Discovery and Architecture

Section 5: Use in Requirements Gathering and Discovery

5.1 Core uses

Codex and Claude can help with:

• converting business needs into epics, features, user stories, and tasks
• surfacing ambiguous statements and missing decisions
• generating acceptance criteria
• identifying edge cases and exception paths
• preparing workshop questions
• creating discovery questionnaires
• mapping business processes to Salesforce capabilities
• producing BRD/FRD drafts
• performing gap analysis between current state and target state

5.2 Recommended split

• Claude: discovery synthesis, BRD/FRD drafting, workshop questions, process decomposition
• Codex: validating whether requested behavior maps to current metadata/code patterns, identifying technical debt in existing implementation

5.3 Use case examples

#### Lead management

AI can:

• convert lead lifecycle workshops into stories like dedupe, routing, SLA follow-up, and conversion
• identify missing requirements such as round-robin logic, territory exceptions, duplicate handling, and assignment overrides
• map needs to Sales Cloud, Assignment Rules, Flow, duplicate rules, and custom objects where needed

Sample prompt

Act as a senior Salesforce business analyst and solution architect.
Using the workshop notes below, create:
1. epics
2. features
3. user stories
4. acceptance criteria
5. edge cases
6. Salesforce feature mapping

Business domain: lead management
Priority: standard Salesforce first, custom only where necessary
Workshop notes:
[paste notes]

#### Loan processing

AI can:

• decompose intake, document validation, underwriting review, approval routing, and disbursement
• identify where standard objects are insufficient and custom objects are needed
• generate process questions around KYC, consent, audit, and exception handling

#### Customer service workflow

AI can:

• define service request lifecycle
• map case types, entitlement checks, escalation triggers, and omni-channel routing
• identify KB, bots, handoff, and SLA reporting needs

#### Approval flows

AI can:

• identify approval levels, delegated approvers, fallback approvers, bypass logic, and audit requirements
• draft approval matrices

#### Partner portal requirements

AI can:

• convert partner onboarding notes into Experience Cloud capability backlog
• identify profile, permission, sharing, and role model questions

5.4 Discovery framework

5.5 Checklist

• Have all primary personas been identified?
• Are exception paths documented?
• Are success metrics defined?
• Are integration boundaries identified?
• Are compliance constraints captured?
• Are assumptions and open questions logged?

Section 6: Use in Solution Design and Architecture

6.1 What AI can do well in architecture

AI can rapidly turn requirements into:

• solution options
• standard-vs-custom recommendations
• object and field models
• automation strategy options
• security model candidates
• integration patterns
• migration strategies
• multi-org versus single-org considerations
• NFR lists
• risk matrices

6.2 Standard versus custom analysis

Prompt template

Act as a Salesforce solution architect.
Given these requirements, propose three solution options:
1. standard-first
2. hybrid
3. custom-heavy

For each option include:
- Salesforce features used
- custom build required
- pros
- cons
- risks
- scalability implications
- maintainability implications
- security considerations
- recommended option

6.3 Design areas

#### Object model and field design

AI can draft:

• entity relationships
• field purpose tables
• ownership rules
• indexing considerations
• archival concerns

#### Automation strategy

AI can help decide:

• record-triggered flow versus Apex
• before-save versus after-save flow
• when subflows are appropriate
• where async patterns are needed

#### Integration architecture

AI can outline:

• sync versus async tradeoffs
• middleware versus direct API
• platform event use
• retry and idempotency strategy
• ownership of transformation logic

#### Security model

AI can structure:

• role hierarchy
• profile/permission set strategy
• sharing model
• field-level security matrix
• sensitive data controls

#### Service architecture

AI can help with:

• omni-channel design
• case routing
• entitlement model
• escalation model
• service console design

#### Experience Cloud design

AI can support:

• audience segmentation
• external user sharing
• authentication patterns
• content architecture
• portal self-service journeys

#### Org strategy and operating topology

AI can help frame:

• single-org versus multi-org decision factors
• regional, business-unit, or regulatory partitioning needs
• integration and data residency implications
• support and CoE operating model consequences

#### Non-functional requirements and scalability

AI can draft NFR sets covering:

• transaction volumes
• latency expectations
• resiliency and retry posture
• auditability
• maintainability
• observability
• supportability
• large data volume considerations

#### Performance considerations

AI can help identify:

• Flow recursion and transaction depth risks
• unselective query patterns
• synchronous integration bottlenecks
• UI rendering and LWC state churn issues
• high-volume sharing recalculation concerns

#### Data migration planning

AI can structure:

• object sequencing
• mapping sheets
• validation rules to disable/enable
• cutover windows
• reconciliation metrics

6.4 Architecture review use

AI is valuable in architecture review boards for:

• design challenge questions
• anti-pattern detection
• tradeoff summaries
• readiness checklists
• ADR drafting

6.5 Pros/cons matrix example

6.6 Human review required

• Validate every design option against actual licensing, org constraints, and integration ownership.
• Do not accept AI recommendations for sharing or encryption without security review.
• Pressure-test AI-generated designs against transaction volume and support model.