83 automations. 48 running right now.

Every new lead lands in the CRM already rated cold, warm or hot. If the model goes down, a backup takes over; if the JSON comes back broken, a fixer repairs it.

Before: research every company before deciding who to call. After: the card already says where to start.

1. CRM webhook
2. Input guard(failure handling or safety)
3. Public registry + bureau
4. Response cache(failure handling or safety)
5. LLM with structured output(AI step)
6. Backup model + JSON fixer(failure handling or safety)
7. Cold · warm · hot
8. Update the card

Leads are split by score and workload before the day starts, on business days only, in batches of 10. Stalled cards move on their own, and an alert fires when the lead stock runs out.

Before: split leads by hand and chase stalled cards. After: the queue arrives ready and the funnel moves itself.

1. Scheduled at 7:30
2. Business days only(failure handling or safety)
3. Fetch reps and leads
4. Score filter
5. Set aside risk-blocked leads(failure handling or safety)
6. Distribute in batches of 10
7. Log + stock alert

Every opportunity is checked against a master table, with the company ID validated and branches detected. Worse data never overwrites better data.

Before: everyone filled in the origin their own way. After: one table decides and the CRM follows.

1. CRM webhook
2. Validate company ID(failure handling or safety)
3. Look up the master table
4. Detect a branch of another deal
5. Classify the origin(AI step)
6. Idempotency + downgrade block(failure handling or safety)
7. Update card and master table

Create or edit the card and the credit-platform registration comes out ready and enriched. Sends retry on failure and, if that is not enough, the error is logged instead of vanishing.

Before: type the same registration into two systems. After: the card becomes a registration on its own.

1. Create/update webhook
2. Fetch deal, fields and users
3. Query the public registry
4. Bounded polling(failure handling or safety)
5. Build the payload
6. Send with retry(failure handling or safety)
7. Respond or log the error(failure handling or safety)

News from the last 48 h, classified by AI, ranked by relevance, cross-checked with the CRM and never repeated.

Before: dig through news on several sites. After: the digest arrives ready on the phone.

1. Scheduled at 8:00
2. Fetch last 48 h of news
3. Normalize and dedupe(failure handling or safety)
4. Classify the opportunity(AI step)
5. Validate JSON and score(failure handling or safety)
6. Cross-check with the CRM
7. Send without repeats(failure handling or safety)

Change the stage in the CRM and the row shows up in the dashboard spreadsheet, with sector and revenue already normalized by AI and no duplicates.

Before: copy from the CRM into the spreadsheet. After: the row shows up on its own.

1. CRM webhook
2. Normalize fields and dates
3. Query the public registry
4. Classify sector and term(AI step)
5. Normalize revenue and referral(AI step)
6. Route by funnel stage
7. No duplicate rows(failure handling or safety)
8. Write to the spreadsheets

A form or channel post becomes an enriched company and card, and the answer comes back mentioning whoever asked. If it is a duplicate, the flow warns and stops.

Before: duplicate cards. After: the flow warns and stops.

1. Form response
2. Query the public registry
3. Normalize address and contacts
4. Check for an existing company(failure handling or safety)
5. Warn about duplicate and stop(failure handling or safety)
6. Create or update company and card
7. Post to the channel with @mention
8. Return the verdict in the thread

Contracts, statements and notices come out of a single response: the flow checks the content and usually hands it to a desktop bot that generates the document.

Before: build each document from the form by hand. After: the document comes out of the response itself.

1. Form response
2. Parse the fields
3. Identify the sender
4. Read each attachment
5. Check the content(failure handling or safety)
6. Trigger the desktop bot

Both owners get the invite, and a meeting already booked is never created twice. A scheduled routine also cross-checks the base with the public registry to update leads.

Before: book it in the CRM, then again in the calendar. After: the invite comes from the card itself.

1. CRM webhook
2. Convert dates to UTC
3. Fetch the attendees
4. Resolve the emails(AI step)
5. Check for an existing meeting(failure handling or safety)
6. Stop if an attendee is missing(failure handling or safety)
7. Create the Teams meeting
8. Log to the spreadsheet

The lead moves forward, the company folder appears with the reports and the link goes back to the card.

Before: gather reports and send the link by hand. After: the folder is already on the card.

1. CRM webhook
2. Create the company folder
3. Save each credit report
4. Save each bureau report
5. Generate a sharing link
6. Update the card

Create a folder, create a file, fetch emails: subflows the others call over HTTP. Fix it in one place and it is fixed everywhere.

Before: the same logic copied into every flow. After: one fix applies to all of them.

1. Receive HTTP call
2. Run the single action
3. Respond to the caller