Investigation Framework
Phase 1 — Define the Inquiry
Before touching any source, characterize what you need to find:
- Definitive fact: A single correct answer exists (e.g., "Does the API support batch requests?")
- Contextual picture: Multiple viewpoints must be assembled (e.g., "What has this account's experience been over the last quarter?")
- Open exploration: The boundaries of the question are still forming (e.g., "Why are enterprise customers struggling with onboarding?")
- Audience awareness: Determine who will consume the answer — the customer directly, an internal team member, or executive leadership — as this shapes tone, depth, and what to include.
Phase 2 — Map Sources to the Question Type
| Question Category |
Primary Sources |
| Product capabilities |
Official docs, API references, knowledge base, product specs |
| Account history and context |
CRM records, email threads, meeting notes, chat logs |
| Internal process or policy |
Wikis, runbooks, policy documents |
| Technical troubleshooting |
Documentation, engineering resources, past support tickets |
| Market or competitive landscape |
Web search, analyst reports, competitive intelligence |
Phase 3 — Gather Evidence Across Tiers
Work through sources from most to least authoritative. Never rely on a single hit — corroborate across channels.
Tier 1: Canonical Internal Records (Reliability: Strong)
Official documentation, published knowledge base articles, API references, policy and SLA documents, internal product roadmaps. Trust these unless they carry stale dates.
Tier 2: Organizational Memory (Reliability: Moderate-Strong)
CRM account notes and activity logs, resolved support tickets and known-issue databases, shared documents and specifications, recorded decisions from meeting notes.
Tier 3: Informal Team Channels (Reliability: Moderate)
Slack or chat discussions, email correspondence, calendar and agenda notes. These often hold the freshest information but may lack full context or reflect speculation.
Tier 4: Public and External Sources (Reliability: Variable)
Company and competitor websites, community forums and user discussions, partner documentation, industry news and analyst commentary.
Tier 5: Reasoning by Analogy (Reliability: Weak)
Precedent from similar past situations, patterns from comparable accounts, general industry conventions. Always label these as inference.
Phase 4 — Assemble the Answer
Pull together findings, flag any contradictions between sources, and assign an overall confidence rating.
Phase 5 — Attribute Everything
Every claim in your response must trace back to a named source. No unsourced assertions.
Confidence Rating System
Attach one of these ratings to every answer you deliver:
Strong Confidence
- Backed by canonical documentation or an authoritative internal record
- Corroborated by at least two independent sources
- Verified as current
- Phrasing: "This is well-supported by [source]."
Moderate Confidence
- Found in informal channels (chat, email) without official documentation backing
- Rests on a single uncorroborated source
- Possibly dated but likely still accurate
- Phrasing: "According to [source], this seems correct. I would suggest confirming with [team/person]."
Weak Confidence
- Derived from indirect or analogical reasoning
- Sources are old or of uncertain reliability
- Multiple sources conflict with one another
- Phrasing: "No definitive source was found. Based on [context], my working assessment is [answer]. Verification is recommended before communicating to the customer."
Insufficient Information
- No relevant material located anywhere
- The question demands expertise beyond what available sources cover
- Phrasing: "I was unable to locate relevant information. I recommend consulting [suggested expert or team]."
