Pricing Model

Pricing Models do not exist as documented artifacts. Loan officers set rates based on personal experience, gut feel, and whatever competitor rate sheets they last saw. A commercial lender might price a five-year term loan at prime plus 200 basis points because 'that is what we always charge good customers,' while another lender in the same branch prices a similar deal at prime plus 275 because they heard rates were going up. There are no written spread matrices, no documented profitability thresholds, no formal relationship pricing guidelines, and no minimum return-on-equity calculations. When a borrower asks why their rate is higher than a competitor's offer, the loan officer improvises an explanation. Management has no way to know whether the institution is systematically underpricing risk or leaving margin on the table because no pricing framework exists to measure against. Fee schedules for origination fees, unused line fees, and prepayment penalties are equally informal — each loan officer has their own sense of what to charge. Exam findings about inconsistent pricing across protected classes cannot be defended because there is no documented methodology to demonstrate objectivity or fair lending compliance. Every loan is priced as a one-off negotiation, and the institution's aggregate pricing strategy is nothing more than the sum of individual lender habits.

None — AI cannot optimize or even analyze pricing because no formal pricing model, rate sheet, or spread matrix exists. Every deal is a black box of individual judgment.

A basic Pricing Model exists as a spreadsheet or PDF rate sheet maintained by the CFO or treasury team. It specifies a base index — typically prime rate or SOFR — and a minimum spread for each major product category such as commercial real estate, C&I lines, and consumer installment loans. The rate sheet is distributed quarterly via email, though many loan officers still reference the prior quarter's version without realizing it has changed. Relationship pricing exceptions are handled informally: a senior lender calls the credit administrator and negotiates a rate concession for a large depositor, but this conversation is not documented as a formal exception. These exceptions are not tracked systematically, so the institution cannot measure how much revenue is being conceded through relationship pricing. The pricing model does not incorporate borrower risk grade, loan-to-value ratio, collateral type, or any profitability analysis. It is a blunt instrument — one spread for all commercial real estate regardless of property type, leverage, tenant quality, or market conditions. Fee schedules for origination, closing, and ongoing fees exist in a separate document that may or may not align with the rate sheet assumptions. The model provides a floor but not the analytical rigor to ensure pricing decisions generate adequate risk-adjusted returns.

AI can look up the current base rate and apply the stated spread, but cannot risk-adjust pricing or evaluate whether exceptions are eroding profitability because the model lacks risk segmentation and exception tracking.

The Pricing Model is a structured document with spread matrices segmented by product type, borrower risk grade, loan-to-value tier, and term. A commercial real estate loan for an investment property with a risk grade 5 borrower and 75% LTV has a different target spread than a risk grade 3 owner-occupied deal at 60% LTV. The model specifies the base index (SOFR, prime, or Treasury), the credit spread component, a liquidity premium for longer terms, and a minimum return-on-equity threshold that each deal must clear. Fee schedules — origination fees, unused line fees, prepayment penalties, and annual renewal fees — are codified in the same document with clear ranges by product and deal size. Relationship pricing exceptions require documented approval with a stated rationale and a compensating factor such as deposit balances or fee income. However, the pricing model is a standalone artifact disconnected from the loan origination system, the treasury cost-of-funds calculations, and portfolio performance data. Loan officers consult the matrix and manually key the rate into the origination system, introducing transcription errors and interpretation inconsistencies. The model assumes a static cost of funds that treasury updates quarterly, which means mid-quarter rate movements are not reflected in pricing guidance. There is no feedback loop from loan performance to validate whether the spread matrix produces adequate returns.

AI can recommend a target rate for a given deal based on the matrix inputs, but cannot validate whether the recommended rate achieves the institution's actual cost-of-funds hurdle because the model is not connected to treasury or portfolio performance systems.

