Skip to content

Deal desk recommendation support

Evaluate commercial options, policy constraints, and approval thresholds to recommend a governed deal path without directly executing the transaction.

Metadata

  • Pattern id: deal-desk-recommendation-support
  • Pattern family: Recommend / Decide / Escalate
  • Problem structure: Recommendation and decision support (recommendation-and-decision-support)
  • Domains: Finance (finance), Compliance (compliance), Operations (operations)

Workflow goal

Produce a defensible recommendation for how to structure, approve, or escalate a proposed commercial deal based on pricing guardrails, policy constraints, and downstream operational implications.

Inputs

Proposed deal package

  • Description: The requested commercial terms, products, quantities, discounts, commitments, and exceptions under consideration.
  • Kind: proposal
  • Required: Yes
  • Examples:
  • Enterprise renewal with expanded seats and custom payment terms
  • Net-new services package requesting a nonstandard discount

Policy and approval framework

  • Description: Pricing guardrails, authority matrices, legal constraints, revenue rules, and escalation thresholds that shape what can be recommended.
  • Kind: policy
  • Required: Yes
  • Examples:
  • Discount thresholds requiring finance approval
  • Contracting rules for nonstandard indemnity language

Historical and contextual evidence

  • Description: Prior deal precedents, customer context, forecast impact, and operational delivery considerations relevant to the decision.
  • Kind: record-set
  • Required: No
  • Examples:
  • Similar closed deals with approved exception paths
  • Margin impact analysis and implementation capacity notes

Stakeholder feedback

  • Description: Inputs from account teams, finance reviewers, compliance owners, or operations leads about risks and trade-offs.
  • Kind: narrative
  • Required: No
  • Examples:
  • Sales notes on strategic customer importance
  • Compliance concern about export or data-handling obligations

Outputs

Ranked deal path recommendation

  • Description: Ordered options with rationale, trade-offs, and a recommended approval or escalation route.
  • Kind: recommendation
  • Required: Yes
  • Examples:
  • Recommend standard terms with a capped discount and finance sign-off
  • Recommend executive escalation because requested terms exceed delegated authority

Decision support packet

  • Description: Evidence bundle linking the recommendation to pricing rules, precedents, risk notes, and unresolved questions.
  • Kind: case-packet
  • Required: Yes
  • Examples:
  • Margin summary with cited guardrail exceptions
  • Approval matrix showing which stakeholders must review the deal

Escalation trigger list

  • Description: Explicit conditions that should stop local approval and route the deal to a higher authority.
  • Kind: escalation-rules
  • Required: Yes
  • Examples:
  • Escalate if requested discount exceeds the regional threshold
  • Escalate if nonstandard payment timing changes revenue recognition treatment

Environment

Operates in governance-sensitive commercial workflows where deal quality depends on balancing win probability, margin, policy compliance, and operational feasibility without hiding the rationale.

Systems

  • CRM and quoting systems
  • Pricing and approval policy repositories
  • Finance and contract review tools
  • Forecasting and delivery planning systems

Actors

  • Account team or requester
  • Deal desk analyst
  • Finance or compliance approver
  • Commercial operations lead

Constraints

  • Recommendations must reflect current approval matrices and policy thresholds.
  • Strategic context may influence option ranking but should not override non-waivable constraints silently.
  • The output must separate recommendation rationale from final decision authority.
  • Exceptions and unresolved policy issues must remain visible to approvers.

Assumptions

  • Relevant pricing, approval, and margin data can be retrieved in time to support deal review.
  • Final decision rights are clearly assigned outside the agent workflow.
  • Reviewers can inspect the evidence packet before acting on the recommendation.

Capability requirements

  • Retrieval (retrieval): The workflow needs historical precedents, policy thresholds, and contextual records before a recommendation can be justified.
  • Synthesis (synthesis): Commercial, operational, and governance inputs must be combined into a decision-ready packet rather than remaining as disconnected facts.
  • Recommendation (recommendation): The central deliverable is a ranked next-best path with clear trade-offs and escalation guidance.
  • Policy and constraint checking (policy-and-constraint-checking): Pricing guardrails, approval limits, and compliance rules materially determine which options are viable.
  • Verification (verification): Key assumptions such as discount bands, precedent fit, and required approvers should be checked before advice is surfaced.
  • Memory and state tracking (memory-and-state-tracking): The workflow must preserve rationale, prior exceptions, and reviewer feedback across iterative deal revisions.

Execution architecture

  • Tool-using single agent (tool-using-single-agent): One recommendation agent can usually retrieve deal context, compare options, and prepare a coherent decision packet within a bounded review loop.
  • Human in the loop (human-in-the-loop): Human stakeholders remain embedded because the workflow is explicitly about informing, not replacing, governed commercial decisions.

Autonomy profile

  • Level: Recommendation only (recommendation-only)
  • Reversibility: Recommendations can be revised as deal terms change, but weak advice can still waste negotiation time, distort forecasts, or bias approvers toward a poor outcome.
  • Escalation: Escalate when non-waivable policy constraints are implicated, precedent fit is weak, requested terms exceed delegated authority, or material legal or revenue-treatment uncertainty remains.

