AI & RAG
Flag help articles whose RAG keeps missing real user phrasings
Pulls recent Intercom conversations where the AI answer bot scored low or got a thumbs-down, clusters the unmatched user phrasings against your help center.
How it runs
The automated pipeline, trigger to output.
- TriggerWeekly audit schedule
- ActionFetch recent answer-bot conversations + CSATIntercom
- LogicKeep low-confidence or thumbs-down replies
- ActionCluster missed phrasings, map to articlesOpenAI
- ActionStore article-to-phrasing mapPostgres
- OutputOpen a Linear ticket per weak articleLinear
What it does
It audits which help-center articles your retrieval layer should have surfaced but didn't. It reads recent answer-bot conversations, isolates the questions where retrieval missed or the customer rated the reply unhelpful, then maps each miss to the article that *should* have answered it and the exact phrasing the user actually typed. Each weak article becomes a Linear ticket with the missed phrasings attached.
When to use it
Run it weekly when your Intercom answer bot deflection rate is plateauing and you suspect the content is right but the wording doesn't match how customers ask. Ideal for support-content and knowledge-ops teams who own article quality.
How it works
- 1A weekly schedule fires the audit.
- 2Fetch the last 7 days of Intercom answer-bot conversations plus their resolution and CSAT signals.
- 3Filter to conversations with a low retrieval-confidence score or a thumbs-down rating.
- 4An OpenAI step clusters the unmatched user phrasings and matches each cluster to the closest existing article.
- 5Persist the article-to-missed-phrasing map in Postgres for trend tracking.
- 6Open one Linear ticket per under-performing article listing the phrasings to add as synonyms or headings.
Set it up
What you configure once, before turning it on.
- 1Connect IntercomConversations, contacts, articles.
- 2Connect OpenAIModels, embeddings, files.
- 3Connect PostgresAny Postgres URL — query, write, migrate.
- 4Connect LinearIssues, projects, cycles, triage.
- 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 AI & RAG workflows
Publish a Grounded API FAQ Page to Confluence Weekly
Each week, clusters the top unanswered or repeated API questions, generates spec-grounded answers with citations.
Detect Breaking API Changes from Spec Diffs and Alert Owners
Compares the new OpenAPI spec against the previous version on each GitLab merge, uses retrieval over the changelog to classify whether changes are breaking.
Pre-meeting prep brief grounded in Coda and CRM
Before each booked sales meeting, builds a one-page prep brief by combining the account's HubSpot context with grounded talking points and objection responses pulled from your…
Coda-grounded sales answer bot with citations in Slack
Reps ask product, pricing, or competitive questions in Slack and get an answer drawn only from your Coda knowledge hub, with links to the exact docs and rows it pulled from.
Weekly knowledge-gap digest from unanswered rep questions
Each week, scans rep questions the answer bot couldn't ground in Coda, clusters the recurring gaps.
RFP and security questionnaire drafter grounded in Coda
Drafts answers to inbound RFP and security questionnaire questions by retrieving approved language from your Coda hub, then files the cited draft for review before a rep sends it.
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.