DealScopeAI/dealscope-salesforce-ai-brief-dataset-v1

---
pretty_name: DealScope Salesforce AI Brief Dataset v1
language:
- en
tags:
- salesforce
- crm-ai
- enterprise-llm
- text-generation
- synthetic
task_categories:
- text-generation
size_categories:
- n<1K
---

# DealScope Salesforce AI Brief Dataset v1

## Dataset Summary

This dataset contains `25` structured Salesforce-record brief examples in the DealScope output format.

Each record is shaped like a real DealScope API response and includes:

- record metadata
- buying signals
- risks
- stakeholders
- a draft follow-up email
- a multi-line summary

The dataset is intended as a public retrieval and reference asset for Salesforce-focused AI brief workflows.

## What Is In This Release

- `2` anonymized records derived from real DealScope API outputs and then redacted for public release
- `23` synthetic records generated in the same public schema to cover a wider range of DealScope use cases

The synthetic records were created from the real DealScope response shape and from documented product workflows such as:

- lead and opportunity briefs
- account handoff and review summaries
- risk and buyer-signal extraction
- follow-up drafting
- LinkedIn context import
- HubSpot context import

## Intended Use

This release is designed for:

- public product reference and retrieval
- demos of structured CRM brief outputs
- experiments around Salesforce-oriented summarization and signal extraction
- comparison pages, notebooks, and benchmark examples

## Not Intended Use

This release should not be treated as:

- a source of real customer identities
- a full CRM activity archive
- an authoritative benchmark of production quality across all organizations
- legal, procurement, or compliance guidance

## Data Structure

Each line in `train.jsonl` is one JSON object with this high-level shape:

```json
{
"version": "1.0",
"anonymized": true,
"record": {
"type": "Opportunity",
"recordId": "opp-5751c636"
},
"signals": {
"buying": [],
"risks": []
},
"stakeholders": [],
"draftEmail": {
"subject": "",
"body": ""
},
"summary": "",
"metadata": {
"summariesUsed": 32,
"summariesLimit": 1000
}
}
```

## Field Notes

- `record.type` is one of `Lead`, `Opportunity`, or `Account`
- `record.recordId` is always anonymized
- `signals.buying` and `signals.risks` contain structured reasoning blocks with `label`, `evidence`, `severity`, and `why`
- `stakeholders` contains anonymized participant labels plus role and confidence
- `draftEmail` is a ready-to-edit follow-up suggestion
- `summary` is a short multiline account, lead, or opportunity brief
- `metadata` contains lightweight usage telemetry where available

## Data Creation

The public schema follows the real DealScope API response format.

The release process:

1. start from the real DealScope output shape
2. anonymize record ids, people, company references, and sensitive evidence strings
3. preserve the structured signal and follow-up logic
4. generate additional synthetic examples in the same schema to improve coverage

## Anonymization And Safety

The release uses these protections:

- Salesforce ids are replaced with synthetic hashed ids
- contact names are replaced with placeholders like `{{primary_contact}}`
- internal sender names are replaced with `{{seller_name}}`
- company references are normalized with `{{account_name}}` where needed
- raw note bodies are redacted when they contain sensitive or hostile text

## Limitations

- most records are synthetic, not direct customer exports
- the dataset reflects a specific structured brief format rather than full CRM history
- some fields intentionally compress or neutralize original evidence for safety
- the release is small and should be treated as `v1`, not a complete corpus

## Release Contents

- `README.md`
- `train.jsonl`
- `field-descriptions.json`

## Suggested Citation

If you reference the dataset publicly, cite it as:

`DealScope Salesforce AI Brief Dataset v1`