Skip to main content
Technology

HubSpot Custom Objects for AI: How to Structure Conversation Data

HubSpot's standard objects aren't built for AI conversation data. Learn exactly which custom objects to create, and follow a step-by-step implementation.

Author: Agent360 editorial team
Published: 20 April 2026
Last updated: 8 November 2026
Reading time: 16 min read

Why should you use HubSpot Custom Objects for AI data?

You structure AI conversation data in HubSpot by creating Custom Objects instead of squeezing transcripts, sentiment scores and SPIN notes into the standard objects. The standard objects aren't built for the depth and structure that AI analyses produce, and a dedicated custom object lets you store, relate and report on conversation data directly in the CRM.

You structure AI conversation data in HubSpot by creating Custom Objects (user-defined data objects such as a "Sales Call" object) instead of squeezing transcripts, sentiment scores and SPIN notes into the standard objects (Contacts, Companies, Deals). The standard objects aren't built for the depth and structure that AI analyses produce. A dedicated custom object lets you store, relate and report on conversation data directly in the CRM, without hacking existing fields.

TL;DR, the key points in 30 seconds:

  • HubSpot's standard objects (Contacts, Companies, Deals, Tickets) lack structure for AI-generated conversation data such as transcript segments, sentiment scores and SPIN notes
  • Custom Objects let you store and relate conversation data directly in HubSpot, without hacking existing fields
  • Properly structured conversation data significantly reduces manual CRM updating time and markedly increases data quality

Last updated: March 2026

What are HubSpot Custom Objects?

HubSpot Custom Objects are an Enterprise feature in HubSpot CRM that lets you define entirely new data objects beyond the four standard objects (Contacts, Companies, Deals and Tickets). A custom object works like a new "table" in your CRM database, with its own properties, its own relations to other objects and its own views in the HubSpot interface.

Think of it as the difference between using a standardised spreadsheet and designing your own database. The standard objects cover 80% of needs, but the remaining 20%, which typically involve complex business processes or advanced AI workflows, require custom objects.

Read more about how to make use of HubSpot CRM automation to maximise the effect of your workflows.

For AI-driven sales this means, concretely: you can create a "Sales Call" custom object that stores the transcript, sentiment score, identified SPIN elements, next steps and a relation to both Contact and Deal, all as structured, searchable data in HubSpot.

What does HubSpot's standard data model lack for AI?

HubSpot's four standard objects are built to track who you're talking to and what you're selling. They aren't built to track what is said, how it's said, and what the AI system concludes. According to HubSpot's own statistics, sales teams spend on average 4.8 hours a week on manual CRM updating, time that can be eliminated directly with properly structured custom objects connected to AI. Here are the precise gaps in the standard model:

1. No native structure for conversation data The Deal object has fields for deal stage, amount and close date. It has no fields for "AI-identified objections", "customer's budget mentioned in conversation" or "SPIN: Implication Questions asked". You can add custom properties to existing objects, but you end up with 40+ extra fields on the Deal object, which makes dashboards unreadable. Understand the shift from rule-based workflows to autonomous agents in HubSpot Workflows for AI Agents.

2. Missing 1:many relation to conversations A deal can have 15 calls over 3 months. With standard objects you can only log these as activities (notes/calls), not as structured data points you can analyse across. A "Sales Call" custom object lets you attach 15 separate, structured conversation objects to one deal.

3. No versioning of conversation data AI analysis of sales conversations typically produces: transcript (raw text), summary (structured summary), entity extraction (names, numbers, dates) and sentiment scores. All of this must be stored with a timestamp and version, something standard objects don't support.

4. Scaling to enterprise volume According to Gartner's CRM report 2025, companies with 10+ salespeople have on average 2,400 sales-related activities per month. Squeezing 2,400 AI-analysed conversations into deal notes is a data quality problem, not a solution.

Comparison: HubSpot standard objects vs. custom objects for AI conversation data

Dimension Standard Objects (Deals/Contacts) Custom Objects (Sales Call)
Transcript storage Note/Activity (unstructured text) Dedicated property (structured JSON/text)
Sentiment score Not natively supported Number property (0-100 scale)
Relation to multiple calls 1 deal → N activities (no structure) 1 deal → N call objects (queryable)
AI-identified entities Typed manually in notes Structured properties (budget, timeline, etc.)
Reporting and analysis Not possible across calls Native HubSpot reports on all properties
API access Activity API (limited) Custom Objects API (full CRUD)
Workflow triggers No (activities) Yes, workflows based on any property
Requires HubSpot plan All plans Enterprise (from ~USD 1,200/month)

Which custom objects do you need for conversation data?

