts When Notion Naming Doc Changes
Watches a Notion product-naming page and, whenever a product or feature name changes, regenerates the affected ElevenLabs IVR audio prompts and saves the new MP3s to Dropbox.
Pre-register a visitor, email a QR badge, and alert the host on arrival
When a host submits a visitor pre-registration form, this creates a visitor record, emails the guest a scannable QR badge with arrival instructions.
Sweep stale visitor check-ins, auto-checkout, and flag overdue guests
On a recurring schedule this finds visitors still marked on-site past their expected departure, auto-checks-out anyone past end-of-day.
Slack-Approved IVR Re-Voicing After a Product Rename
On demand from Slack, drafts updated IVR prompt scripts for a renamed product, posts them for human approval.
Run it inside a business
This workflow drops into a full company template. Import the org, and this is one of the playbooks its agents run.
Run this workflow in your colony.
14-day trial. No DevOps. No Sales call. Provisioned in under a minute.