Human checkpoints

  • Confirm the deal scope, strategic context, and authority boundary before recommendation generation begins.
  • Review the ranked recommendation and evidence packet before any quote, contract change, or escalation action is taken.
  • Approve or reject proposed exception rationales when the requested terms exceed standard policy bands.

Risk and governance

  • Risk level: High (high)
  • Failure impact: Poor recommendations can erode margin, create compliance or contract exposure, delay revenue, and push approvers toward decisions that are difficult to unwind after customer communication.
  • Auditability: Preserve the options considered, policy checks applied, historical precedents consulted, reviewer comments, and final rationale for why one path was recommended or escalated.

Approval requirements

  • Human approval is required before recommendation output is used to finalize pricing, contract exceptions, or customer commitments.
  • Policy owners must approve any override path that treats strategic context as sufficient reason to bypass standard thresholds.

Privacy

  • Limit exposure of sensitive customer, pricing, and contractual information to the minimum needed for governed review.
  • Apply retention and access controls consistent with commercial confidentiality obligations.

Security

  • Restrict write access so recommendation tooling cannot directly alter quotes or approval records.
  • Log policy lookups, recommendation revisions, and human overrides for later inspection.

Notes: High-risk governance is appropriate because the pattern influences consequential commercial decisions even though it stops short of execution.

Why agentic

  • The workflow must weigh policy limits, commercial context, and operational trade-offs rather than apply one static rule.
  • Useful recommendations depend on stateful comparison to precedents, reviewer feedback, and evolving deal terms across multiple revisions.
  • The system must know when confidence or authority is insufficient and surface escalation instead of forcing a misleading answer.

Failure modes

The recommendation overweights win probability and underweights governance constraints

  • Impact: Approvers are nudged toward an attractive but noncompliant or financially unsound deal structure.
  • Severity: high
  • Detectability: medium
  • Mitigations:
  • Keep non-waivable constraints explicit and impossible to downgrade silently.
  • Show trade-offs and blocked options in the decision packet rather than only the preferred path.

Historical precedents are applied out of context

  • Impact: The workflow recommends terms that appear justified but do not match current policy or delivery realities.
  • Severity: medium
  • Detectability: medium
  • Mitigations:
  • Compare precedent age, segment, and approval basis before using it as support.
  • Require explicit notes when a recommendation relies on analogical rather than directly matching history.

Required approvers or escalation paths are omitted

  • Impact: The deal advances with incomplete governance coverage and must be reworked late in the cycle.
  • Severity: high
  • Detectability: high
  • Mitigations:
  • Verify approver determination against the current authority matrix.
  • Include unresolved approval dependencies in the output packet.

Recommendation state is not updated after deal terms change

  • Impact: Reviewers act on stale rationale that no longer matches the current proposal.
  • Severity: medium
  • Detectability: high
  • Mitigations:
  • Tie recommendation versions to explicit deal revisions.
  • Re-run policy and margin checks whenever material terms change.

Evaluation

Success metrics

  • Reviewer acceptance rate of recommendations without major policy or approver corrections.
  • Time to produce a complete recommendation packet for governed deal review.
  • Frequency with which escalations are triggered before customer-facing commitments exceed authority.

Quality criteria

  • Recommendations clearly separate viable options, blocked options, and the rationale for the preferred path.
  • Policy constraints, approval needs, and unresolved uncertainties remain visible to decision-makers.
  • Recommendation outputs stay synchronized with the current deal revision and supporting evidence.

Robustness checks

  • Test with deals that mix strategic importance and hard policy limits to ensure the workflow escalates rather than rationalizes exceptions silently.
  • Test sparse precedent coverage and verify the output degrades into explicit uncertainty instead of false precision.
  • Test late-stage term changes and confirm approval routing and rationale are recomputed.

Benchmark notes: Strong evaluation should measure both commercial usefulness and governance discipline; faster approval support is not helpful if it normalizes unsafe exceptions.

Implementation notes

Orchestration notes

  • Keep retrieval, option comparison, policy checking, and escalation packaging as explicit stages so reviewers can inspect where a recommendation came from.
  • Preserve prior recommendation versions alongside reviewer feedback to avoid losing decision context across deal revisions.

Integration notes

  • Common implementations integrate CRM, quoting, pricing, contracting, and approval matrix systems.
  • Keep the pattern neutral about any specific CPQ, CRM, or contract platform.

Deployment notes

  • Start with recommendation visibility only before connecting the workflow to approval-routing automation.
  • Monitor disagreement between recommendations and final approved outcomes to tune policy interpretation carefully.

References

Example domains

  • Finance (finance): Recommend a renewal structure that balances discount policy, margin targets, and approval thresholds.
  • Compliance (compliance): Evaluate whether a cross-border commercial exception requires escalation because contractual commitments alter compliance exposure.
  • Operations (operations): Recommend whether a proposed services package is operationally feasible before routing it for executive approval.

Grounded instances

Canonical source

  • data/patterns/recommend-decide-escalate/deal-desk-recommendation-support.yaml