Domain 1: Sales Lifecycle
- Opportunity stages should mark buyer-verifiable progress, not internal handoffs; stages driven by internal activity inflate the count and blur the probability mapping that forecast categories depend on.
- Forecast categories map many stages to one bucket, which is what keeps the forecast readable when a process has many stages; the value derives from the stage and can be overridden per record.
- The opportunity amount becomes a roll-up of line items once opportunity products are used, so the design must make adding products part of the process rather than optional.
- An opportunity can carry several quotes with exactly one synced, which is what stops competing proposals all claiming to be the deal's value at once.
- Lead conversion carries custom fields across only where a mapping is defined with compatible data types; without one the value stops at the lead, which is why data appears to vanish.
- Path puts stage-specific key fields and guidance on the record, which is how a sales methodology becomes something reps follow rather than something they were taught once.
- Campaign attribution needs both halves: campaign members link people to activity, campaign influence links campaigns to the opportunities those people are on.
- Record types with their own sales processes are the standard answer where new business and renewals genuinely differ; one shared stage list makes stage reporting ambiguous for both.
Domain 2: Consulting and Implementation Strategies
- A requirements conflict is a governance event: document both positions and escalate to the body that holds authority, rather than settling it by seniority or building both.
- Sandbox choice follows the data requirement and the refresh interval together: Full copies everything but refreshes every 29 days, Partial Copy samples via a template and refreshes every 5.
- Validation-only deployment runs the whole process including tests then discards the result, and a passing validation stays eligible for quick deploy for four days.
- Phasing buys feedback and concentrates support at the cost of a longer split state, which delays consistent reporting and usually adds temporary bridging integration.
- Load parents before children, and use an external ID marked unique so repeat loads upsert rather than duplicate; migrations are almost always run more than once.
- High logins with low record activity means access works and the process does not fit the work; that is an enablement and process finding, not a technology one.
- Middle management is the leverage point for adoption, because teams take their cue from their direct manager regardless of how much end-user training is delivered.
- Cutover plans need a named decision owner and pre-agreed go or no-go criteria, because nobody designs a safe rollback under time pressure on the night.
Domain 3: Practical Application of Agentforce Sales Expertise
- An agent is structured as topics, each carrying its own instructions and the actions available inside it; scoping actions to the topic is the main control on predictable behaviour.
- Actions are backed by flows, Apex, API calls, standard actions or prompt templates, and parameter descriptions matter because the reasoning engine uses them to decide what to pass.
- Confidently wrong answers usually mean missing grounding, not insufficient instructions: a model generates plausible text unless supplied with retrieved facts.
- Wrong topic selection is a topic description problem; the engine chose before any instruction inside the topic applied, so sharpen or merge the descriptions.
- The Trust Layer covers the AI path specifically: masking outbound, toxicity scoring inbound, zero retention, prompt injection defence and an audit trail of prompts and responses.
- Agent data access follows the permissions of the identity it runs as, so that identity is a security design decision rather than a setup detail.
- An absolute prohibition is enforced by removing the capability from topics and actions, not by instructing against it; instructions are followed probabilistically.
- Agentforce is consumption priced, so design efficiency has a direct commercial effect and volume forecasting belongs in the design rather than the invoice.
Domain 4: Data Management
- Roll-up summary fields require master-detail, which is why the account cannot roll up opportunity amounts natively: that relationship is a lookup.
- Query efficiency at volume needs a selective filter on an indexed field; a non-selective filter forces a full scan however well indexed the field is.
- Ownership skew slows sharing recalculation when one user owns very many records; account data skew causes parent lock contention when one parent has very many children.
- Salesforce Connect leaves data in the source and retrieves on access, so it consumes no storage and inherits the source's availability and response time.
- Removing a field from a page layout is not a security control: the field stays reachable through reports, list views and the API. Field-level security removes all of those.
- Restriction rules only narrow access, layering on top of the sharing model rather than replacing the organisation-wide default.
- Merging reparents the losing record's related records to the survivor and lets the user pick which field values live, which is why merging beats deleting a duplicate.