The minimal AI-optimised data structure in HubSpot typically consists of three custom objects. Depending on your sales process and its complexity, you can add a fourth.

Custom Object 1: Sales Call (the core object)

This is the foundation. A Sales Call object represents a single sales conversation and stores everything the AI system produces from it.

Required properties:

  • call_date (DateTime), Time of the conversation
  • duration_seconds (Number), Duration in seconds
  • transcript_raw (Multi-line text), Raw transcription from the AI
  • summary_ai (Multi-line text), AI-generated summary
  • sentiment_score (Number, 0-100), Overall sentiment
  • call_type (Dropdown): Outbound, Inbound, Demo, Discovery, Closing
  • ai_next_steps (Multi-line text), Identified next steps
  • budget_mentioned (Number, DKK), Budget mentioned in the conversation
  • timeline_mentioned (Single-line text), Timeframe mentioned
  • objections_identified (Multi-line text), Identified objections
  • spin_situation (Multi-line text), SPIN: Situation questions noted
  • spin_problem (Multi-line text), SPIN: Problem questions noted
  • spin_implication (Multi-line text), SPIN: Implication questions noted
  • spin_need_payoff (Multi-line text), SPIN: Need-Payoff noted
  • recording_url (URL), Link to the audio recording

Relations: Sales Call → Contact (many-to-one), Sales Call → Deal (many-to-one)

Custom Object 2: AI Insight (aggregated insights)

Whereas Sales Call stores per-conversation data, AI Insight stores aggregated conclusions across conversations for a given deal or contact.

Required properties:

  • insight_type (Dropdown): Deal Risk, Buying Signal, Competitor Mention, Price Sensitivity
  • confidence_score (Number, 0-100), The AI's degree of confidence
  • evidence_calls (Multi-line text), Reference to specific call IDs
  • generated_date (DateTime), When the insight was generated
  • action_taken (Checkbox), Whether the salesperson has acted on the insight

Relations: AI Insight → Deal (many-to-one), AI Insight → Contact (many-to-one)

For a complete overview, see our guide to the best CRM automation in 2026.

Custom Object 3: Coaching Session (salesperson performance)

For internal AI coaching you need an object that stores coaching data per salesperson.

Required properties:

  • rep_id (HubSpot User lookup)
  • period (Single-line text), e.g. "2026-W18"
  • calls_analyzed (Number)
  • avg_sentiment (Number)
  • talk_ratio_pct (Number), Salesperson's share of talk time
  • top_strength (Multi-line text), AI-identified strength
  • top_improvement (Multi-line text), AI-identified area for improvement
  • coaching_notes (Multi-line text), Manager notes

Relations: Coaching Session → HubSpot User

Custom Object 4 (optional): Competitor Intelligence

If your AI analyses competitor mentions in conversations, a dedicated Competitor Intelligence object lets you track when and in what context competitors are mentioned.

Step by step: build the AI data structure in HubSpot

Here is the precise implementation order. According to Salesforce's State of Sales 2025, companies spend an average of 3.2 months on CRM implementations, but with this plan you can have a working AI data structure ready in 2-3 weeks.

Step 1: Plan your custom object schemas (Days 1-2)

Before you open HubSpot, document your schemas in a spreadsheet:

  • Object name, internal API name
  • All properties with type and validation
  • All relations (which standard objects are linked)
  • Required fields and default values

Use API-friendly names from the start: sales_call, ai_insight, coaching_session. It saves you problems when you build API integrations.

Step 2: Create custom objects via the API (Days 3-5)

HubSpot's Custom Objects interface is usable for simple cases, but for complex schemas we recommend using HubSpot's CRM API directly. It gives version control and documentation.

``` POST https://api.hubapi.com/crm/v3/schemas Authorization: Bearer YOUR_PRIVATE_APP_TOKEN ```

Use HubSpot's official Custom Objects API documentation as a reference. An important detail: primaryDisplayProperty determines what is shown as the "title" of the object in HubSpot, use call_date or summary_ai for Sales Call.

Compare the leading CRM platforms in our guide to CRM systems in Denmark.

Step 3: Set up associations (Days 5-7)

Associations are the relations between your custom objects and standard objects. Create associations via:

``` POST https://api.hubapi.com/crm/v3/schemas/{objectType}/associations ```

Critical: always create both directions of an association. Sales Call → Contact AND Contact → Sales Call. HubSpot doesn't handle both directions automatically.

