The commercial decision
Contract renewals bring together usage, contract terms, account history and client context. The difficulty is deciding what that evidence means for the relationship. A complete set of facts can still leave the assumptions behind a recommendation unclear.
I initiated the project and was its sole lead, owning problem definition, commercial requirements, workflow design, AI-assisted development and iterative evaluation. I used AI development tools to build, test and refine the prototype. Data specialists advised on source definitions and retrieval; commercial stakeholders contributed feedback.
Combining evidence with human context
I designed a two-pass workflow because account records alone cannot explain a client’s objectives or the reasons behind a change in demand. A recommendation needed to remain open to information from the seller.
- Initial assessment. Assemble the evidence, offer a provisional view and ask focused questions about missing commercial context.
- Recommendation for review. Incorporate the seller’s answers, reassess the position and prepare the case for a pricing reviewer.
This separates evidence from interpretation and interpretation from approval. The prototype can question assumptions and suggest a position; the seller can amend it, and people retain the commercial decision.
Feedback from commercial colleagues helped me focus the brief on what mattered for the renewal decision, with supporting evidence kept separate. I weighed whether deeper analysis would help someone make a better decision enough to justify the effort of producing and reviewing it.
Allocating work between code, AI and people
I assigned bounded calculations and validation checks to executable code, while using AI to interpret evidence, ask questions and draft the recommendation. That division made the numerical work inspectable without treating the model’s commercial judgement as authority.
Requirements for source meaning and missing information supported that design. I used case testing and targeted technical evaluation to refine the logic and the recommendations, assessing the complete answer as well as its component calculations.
Evaluating the whole decision process
I defined value in terms of the combined work of the seller and reviewer, alongside decision quality. A shorter exchange or a quick approval could be a misleading success measure: less preparation by one person can create more work for another, and a difficult case may deserve more discussion.
The evaluation approach therefore considered whether the advisor helped assemble relevant evidence, expose assumptions and produce a reviewable recommendation. These were criteria for testing the intended organisational benefit, rather than claims of measured savings.
Work produced
I built a versioned AI advisor prototype with structured evidence requirements, executable calculation and validation helpers, a two-pass workflow and a documented evaluation approach. The work connects commercial requirements to a practical decision-support product and a way to assess its usefulness.