Deal Investigator

Deal Investigator

What this skill does

For a single opportunity, produces a structured review answering five questions a CRO would ask in a deal review:

1. Is this deal real?
2. Is the close date credible?
3. What is the actual risk?
4. What do comparable deals tell us?
5. What is the recommendation: commit, best case, omit, or escalate?

The output is a one-page deal memo, formatted for a deal review meeting.

Upstream context

This skill is typically invoked by salesforce-revops-audit once per flagged opportunity when the audit's Deal Integrity dimension surfaces individual deals with 3+ risk flags. It can also be run directly when a manager wants to pressure-test a specific deal they have in mind. Unlike the other skills in this pack, this one operates on a single opportunity, so the audit will recommend running it multiple times in sequence when multiple deals are flagged.

When to invoke

Invoke when the user says any of:

• "investigate this deal"
• "deal review for [opp name or id]"
• "is this deal real"
• "pressure-test [opp name]"
• "deep dive on this opportunity"
• "should we call this in commit"

Do not invoke for:

• Full-team forecast prep (use forecast-call-prep)
• Pipeline-wide hygiene scanning (use pipeline-hygiene-audit)
• Multi-deal portfolio analysis

Data sources, in order of preference

1. Salesforce MCP: query the specific opportunity, its account, recent activity, and comparable closed opportunities (SOQL in Appendix A).
2. CSV or JSON: read opportunities.csv and identify the target opp by id or name.
3. Sample data: bundled synthetic dataset, with the user providing an opp_id from the dataset.

Column discernment

Real Salesforce exports rarely match canonical names exactly. Custom suffixes (__c), renamed fields, and different cases are normal. Before parsing any data file, read docs/column_mapping.md and use it to map the export's actual headers to the canonical fields this skill needs.

Procedure (full detail in docs/column_mapping.md):

1. Normalize each header in the export (lowercase, strip __c, replace _ and . with space, drop noise tokens).
2. Score each header by token overlap against the canonical field's header_tokens, subtracting for exclusion_tokens hits.
3. Confirm the top candidate with a value fingerprint (pull two or three sample rows and check the values against the catalog's value_fingerprint).
4. If two headers tie above threshold, ask the user one question to disambiguate. If no header passes for a required field, ask the user to name the column. Do not guess.

Note the mapping in a one-line footnote at the bottom of the deal memo: Mapping: matched N of M required fields from .

Process

Step 1: Identify the target opportunity

User provides either an opp ID (opp_0042), an account name ("Acme Industries"), or partial description ("the Acme renewal Diego is working on"). If ambiguous, list the top three matches and ask which one. If exact match, proceed.

Step 2: Gather the deal facts

Pull from the data source:

• Opportunity name, account, owner, segment, amount, stage, probability, forecast category
• Created date, close date, last activity date, days since last activity
• Next step
• Stage history if available
• Activity history if available

Step 3: Identify comparable deals

Find 3-5 closed opportunities (won and lost) that share at least two of: same owner, same segment, similar amount band (within 50% of the target), same general industry signal from account name. These become the comparable set.

For each comparable, note: outcome (won/lost), days from creation to close, stage at which it was lost (if lost), final amount.

Step 4: Compute the risk indicators

Score each indicator. Each is a binary risk flag.

| Indicator | Risk if | |---|---| | Stale activity | dayssincelastactivity > 14 | | Missing next step | nextstep is empty | | Close date plausibility | closedate 1.5x average for that stage for this owner | | Amount vs comparable | target amount > 2x typical comparable amount | | Forecast category alignment | category = Commit or BestCase but probability = LASTN_DAYS:60 ORDER BY ActivityDate DESC

-- Comparable closed deals from same owner and segment SELECT Id, Name, Account.Name, Amount, StageName, CloseDate, CreatedDate, IsWon FROM Opportunity WHERE OwnerId = '[OWNERID]' AND IsClosed = true AND CloseDate = LASTN_DAYS:365 ORDER BY CloseDate DESC LIMIT 20

If stage history is needed and the org has the OpportunityHistory object, supplement with a query against that.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [elijeangilles](https://github.com/elijeangilles)
- **Source:** [elijeangilles/revops-skills](https://github.com/elijeangilles/revops-skills)
- **License:** MIT
- **Homepage:** https://revopseval.com

Install and usage instructions live in the source repository linked above.