The Pricing Model is an integrated component of the loan origination workflow. When a loan officer structures a deal, the system automatically calculates the target rate by pulling the current base index from treasury, applying the credit spread from the risk-grade matrix, adding a liquidity premium based on term and repricing frequency, and comparing the result against the minimum return-on-equity threshold. Relationship pricing concessions are formally governed — each exception requires a documented business justification, approval authority based on the concession magnitude, and an offset analysis showing compensating revenue from deposits, fee income, or cross-sell opportunities. The model links to the ALCO cost-of-funds curve so that pricing reflects the institution's actual marginal funding cost for each tenor on the yield curve. Competitive rate intelligence from rate-watch services feeds into a comparison module that shows where the institution's offer sits relative to local and national market benchmarks. The pricing engine calculates the deal's projected return on equity, return on assets, and economic value added before the loan officer presents the offer to the borrower. Portfolio concentration limits influence pricing recommendations — when the institution approaches its CRE concentration ceiling, pricing for incremental CRE deals tightens automatically. However, the pricing model does not yet incorporate dynamic borrower lifetime value modeling or real-time macroeconomic scenario stress testing.

AI can auto-price deals within policy, flag exceptions that require approval, simulate profitability under different rate scenarios, and benchmark against competitive offerings. Cannot yet dynamically adjust pricing based on real-time portfolio concentration or borrower lifetime value models.

The Pricing Model exists as an entity in a knowledge graph with typed relationships to the cost-of-funds curve, portfolio concentration maps, borrower lifetime value models, competitive rate intelligence feeds, macroeconomic scenario engines, and regulatory fair-lending compliance monitors. Pricing is no longer a lookup — it is a computation that considers the deal's marginal contribution to portfolio risk, the borrower's total relationship value including deposits and fee revenue, the institution's current liquidity position, and the competitive landscape for the specific geography and product type. The graph encodes which pricing strategies have historically maximized risk-adjusted return by borrower segment, enabling the system to learn from outcomes across thousands of deals. Fair-lending analytics run continuously against the pricing graph, flagging statistical disparities across protected classes before they become compliance issues or examination findings. An AI agent can query: 'For 7-year fixed-rate CRE loans to risk grade 4 borrowers with relationship deposits exceeding $1M, what pricing strategy maximizes lifetime value while maintaining competitive positioning in the Dallas MSA?' The graph provides not just a rate but an explained pricing strategy grounded in historical outcomes, competitive context, and regulatory constraints. ALCO strategy translates directly into pricing parameters that the graph enforces across every deal.

AI can autonomously price deals considering portfolio-level risk, borrower lifetime value, competitive positioning, and fair-lending compliance. It can simulate the portfolio impact of pricing strategy changes and recommend ALCO-level adjustments to the spread framework.

The Pricing Model is a living, continuously recalculating system. There is no static rate sheet — pricing is generated dynamically for each deal at the moment of structuring, reflecting real-time SOFR and Treasury curves, the institution's current cost-of-funds by tenor, today's portfolio concentration position, the borrower's complete relationship value across all products and services, competitive offers currently in the market, and fair-lending compliance constraints across all protected classes. When the Federal Reserve announces a rate change, every pipeline deal's pricing adjusts within minutes without manual intervention. When a large loan funds and shifts portfolio concentration toward a regulatory threshold, pricing for similar deals tightens automatically to discourage further concentration. The system learns from every funded loan's actual performance — deals that outperform their projected ROE strengthen the pricing parameters that produced them, while underperformers trigger spread adjustments for analogous future deals. Pricing is simultaneously optimized across four dimensions: deal-level profitability, portfolio-level risk management, competitive positioning, and regulatory compliance. The institution's pricing strategy is not a document reviewed quarterly — it is a continuously evolving computation that expresses ALCO's risk appetite through every rate offered to every borrower in real time.

Fully autonomous pricing optimization. AI continuously recalculates, recommends, and learns from every pricing decision across the portfolio, achieving real-time balance between profitability, risk, competitiveness, and compliance.

Ceiling of the CMC framework for this dimension.