Step 4: Build HubSpot workflows on custom objects (Days 8-10)

With Sales Call custom objects you can now build automations that were impossible with standard activities:

  • Deal Stage Update: When sentiment_score falls below 40 on three consecutive calls → Move the deal to "At Risk"
  • Alert to manager: When objections_identified contains "competitor" → Send a Slack notification
  • Coaching trigger: When talk_ratio_pct > 70 (the salesperson talks too much) → Create a Coaching Session task

Step 5: Set up dashboards and reports (Days 11-14)

Create dedicated HubSpot reports that make use of your new data structure:

  • Average sentiment_score per deal stage
  • Conversion rate sorted by talk_ratio_pct
  • Budget quality: deals with budget_mentioned vs. deals without
  • SPIN adoption: which salespeople ask the most spin_implication questions

See our related guide on HubSpot CRM automation in 2026 for a broader review of the automation options.

According to Gartner's analysis of CRM ROI (2025), companies that invest in data quality and structured CRM schemas see a 23% higher win rate and 19% shorter sales cycles than the industry average.

For a deeper understanding of conversation intelligence as the foundation for the data structure, read our guide to what conversation intelligence is.

We have also compared the best CRM automation tools in 2026, if you're considering alternatives to HubSpot.

Technical considerations and pitfalls

Three mistakes that recur when companies implement custom objects for AI data:

Mistake 1: Too many properties on one object It's tempting to gather everything in one big Sales Call object. The problem is that HubSpot's reporting engine has limitations on complex objects with 50+ properties. Keep each object focused and related. Spread data across the three objects described above.

Mistake 2: Forgetting the indexing consequences HubSpot's search function only indexes properties you explicitly mark as searchable. Make sure summary_ai, objections_identified and ai_next_steps are marked for search, otherwise you can't find specific conversations via HubSpot's universal search.

Mistake 3: No data retention policy AI-generated conversation data contains personal data and is subject to GDPR. That requires you to define and implement retention periods at custom object level. HubSpot doesn't support automatic deletion of custom objects, it requires a separate cron job or workflow that calls the Custom Objects DELETE API.

This article is part of our HubSpot series for Danish B2B companies:

See also our complete overview of CRM automation.

Related CRM articles:

Q: What do HubSpot Custom Objects cost? HubSpot Custom Objects are an Enterprise feature and require HubSpot Sales Hub Enterprise or CRM Enterprise. The price starts from about USD 1,200/month for 10 seats (list prices March 2026). With Enterprise you get 10 custom object types and up to 1,000 properties per object. HubSpot typically offers a 10-20% discount on an annual contract and a further discount at volume.

Q: Can I build custom objects on HubSpot Starter or Professional? No. Custom Objects require Enterprise. On Professional you can add custom properties to existing standard objects (Contacts, Companies, Deals, Tickets), but you can't create new object types. It is possible to get a long way with custom properties on the Deal object for simple use cases, but for AI conversation data (with a need for 1:many relations and structured schemas) Enterprise is necessary. Alternatively you can use a separate system (e.g. Airtable or a custom database) connected to HubSpot via Make or Zapier, but this markedly increases maintenance complexity.

Q: How many custom object records can I store? HubSpot Enterprise includes an unlimited number of custom object records (as of March 2026). There is no per-record price. With 10 salespeople making 5 calls a day, you generate ~14,600 Sales Call records per year, which is unproblematic in HubSpot Enterprise. Do note the size limits on individual text properties, though: multi-line text properties are limited to 65,536 characters, which is typically enough for a transcript of ~10,000 words (about a 45-50 minute conversation).

Q: What is the difference between HubSpot Custom Properties and Custom Objects? Custom Properties are extra fields on existing object types. You can, for example, add a "Sentiment Score" property to the existing Deal object. Custom Objects are entirely new object types with their own properties and their own relations. The difference is decisive for AI conversation data: with custom properties on Deal you can only have one sentiment score per deal. With a Sales Call custom object you can have 15 separate sentiment scores (one per conversation) and analyse the development over time.

