The Business Challenge
The business situation the organisation faced before the project began.
Procurement teams often spend significant time manually reviewing purchase requests, interpreting requirements, checking completeness, and determining appropriate procurement actions. This can create inconsistent reviews, repetitive work, and operational delays.
Why AI Was the Right Solution
Why AI was the right approach — and what alternatives were considered.
AI was appropriate because purchase requests are not always expressed in a uniform structure or language. Procurement analysis often requires interpreting context, understanding the relationship between requirements, identifying missing information, and forming recommendations from incomplete or varied inputs.
A traditional rules-based approach could handle clearly defined validation rules, but it would become increasingly complex when dealing with variations in request descriptions and procurement context. AI provides greater flexibility for interpreting unstructured or semi-structured information while still allowing structured workflow steps and human review to remain part of the process.
The objective was therefore not to replace procurement expertise, but to augment it by handling repetitive initial analysis and presenting information in a more consistent structure for human decision-making.
How the Solution Was Delivered
Discovery, design, development, testing and rollout — the journey, not the tooling.
- 1Reviewed real-world procurement processes and identified repetitive activities within purchase request analysis, including requirement interpretation, completeness checks, and procurement action assessment.
- 2Defined the assistant's business objectives and structured the analysis process around extracting key requirements, identifying missing information, and generating consistent procurement recommendations.
- 3Designed an AI-assisted workflow that transforms incoming purchase request information into structured analysis and actionable procurement insights.
- 4Developed the solution as a personal AI project using procurement domain knowledge to guide the workflow and keep recommendations aligned with practical procurement processes.
- 5Tested the workflow using representative purchase request scenarios to evaluate whether the assistant could consistently identify relevant information and produce useful procurement-oriented outputs.
- 6Reviewed the results against expected procurement reasoning and refined the workflow to improve consistency, clarity, and usefulness of the recommendations.
- 7Established the solution as a foundation for future integration with broader procurement workflows, business data, approval processes, and performance measurement.
Key Technical & Architecture Decisions
Architecture, model selection, workflow and trade-offs.
The solution was designed around a separation between AI-assisted analysis and workflow orchestration. This allows the reasoning-oriented tasks, such as interpreting purchase request information and generating procurement-oriented recommendations, to be combined with structured workflow steps for consistent processing.
A key decision was to use structured inputs and outputs rather than relying solely on free-form AI responses. This makes the results easier to review, compare, and integrate with future procurement processes.
Another important decision was to keep human validation in the process. Procurement recommendations can have operational and commercial consequences, so the assistant is positioned as decision support rather than an autonomous authority.
The solution was built incrementally, starting with the core purchase request analysis flow before considering broader integrations. This reduced unnecessary complexity and created a foundation that can be extended as real usage data and requirements become available.
Challenges & How They Were Solved
Obstacles hit along the way and how they were overcome.
One of the main challenges was translating practical procurement reasoning into a structured AI-assisted process. Procurement decisions often depend on context rather than a single fixed rule, so the workflow needed to distinguish between information that could be processed consistently and decisions that still require human judgement.
Another challenge was avoiding overly confident AI recommendations when a purchase request contains incomplete information. The solution therefore emphasises identifying missing information and presenting structured recommendations rather than treating every AI output as a final decision.
A further challenge was defining meaningful measurement without overstating results. Because the current deployment is a personal implementation, financial savings and automated hours were not assumed. Instead, the solution establishes a basis for future comparison of manual versus AI-assisted processing time and output quality.
User Adoption & Change Management
How users responded — training, change management and feedback loops.
As a personal AI implementation, the initial focus was on validating the solution from a procurement practitioner's perspective rather than conducting a formal organisational rollout.
Trust was treated as an important design consideration. The assistant is positioned as a support tool, with human review remaining important for procurement decisions. This helps prevent users from treating AI-generated recommendations as automatically authoritative.
Future adoption would benefit from introducing representative users, collecting structured feedback, comparing AI-assisted outputs with existing manual processes, and progressively refining the workflow based on observed usage.
Business Outcomes & Impact
The measurable outcomes from the underlying AI Deployment — with the story behind them.
The solution is designed to improve procurement productivity by reducing repetitive manual review of purchase requests, improving consistency in requirement analysis, and providing structured AI-assisted procurement recommendations. The measurable impact will be evaluated by comparing manual and AI-assisted processing time and output quality
Business Outcome Categories
The story behind the numbers
The solution establishes a structured and repeatable approach to purchase request analysis, reducing the need to perform the same initial review activities manually for every request. By assisting with requirement extraction, completeness checks, and procurement recommendations, the approach is intended to improve consistency and provide faster decision support.
For procurement practitioners, the main value is the ability to standardise part of the initial analysis process while keeping human judgement involved in procurement decisions. This creates a clearer foundation for measuring processing time, output quality, and productivity as the solution is further developed and used with representative or real operational data.
The current deployment therefore focuses on establishing the capability and measurement framework rather than claiming unverified financial savings or automated hours.
Lessons Learned
What surprised the team, what worked well, and what would be done differently.
- Start with the business process before selecting the AI approach.
- Use domain expertise to define what AI should assist with and what should remain under human judgement.
- Structure AI inputs and outputs so results can be reviewed consistently.
- Treat incomplete information as a normal procurement scenario rather than forcing the AI to produce a definitive answer.
- Measure real operational impact before claiming productivity, cost, or financial improvements.
- Build incrementally and establish a reliable core workflow before expanding into broader automation.
- Keep human validation in the loop when AI recommendations can influence commercial or operational decisions.
Future Opportunities
Where this solution could go next.
The next stage would be to connect the assistant with a broader procurement workflow, including Purchase Requisitions, supplier information, historical purchasing data, approval processes, and procurement reporting.
Future improvements could include:
- Automated completeness and policy checks;
- Integration with procurement or ERP systems;
- Supplier and historical purchase analysis;
- Automated routing of requests based on defined criteria;
- Human-in-the-loop approval workflows;
- Procurement dashboards and performance monitoring;
- Evaluation datasets for measuring AI output quality over time;
- Agentic workflows for selected low-risk procurement activities.
As usage data becomes available, the solution could also establish stronger quantitative measurement of processing time, consistency, recommendation quality, and productivity improvements.
Professional Reflections
Guidance for another builder tackling a similar problem.
This project reinforced the importance of starting AI initiatives with the business process rather than the technology. Understanding how procurement professionals actually review and interpret purchase requests made it easier to identify where AI could provide meaningful assistance and where human judgement should remain essential.
One of the most important lessons was that successful AI implementation is not simply about generating technically impressive outputs. The solution must fit the user's workflow, produce understandable results, and provide enough structure for people to validate its recommendations.
For other AI Builders, my advice is to begin with a clearly defined business problem, use domain knowledge to shape the solution, introduce measurement early, and avoid claiming impact before it has been properly validated. A smaller, well-understood workflow can provide a stronger foundation for future automation than an overly ambitious solution with unclear outcomes.