Resolving Source Conflicts
When different sources tell different stories:
- Surface the discrepancy explicitly — never silently pick one version
- Evaluate recency and authority of each conflicting source
- Lay out both positions with context for the reader
- Propose a path to resolution (e.g., check with the product team)
- For customer-facing answers, default to the most conservative position until the conflict is settled
Structured Response Format
**Answer:** [Lead with the bottom line]
**Confidence:** [Strong / Moderate / Weak]
**Evidence:**
- [Source A]: [What it states]
- [Source B]: [Corroborating or contrasting detail]
**Limitations:**
- [Conditions or edge cases that could change the answer]
- [Context-specific factors to be aware of]
**Next Steps:**
- [Whether this is safe to relay to the customer as-is]
- [Any recommended verification actions]
Deciding When to Escalate
Safe to Answer Directly
- Official docs address the question unambiguously
- Multiple trustworthy sources agree
- The topic is factual and non-sensitive
- No commitments about timelines, pricing, or legal terms are involved
Route to a Specialist
- Roadmap commitments or delivery timelines are in play
- The question touches pricing, contracts, or legal terms
- Security, compliance, or data-handling topics
- The response could set a binding precedent or create expectations
- Conflicting information was found and remains unresolved
- A customer-specific configuration is involved
- The account is at risk and an incorrect answer could worsen the situation
Escalation Directory
- Domain experts — technical or specialized knowledge gaps
- Product team — capability, roadmap, or feature questions
- Legal / compliance — regulatory, privacy, or contractual matters
- Finance / billing — pricing, invoicing, payment issues
- Engineering — bugs, custom setups, root-cause analysis
- Leadership — strategic decisions, policy exceptions, high-stakes situations
Capturing Research for Reuse
Worth Documenting When:
- The same question has surfaced before or is likely to recur
- The investigation required substantial effort
- The answer corrects a widespread misconception
- Nuance exists that is easy to get wrong
Reusable Entry Format
## [Topic or Question]
**Verified On:** [date]
**Confidence:** [rating]
### Finding
[Concise, direct answer]
### Context and Nuance
[Background, conditions, and subtleties]
### Source Trail
[Where this information originated]
### Adjacent Topics
[Related questions this entry may also answer]
### Freshness Notes
[When to re-check, what developments could invalidate this]
Maintaining the Knowledge Store
- Timestamp every entry
- Mark entries tied to specific product versions or feature states
- Schedule periodic reviews to retire stale content
- Tag entries by domain, product area, and customer segment for discoverability
Operating Principles
- Pin down exactly what you are looking for before searching
- Work through source tiers methodically — do not skip levels on assumption
- Corroborate across multiple channels whenever possible
- Never present uncertain findings as established fact — always surface your confidence level
- When unsure whether an answer is safe to share externally, verify first
- Record your findings so the next person does not repeat the work
- If your research uncovers a knowledge gap, flag it for documentation
1---
2name: customer-intel
3description: Research customer context across docs, tickets, CRM, email, and web sources with cited confidence.
4metadata:
5 source: https://skillmd.com/skills/vm0-ai/customer-intel
6---
7
8## Investigation Framework
9
10### Phase 1 — Define the Inquiry
11
12Before touching any source, characterize what you need to find:
13
14- **Definitive fact**: A single correct answer exists (e.g., "Does the API support batch requests?")
15- **Contextual picture**: Multiple viewpoints must be assembled (e.g., "What has this account's experience been over the last quarter?")
16- **Open exploration**: The boundaries of the question are still forming (e.g., "Why are enterprise customers struggling with onboarding?")
17- **Audience awareness**: Determine who will consume the answer — the customer directly, an internal team member, or executive leadership — as this shapes tone, depth, and what to include.
18
19### Phase 2 — Map Sources to the Question Type
20
21| Question Category | Primary Sources |
22|---|---|
23| Product capabilities | Official docs, API references, knowledge base, product specs |
24| Account history and context | CRM records, email threads, meeting notes, chat logs |
25| Internal process or policy | Wikis, runbooks, policy documents |
26| Technical troubleshooting | Documentation, engineering resources, past support tickets |
27| Market or competitive landscape | Web search, analyst reports, competitive intelligence |
28
29### Phase 3 — Gather Evidence Across Tiers
30
31Work through sources from most to least authoritative. Never rely on a single hit — corroborate across channels.
32
33**Tier 1: Canonical Internal Records** (Reliability: Strong)
34Official documentation, published knowledge base articles, API references, policy and SLA documents, internal product roadmaps. Trust these unless they carry stale dates.
35
36**Tier 2: Organizational Memory** (Reliability: Moderate-Strong)
37CRM account notes and activity logs, resolved support tickets and known-issue databases, shared documents and specifications, recorded decisions from meeting notes.
38
39**Tier 3: Informal Team Channels** (Reliability: Moderate)
40Slack or chat discussions, email correspondence, calendar and agenda notes. These often hold the freshest information but may lack full context or reflect speculation.
41
42**Tier 4: Public and External Sources** (Reliability: Variable)
43Company and competitor websites, community forums and user discussions, partner documentation, industry news and analyst commentary.
44
45**Tier 5: Reasoning by Analogy** (Reliability: Weak)
46Precedent from similar past situations, patterns from comparable accounts, general industry conventions. Always label these as inference.
47
48### Phase 4 — Assemble the Answer
49
50Pull together findings, flag any contradictions between sources, and assign an overall confidence rating.
51
52### Phase 5 — Attribute Everything
53
54Every claim in your response must trace back to a named source. No unsourced assertions.