Q: Is AI conversation data in HubSpot GDPR-compliant? It depends on your implementation. HubSpot's platform itself is GDPR-compliant with EU data centre options. The challenge is your AI analysis pipeline: if you transcribe via a third-party AI service (OpenAI, Anthropic, etc.), you must ensure EU data processing agreements and minimise the amount of data sent (GDPR's principle of data minimisation).

Q: How do I ensure data quality in custom objects over time? Data quality in custom objects requires three things: (1) Required fields, mark critical AI properties as required, so no records are created incomplete; (2) Validation, use HubSpot's property validation for number properties (e.g. sentiment_score: 0-100); (3) Monitoring: build a HubSpot report that shows records with empty critical properties, and review it weekly for the first 3 months. The primary cause of CRM data quality problems is typically not bad technology, but a lack of governance processes.

What are the most important HubSpot API endpoints for custom objects?

For technical teams it is essential to know the precise API endpoints for custom objects. Here are the most important:

Create schema: POST /crm/v3/schemas

Define your object with properties, associations and search settings.

Create record: POST /crm/v3/objects/{objectType}

Insert a new Sales Call record with all properties as JSON.

Create association: PUT /crm/v3/objects/{objectType}/{objectId}/associations/{toObjectType}/{toObjectId}/{associationType}

Link Sales Call to Contact and Deal with separate calls.

Search custom objects: POST /crm/v3/objects/{objectType}/search

Run structured searches with filters, sorts and pagination, far more powerful than standard activity search.

HubSpot's Custom Objects API supports full CRUD as well as batch operations, which let you create up to 100 records in one API call, critical for high-volume AI pipelines.

Practical tip: Use HubSpot's Private App authentication (not OAuth) for server-to-server communication. Private Apps have granular scope control and are easier to rotate than OAuth tokens after security incidents.

Comparison: HubSpot Custom Objects vs. the alternatives

Before you invest in HubSpot Enterprise for custom objects, it's worth considering the alternatives:

Approach Advantage Disadvantage Recommendation
HubSpot Custom Objects Native integration, no external database, reporting out of the box Requires Enterprise (~$1,200+/month) Best for mid-market+ with HubSpot
Airtable + Zapier → HubSpot Cheaper, flexible Latency, maintenance, sync errors Only for prototyping
Salesforce Custom Objects More powerful, broader API More expensive and harder Enterprise with a Salesforce base
Custom database (PostgreSQL) + API Full control Requires developer resources Tech-first companies

Comparison of data structures for AI conversation data 2026

For most Danish B2B companies that already use HubSpot Professional or Enterprise, Custom Objects are the clear recommendation. The alternatives introduce synchronisation problems, extra latency and maintenance complexity that over time exceed the cost of the HubSpot Enterprise upgrade.

Experience from Danish sales departments shows that the road from a "custom database prototype" to "migration to HubSpot Custom Objects" often takes longer than planned, time that is wasted on technical debt instead of sales growth.

See our complete guide to HubSpot alternatives in 2026 if you're assessing whether HubSpot is the right platform for your company.

Conclusion: data is the foundation: structure is the key

Most sales departments misjudge the problem. They think they have an AI problem, that they just need to choose the right AI tool. But the real problem is a data structure problem. AI analysis produces data. If your CRM architecture can't receive, relate and report on that data, you lose most of the value.

HubSpot Custom Objects aren't a luxury problem for enterprise companies. They're the foundation for any company that is serious about using AI to improve the sales process. Without the right structure, AI insights end up as notes in a deal, read by the salesperson but invisible to the organisation.

The transformation starts with the data structure. Choose the right custom objects, define your schemas, and build your AI integration on top of a solid foundation.

Next step:

If you're ready to build your AI data structure with HubSpot Custom Objects as the foundation, you can get advice from Agent360.

Book 20 minutes of technical consultation with the Agent360 team, and get a view on how your HubSpot Custom Objects implementation can fit your sales process and volume.

#hubspot#custom objects#crm#ai#conversation data#sales#data structure

Should your CRM keep itself up to date?

We build the layer that moves data into the CRM you already use, without anyone having to type it in. We look at your current set-up first.

How we built it · CRM & customer overview

Customer CRM that updates itself

An internal page per customer with contacts, the agreement, the projects and a timeline where emails, meetings, email drafts and tasks land by themselves. Above the pages sits the board with all customers, owner and latest contact, and the system suggests next steps based on fixed rules.

  1. Email, meeting, task
  2. Lands on the customer’s page
  3. You take the next step
See the customer’s own page →