- Data storage and file storage are allocated separately, so check which is actually constrained before choosing between archiving records and offloading files.
Domain 5: Predictive and Generative AI
- Predictive models learn from historical records and their recorded outcomes, so outcome history is the binding prerequisite and no volume of attribute data substitutes for it.
- Inconsistent outcome recording teaches wrong patterns, and the resulting scores look exactly like good ones, which is what makes label quality the highest-leverage data work.
- Surface the contributing factors alongside a score; a bare number tells a seller where a deal ranks without telling them what to do about it.
- Threshold setting trades the two error types directly, so it is a business decision made on the relative cost of a missed case against a false alarm.
- Predictive estimates an outcome and generative produces content; the common pattern is prediction selecting the deals and generation drafting the outreach.
- Fluent output is generated the same way whether or not it is true, so confidence in tone is not evidence of accuracy and grounding is what makes an answer checkable.
- Every model has a bounded context window covering instructions, grounding and history together, which is why retrieval must be selective rather than exhaustive.
- Benefit is only realised when the replaced manual step actually stops; deployments that show adoption but no benefit have almost always skipped that retirement.
Salesforce Agentforce Sales Consultant exam tips
- Read every scenario for the constraint. A stated latency tolerance, an absolute prohibition, an obligation to evidence a decision, or a note that the data is already reachable is usually what eliminates three otherwise workable options.
- When a question asks what to do first, the answer is nearly always to establish something rather than to build something: recover the underlying need, confirm the obligation, or route the decision to whoever owns it.
- Watch the boundary between what an agent may do and what needs a person. Retrieval, drafting and summarising are safe because they are reviewed before they matter; commitments, concessions and customer-facing decisions are not.
- Distinguish detection from prevention. Reports and audits find breaches afterwards; validation rules, approval processes and removing a capability from an agent's action list stop them happening.
- For anything AI, ask whether the failure would be a data problem or a configuration problem. If the records support the correct answer and the output is still wrong, it is configuration; if they do not, it is data.
- Several answers will be technically possible. Prefer the one whose cost of ownership is lowest over years: configuration before code, one authoritative field before two copies, and the platform's own mechanism before a hand-built equivalent.
Study guide FAQ
What format and passing requirements does the Salesforce Certified Agentforce Sales Consultant exam have?
The exam is 60 scored multiple-choice questions plus up to 5 unscored, with 105 minutes to complete it and a passing score of 73% in English (69% in Japanese). Registration is 200 USD plus applicable taxes, and the Salesforce Administrator certification is a prerequisite.
Is this the same exam as Sales Cloud Consultant?
It is the renamed successor, but the rename came with a re-scope rather than being a relabel. Sections 3 and 5 put roughly 37% of the blueprint on Agentforce and AI content the classic Sales Cloud Consultant exam did not cover, so preparation based only on older Sales Cloud material leaves a substantial gap.
How much of the exam is about Agentforce specifically?
Practical Application of Agentforce Sales Expertise is 24% and Predictive and Generative AI is a further 13%. The first covers designing, deploying, testing and governing agents in sales work; the second covers the predictive scoring and generative capabilities underneath, including what each needs from the data.
Do I need hands-on Agentforce experience to pass?
The questions are consultant-level rather than administrator-level, so they test design judgement about scope, grounding, handoff and measurement more than click paths. Hands-on time still helps considerably, because the reasoning about topic selection and grounding quality is much easier to hold once you have watched an agent route a request badly.
Which section deserves the most preparation time?
Consulting and Implementation Strategies is the largest at 25% and the broadest, running from discovery and governance through sandboxes, deployment, migration, change management and cutover. If your background is configuration rather than delivery, that section is usually where the gap sits rather than in the product knowledge.
Related Salesforce resources
- Salesforce Agentforce Sales Consultant practice exam
- Salesforce practice exams
- Certification path
- Salesforce Certified Agentforce Specialist study guide
- Salesforce Certified Data 360 Consultant study guide
- Salesforce Certified Platform Administrator (ADM-201) study guide
- Certification exam guides & tips
- Pricing & plans
- FAQ