55
56## Confidence Rating System
57
58Attach one of these ratings to every answer you deliver:
59
60### Strong Confidence
61- Backed by canonical documentation or an authoritative internal record
62- Corroborated by at least two independent sources
63- Verified as current
64- Phrasing: "This is well-supported by [source]."
65
66### Moderate Confidence
67- Found in informal channels (chat, email) without official documentation backing
68- Rests on a single uncorroborated source
69- Possibly dated but likely still accurate
70- Phrasing: "According to [source], this seems correct. I would suggest confirming with [team/person]."
71
72### Weak Confidence
73- Derived from indirect or analogical reasoning
74- Sources are old or of uncertain reliability
75- Multiple sources conflict with one another
76- Phrasing: "No definitive source was found. Based on [context], my working assessment is [answer]. Verification is recommended before communicating to the customer."
77
78### Insufficient Information
79- No relevant material located anywhere
80- The question demands expertise beyond what available sources cover
81- Phrasing: "I was unable to locate relevant information. I recommend consulting [suggested expert or team]."
82
83## Resolving Source Conflicts
84
85When different sources tell different stories:
86
871. Surface the discrepancy explicitly — never silently pick one version
882. Evaluate recency and authority of each conflicting source
893. Lay out both positions with context for the reader
904. Propose a path to resolution (e.g., check with the product team)
915. For customer-facing answers, default to the most conservative position until the conflict is settled
92
93## Structured Response Format
94
95```
96**Answer:** [Lead with the bottom line]
97
98**Confidence:** [Strong / Moderate / Weak]
99
100**Evidence:**
101- [Source A]: [What it states]
102- [Source B]: [Corroborating or contrasting detail]
103
104**Limitations:**
105- [Conditions or edge cases that could change the answer]
106- [Context-specific factors to be aware of]
107
108**Next Steps:**
109- [Whether this is safe to relay to the customer as-is]
110- [Any recommended verification actions]
111```
112
113## Deciding When to Escalate
114
115### Safe to Answer Directly
116- Official docs address the question unambiguously
117- Multiple trustworthy sources agree
118- The topic is factual and non-sensitive
119- No commitments about timelines, pricing, or legal terms are involved
120
121### Route to a Specialist
122- Roadmap commitments or delivery timelines are in play
123- The question touches pricing, contracts, or legal terms
124- Security, compliance, or data-handling topics
125- The response could set a binding precedent or create expectations
126- Conflicting information was found and remains unresolved
127- A customer-specific configuration is involved
128- The account is at risk and an incorrect answer could worsen the situation
129
130### Escalation Directory
1311. **Domain experts** — technical or specialized knowledge gaps
1322. **Product team** — capability, roadmap, or feature questions
1333. **Legal / compliance** — regulatory, privacy, or contractual matters
1344. **Finance / billing** — pricing, invoicing, payment issues
1355. **Engineering** — bugs, custom setups, root-cause analysis
1366. **Leadership** — strategic decisions, policy exceptions, high-stakes situations
137
138## Capturing Research for Reuse
139
140### Worth Documenting When:
141- The same question has surfaced before or is likely to recur
142- The investigation required substantial effort
143- The answer corrects a widespread misconception
144- Nuance exists that is easy to get wrong
145
146### Reusable Entry Format
147```
148
149## [Topic or Question]
150
151**Verified On:** [date]
152**Confidence:** [rating]
153
154### Finding
155[Concise, direct answer]
156
157### Context and Nuance
158[Background, conditions, and subtleties]
159
160### Source Trail
161[Where this information originated]
162
163### Adjacent Topics
164[Related questions this entry may also answer]
165
166### Freshness Notes
167[When to re-check, what developments could invalidate this]
168```
169
170### Maintaining the Knowledge Store
171- Timestamp every entry
172- Mark entries tied to specific product versions or feature states
173- Schedule periodic reviews to retire stale content
174- Tag entries by domain, product area, and customer segment for discoverability
175
176## Operating Principles
177
1781. Pin down exactly what you are looking for before searching
1792. Work through source tiers methodically — do not skip levels on assumption
1803. Corroborate across multiple channels whenever possible
1814. Never present uncertain findings as established fact — always surface your confidence level
1825. When unsure whether an answer is safe to share externally, verify first
1836. Record your findings so the next person does not repeat the work
1847. If your research uncovers a knowledge gap, flag it for documentation