Treating Deal Discounts and Contract Exceptions as Decisions: The Asterion Deal Desk Demo

🎥 The YouTube version is also available:

Asterion Deal Desk | AI-Assisted Commercial Decision, Approval, and Execution

Book: Decision Trace Model Practical Implementation Guide: A Practical Guide to Connecting AI-Powered Judgment to Real-World Organizational Implementation

Discounts, payment terms, complimentary implementation support, and SLA exceptions are not conditions that sales can resolve alone. Sales wants to avoid losing the revenue opportunity. Finance needs to protect margin and collections. Legal seeks to limit contractual risk. And business leaders must consider the customer strategy that lies ahead. Each perspective is legitimate.

The problem is not deciding which department is right. It is that evidence, alternatives, unresolved questions, and approval paths tend to become scattered across email and chat, making it difficult to explain later why a deal was accepted on particular terms.

Asterion Deal Desk is a deal-decision support demo built around this challenge. It is not a system that lets AI automatically approve a discount. AI presents multiple deal structures and the rationale for each one; people make the decision within their authority. The complete process is retained as a Decision Trace, connecting the eventual outcome to the next decision.

Do not compress a deal into a single approval

Looking only at price is not enough when approving a deal. Even a large discount can have a very different meaning depending on the potential for future expansion, payment terms, the burden of implementation support, and the nature of any contract exceptions.

The demo follows a proposal to Northstar Retail Group for a next-generation store analytics platform. The initial request includes the following issues:

• A 28% discount, exceeding the normal pricing policy
• A projected gross margin of 49.7%, below the 55% threshold
• Payment 90 days after acceptance, increasing collection risk
• Complimentary implementation support, reducing profitability
• A request to treat part of the SLA as an exception to standard terms
• A late-stage competitive selection process and a strategic possibility of expanding to multiple locations in the future

What such a deal needs is not a rushed choice between “approve” and “reject.” It needs a concrete way to preserve the opportunity while identifying which conditions should change, what needs further confirmation, and whose approval is required to proceed.

AI does not declare one answer; it presents comparable deal structures

Rather than passing the initial request through unchanged, Asterion Deal Desk compares several proposals.

In this example, AI recommends Option B. The first-year discount is limited to 22%, followed by a phased return to standard pricing from the second year onward. Payment terms are reduced to 60 days, and complimentary implementation support becomes a paid option. The SLA exception is separated from the pricing decision and routed to legal review as a limited clause. The possibility of expansion to three locations is not treated as established fact; confirmation of the customer’s intent becomes a condition for approval.

The important point is that the AI recommendation is not presented as “the right answer.” Option B is one proposal whose three-year expected gross profit, win probability, collection risk, and contract risk can be compared. People decide whether to adopt it and how far to revise its conditions.

Keep evidence, hypotheses, and unknowns separate

An AI proposal may sound plausible, but it is not useful for decision-making if its composition is unclear. The demo therefore separates each interpretation into the following categories:

• Established rules: pricing policy, credit policy, standard SLA terms, and similar controls
• Evidence-based estimates: gross-margin performance in comparable deals and estimated collection losses arising from payment terms
• Hypotheses and unresolved matters: the possibility of future expansion, competitive-price information, and the detailed scope of an SLA exception

For example, the assessment that “a 90-day payment term increases collection risk” is linked to the credit policy and a loss estimate based on the difference in terms. By contrast, “the deal may expand to three locations” can have strategic value, but it is not a verified fact. Giving the latter the same weight as the former allows an appealing expectation to distort the decision.

The interface displays the evidence consulted, the AI’s interpretation, and the questions that require human confirmation separately. This is also a design for ensuring that what AI does not know is not treated as “no problem.”

Do not leave authority outside the interface

Different roles can handle different parts of a deal change. A sales representative can initiate a request, add supporting evidence, revise terms, and resubmit it. But that person cannot unilaterally approve pricing, payment, and contract exceptions.

When conditional approval is selected in the demo, the required approvers are shown according to the content of the deal:

• Business owner: discount and strategic exceptions
• Finance reviewer: payment terms and credit conditions
• Legal reviewer: contract exceptions such as SLA terms

When the interface switches roles, the available actions change accordingly. Approval outside a user’s authority is blocked in the interface, and approval requests are sent in the required sequence. The key is not merely documenting authority as an operating rule. It is enforcing authority at the boundary between decision and execution.

A Decision Trace is broader than a log

The core of this demo is the Decision Trace. It is not simply an activity history; it preserves the causal path that led to a decision.

Deal signal
  → Evidence collection
  → Interpretation (risks, opportunities, unresolved matters)
  → Multiple proposals
  → Human revision of terms and decision
  → Approval based on authority
  → Execution
  → Outcome and learning

The interface links evidence—such as the customer profile, comparable deals, competitive intelligence, pricing policy, credit policy, and standard SLA terms—to the proposal, the human decision, and each approval. Someone reviewing the case later can see not only who approved it, but also which evidence was used, what remained unresolved, and how the terms were changed.

This structure supports more than auditability and handover. It makes it easier for sales, finance, and legal to share the assumptions behind a decision. When a similar deal arrives, the organization can refer not only to the outcome but also to the conditions that produced it.

Do not perform AI correctness after the deal is won

Evaluating AI only against estimates made at the time of contract can easily turn into a retrospective narrative that favors the system. Asterion Deal Desk records outcomes after execution—such as payment collection, first-year gross profit, the number of departments using the product, and the likelihood of further expansion—and compares them with expectations at the time of the decision.

However, Options A and C were not executed, so they have no observed outcomes. Their comparison values are reference values generated by the model available at that time. This distinction matters. Keeping actual results separate from counterfactual simulations avoids presenting the story as though “AI was always right.”

The outcome is added to the Decision Trace and returned as material for similar future deals. Learning here does not mean only automatically updating a model as a black box. It also means enabling the organization to examine how combinations of conditions affected profitability, collection, contracts, and customer strategy.

Deal AI needs more than discount-rate optimization

When a deal-support AI is designed as a function that produces the “optimal discount rate,” much of the real decision falls outside the frame. Practice requires consideration of the relationship among price, gross margin, payment, contract terms, strategic value, strength of evidence, unresolved matters, approval authority, and outcomes after execution.

Asterion Deal Desk is not a demo designed to replace human decision-makers with AI. It is a workspace for helping people make better decisions and retaining those decisions as organizational knowledge.

AI proposes. People make accountable decisions. Evidence, decisions, authority, execution, and outcomes are retained as one continuous flow. Preserving that order is the starting point for combining speed and control in deal automation.

Chinoba

Chinoba

Intelligence as Relationship

Research Platform

founded by

Masao Watanabe

AI Systems Architecture
Decision Trace
Human–AI Coordination
Algorithmic Governance