CRM Stack Migration

Building Attio for Gen H. Replacing HubSpot with a CRM that models how the business actually works, then layering in automation and AI to make BDMs measurably more effective.

Why We're Moving

HubSpot costs ~$5k/month and fails at all three things it does: CRM doesn't model the business (no broker/brokerage/network hierarchy), email marketing is weak because the data feeding it is weak, and live chat is fine but overpriced. The core problem: bad tooling is actively making the sales team worse. They can't find the right prospects, can't see the right context, can't act on the right signals.

Contract expires December 2026. Comfortable timeline.

Target Stack

What We've Built So Far

Data model (6 custom objects, 49+ attributes)

Seed data

• 10 brokers across 8 brokerages, linked to 5 networks and 5 clubs
• 20 cases across all pipeline stages with realistic data
• Rich interaction notes on brokers (call notes, meeting notes, strategy context)
• Brokerage-level strategy notes (account plans, risk flags)
• Case context notes (underwriter queries, valuations, offers)
• Tasks with deadlines linked to relevant records

Display configuration

Each object's Record Text (Workspace settings > Objects > Appearance) must be set to the human-readable field. Without this, the UI shows UUIDs everywhere.

Attio Pro Platform Capabilities

Workflows (automation engine)

Triggers: Record created, record/attribute updated, recurring schedule (daily/weekly/cron), webhook received, manual button, task created, record added to list.

Actions: Create/update records, create tasks, complete tasks, enrol in sequences, if/else branching, switch (multi-path), filter, delay, formula calculations, aggregate values (sum/avg/min/max), Slack messages, in-app broadcast alerts, HTTP requests, AI actions (classify, summarise, prompt completion), loop, parse JSON.

Limits: 10,000 credits/month on Pro. Find Records capped at 100 results per query. HTTP requests timeout after 20 seconds.

Sequences (email drips)

Multi-step email campaigns. Smart sending (timezone-aware, business hours only). Auto-pause on out-of-office. Auto-exit on reply. Threading (conversation replies). Can be triggered manually or automatically via workflow "Enrol in Sequence" action.

Calculated fields

No native formula or rollup field types. Workaround: workflow-based calculations using Formula, Aggregate Values, and Adjust Time blocks that write results to regular fields. Values update when the workflow runs, not in real time. Good enough for daily metrics.

Kanban

Custom fields on cards (any attribute as a card row). Native "Track time in stage" - toggle per stage, set target duration, counter appears on cards and goes red when exceeded. Sorting by time in stage. Stage-level aggregations (sum, count).

Notifications

In-app notifications for @mentions, task assignments, daily task digest. Condition-based alerts require workflows: Broadcast Message (in-app popup) or Slack integration. No native "notify me when X condition is met" without a workflow.

Open Design Question: Interaction Type Modelling

Sara wants structured tracking of how the team interacts with brokers - not just "a note was added" but "this was a sales meeting" vs "this was a half-day hot-desking session" vs "this was a phone call." With drop-downs, not free text.

The problem: Attio's native activity system (timeline entries, logged meetings, calls) has a fixed schema with no custom attributes. You cannot add an "interaction type" select field to a native activity entry.

Air Call Integration

Air Call is Gen H's exclusive telephony platform. Every broker call goes through it. This is core infrastructure, not a nice-to-have.

Gen H's Air Call plan

Custom Professional, £3,585/month. 31 users, 15 numbers, 9 teams. AI Assist included in plan (call summaries, key topics, sentiment, action items) - was not enabled for any users until 26 Feb. Now enabled for one inbound BDM as a test; will roll out to full team if summary quality is good. No additional cost.

Why the native Attio integration won't work

A first-party Attio + Air Call integration exists (shipped early 2025). It auto-creates notes on Person records after calls. But it only logs to Person records, not custom objects. Our brokers are a custom object. It also doesn't pass through AI Assist summaries, recordings, or transcriptions, and can't trigger Attio workflows.

Recommended: Air Call webhook → Cloudflare Worker → Attio API

Air Call fires webhook events on call end. The payload includes: phone number, direction, duration, timestamp, agent, tags, notes, recording URL, and (with AI Assist enabled) an AI-generated call summary.

A lightweight Cloudflare Worker receives the webhook and:

1. Matches the phone number against Attio's Brokers object
2. Creates an Interaction record (type: Phone Call, duration, direction, AI summary, linked broker) - for structured querying
3. Creates a note on the broker record (clean summary) - for the activity timeline
4. Updates broker attributes (last interaction date, total calls, last call direction)
5. If no matching broker: flags as potential new lead, creates a task for the BDM

This is better than the native integration because we control the schema and get AI summaries flowing through. It also establishes the template for other inbound data feeds (Ignite, email tracking, event attendance) - same pattern. Sub-second latency, near-zero hosting cost, ~100 lines of code. The BDM hangs up and the system does the rest.

The CRM Vision: What "Great" Looks Like for BDMs

The CRM should be proactive, not passive. A BDM opens Attio on Monday morning and their day is planned. They don't have to remember who to call, which cases are stuck, or which brokers are going cold. The system tells them.

Pipeline management (keeping what we have)

Auto-generated tasks at every stage transition and time threshold. The BDM's task list IS their daily plan.

Pipeline filling (getting more)

This is where CRM creates real edge. Four broker segments, each with different data requirements:

Automatic workflows

Email sequences

Sara's taxonomy: welcome, nurture, nudge, win-back. Each sequence needs an ownership decision - who does it send from?

Ownership model (confirmed with Sara 26 Feb):

• General/long-cycle campaigns (win-back, brand nurture): send from GenH Marketing, but include BDM name, details, and "book a meeting" link. Goal: keep the BDM's name in the broker's mind even when marketing owns the campaign.
• Short-cycle, case-specific nudges (DIP follow-up, 3-day chase): send from the BDM personally. Automated but personalised. Rationale: brokers deal with 30+ lenders and forget names. A timely, personal-feeling email from "Sarah" is more effective than one from "hello@genh.com".

Claude Integration Roadmap

Attio proves its value standalone first. Claude is the accelerant, layered in after trust is established.

Example morning brief:

1. GH-2026-016 is stuck. Rebecca Stone's case has been in Full App for 14 days. Lucy Harrison hasn't heard from us. Call her today with a concrete update or she writes us off.

2. David Patel is becoming your best broker. 4 cases in 3 weeks, 100% completion rate. Still off-panel at SPF. The panel review on 15 March is your window - bring his stats.

3. Emma Thompson has gone quiet. Her one case is at Valuation. If it's clean, this is the moment she decides if we're a one-off or a regular lender. Don't let the result land without a call.

Demo Build List

What to build now to demonstrate value to the sales team.

In Attio UI

1. Turn on "Track time in stage" on Cases kanban - set targets: DIP Submitted 3d, Full App 7d, Valuation 5d, Offer Issued 10d. Cards go red when stuck.
2. Workflow: case stuck auto-task - daily schedule, find cases exceeding stage target, create task for BDM.
3. Workflow: new broker first case - on case created where broker case count = 1, create welcome call task.
4. Workflow: gone cold detection - daily schedule, find brokers with last submission > 30 days, flip to "Cooling", create re-engagement task.
5. Workflow: offer issued task - case stage changes to Offer Issued, create relationship call task.
6. Workflow: case completed task - case stage changes to Completion, create post-completion call task.
7. Email sequence: new broker welcome - 3 emails (day 0, 3, 7), wired to workflow #3.
8. Build views - "My Tasks Today" (filtered, sorted by priority), "Pipeline Health" (cases by days in stage), "Broker Activity" (sorted by last submission), "Growth Opportunities" (low-volume brokers by brokerage).

Via API

1. Add "Referred By" field to Brokers - record-reference to Brokers. Backfill Tom Richards → Priya Sharma.

Demo narrative: "You open Attio on Monday morning. Your task list tells you who to call and why. Cases that are stuck are flagged before the broker chases you. New brokers get a welcome sequence without you lifting a finger. Brokers going cold get caught at 30 days, not 90. Every offer and completion triggers a relationship touchpoint. And you can see your entire pipeline health in one view. HubSpot can't do this because it doesn't understand brokers, brokerages, and networks. Attio can because we built it to match how Gen H actually works."

Open Questions for Attio Team

Timeline

The Uncomfortable Question

Attio API Reference

Learned by trial and error against the live API (25 Feb 2026). Treat as ground truth over the Attio docs, which are inconsistent in places.

Basics

• Base URL: https://api.attio.com/v2/
• Auth: Authorization: Bearer <api_key> header. Key generated in Settings > Developers.
• Pro tier required (£68/user/month) for API access, custom objects, and workflows. Free/Plus tiers have no API.
• Rate limits: 100 reads/sec, 25 writes/sec.
• No official Python or Node SDKs. Raw HTTP via Python's urllib. Community TypeScript SDK (npm add attio-js) untested.
• MCP servers exist but untested: official at mcp.attio.com/mcp (OAuth), community attio-mcp on npm (API key).

Object and attribute creation

• Custom objects can only be created in the Attio UI. No POST endpoint for object definitions.
• Objects get an api_slug from their name (e.g. "Brokerages" → brokerages). Use GET /v2/objects to discover slugs.
• Each custom object starts with 3 system attributes: record_id, created_at, created_by.

Creating attributes - POST /v2/objects/{object_slug}/attributes:

Body must be wrapped in data with all fields present:

{
  "data": {
    "title": "Loan Amount",
    "description": "Requested loan amount",
    "api_slug": "loan_amount",
    "type": "currency",
    "is_required": false,
    "is_unique": false,
    "is_multiselect": false,
    "config": {"currency": {"default_currency_code": "GBP", "display_type": "symbol"}}
  }
}

• config is required even if empty ({}). Omitting it causes a validation error.
• For record-reference types, config: {"record_reference": {"allowed_objects": ["target_slug"]}}. Slugs work, not just UUIDs.
• Cannot create via API: email-address and domain types (system-only on People/Companies). Use text as workaround on custom objects.
• Tested types: text, number, currency, date, select, status, phone-number, record-reference.

Select and status options

select attributes require options created before use. Setting a value to a non-existent option returns an error.

Create select options: POST /v2/objects/{object}/attributes/{attribute_id}/options

{"data": {"title": "Active"}}

status attributes use POST /v2/objects/{object}/attributes/{attribute_id}/statuses:

{"data": {"title": "DIP Submitted", "target_archive_state": "active"}}

target_archive_state options: "active", "archived-win", "archived-loss". Setting a status on a record requires the status UUID.

Creating records

POST /v2/objects/{object_slug}/records:

{"data": {"values": {"name": "Sarah Mitchell", "fca_number": "SM001"}}}

Value formats by type:

• text: "name": "Sarah Mitchell"
• number: "ltv": 75
• currency: plain number, not object. "loan_amount": 350000 works. {"currency_code": "GBP", "value": 350000} does NOT.
• date: "submitted_date": "2026-02-10"
• select: string matching option title
• status: "case_stage": [{"status": "uuid"}]
• record-reference: "broker": [{"target_object": "brokers", "target_record_id": "uuid"}]

Querying records

POST /v2/objects/{object_slug}/records/query. Filter syntax uses attribute slugs as keys:

{"filter": {"reference": "GH-2026-003"}}
{"filter": {"property_postcode": {"value": {"$contains": "E"}}}}
{"filter": {"loan_amount": {"currency_value": {"$gt": 400000}}}}
{"filter": {"broker": {"target_record_id": {"$eq": "broker-uuid"}}}}

Sorting:

{"sorts": [{"attribute": "loan_amount", "field": "currency_value", "direction": "desc"}]}

Reading values from results - all values are arrays:

record["values"]["name"][0]["value"]                    # text, number
record["values"]["loan_amount"][0]["currency_value"]    # currency
record["values"]["broker"][0]["target_record_id"]       # record-reference
record["values"]["submitted_date"][0]["value"]          # date

Updating records

PATCH /v2/objects/{object_slug}/records/{record_id}:

{"data": {"values": {"bdm_owner": "Alex Turner", "days_in_stage": 5}}}

Notes and Tasks

Notes: POST /v2/notes - requires format field ("plaintext" or "markdown"). Field is content, not content_plaintext. Notes appear on record timeline.

Tasks: POST /v2/tasks - requires format, assignees (can be empty array), and linked_records.

API limitations

Cannot do: create custom objects, create views/kanban, create workflows, create sequences, create Lists (attempted, payload rejected), create email-address/domain attributes, trigger enrichment.

Untested but available: webhooks (POST /v2/webhooks), assert/upsert (POST /v2/objects/{slug}/records/assert), comments/threads, bulk operations (25 writes/sec = ~90k records/hour).

Shell gotcha

API key corrupted by shell interpolation in bash/curl (especially heredocs). Symptom: 401 Unauthorized with "wrong length". Fix: read key from file in Python.