Purchasing requests under control
How can you prototype an AI-assisted purchasing request workflow without delegating decisions to enter into commitments?
3500 € per session, excl. VAT
In short
Over two days, totalling 14 hours excluding breaks, design a prototype that transforms varied purchase requests into structured, checked files routed to the right approver. You define the business rules, test exceptions and measure errors before preparing a limited pilot, with no mandatory connection to the purchasing system.
Why this training?
Incomplete requests lead to follow-up requests, while incorrect routing delays budget approvals. Formalising the expected fields and escalation rules allows these discrepancies to be addressed before a commitment is made. Generative AI can help extract and rephrase information, but it can also invent a reference or interpret an amount. The solution therefore separates extraction, deterministic checks and human decision-making. Continuing case study: The fictitious organisation Sarnive Assemblage receives requests for parts and services by email. A fictitious corpus combines incomplete quotations, urgent requests, duplicates and ambiguous amounts; the prototype must prepare them for assessment without issuing an order. Application at work: Propose a pilot for one purchasing category, with approval from the purchasing manager and IT department: an authorised corpus, comparison with manual processing, error review and a documented decision on whether to proceed.
Objectives of this training
- Map the processing steps and responsibilities
- Define a data schema and purchasing checks
- Assemble a prototype with human approval
- Test exceptions and calculate quality indicators
- Draft a reversible pilot plan
Who is it for and what are the prerequisites?
Roles and tasks concerned
- Purchasing and procurement managers responsible for the request workflow
- Process leads and IT teams supporting a purchasing pilot
Prior skills
- Know the approval workflows and rules for purchasing commitments.
- Be able to structure an AI instruction and check its response.
- Work with tabular data and understand a conditional sequence.
What you will work on
Define the workflow boundaries
Objective: Define the scope for automation. Receipt, qualification, approval; commitments excluded.
Structure the file
Objective: Define the information required. Subject, quantity, amount, currency, budget, deadline and supporting documents.
Formalise the checks
Objective: Make routing verifiable. Approval thresholds, duplicates, mandatory fields; calculations outside the generation process.
Prototype the extraction
Objective: Produce structured output with source references. Instructions, expected format, abstention and references to supporting documents.
Assemble the workflow
Objective: Link extraction and assessment. File statuses, checks and human approval queue.
Handle exceptions securely
Objective: Prevent risky processing. Malicious instructions in documents, sensitive data, permissions and logging.
Prepare follow-up requests
Objective: Request only useful additional information. Contextualised drafts, approval before sending and history.
Run the tests
Objective: Evaluate on held-out cases. A dataset separate from tuning, expected results and critical errors.
Correct and rerun
Objective: Verify corrections without regression. Error analysis, version control and repeatability.
Define the pilot framework
Objective: Prepare a limited trial. Responsibilities, acceptance, supervision, stopping and returning to manual processing.
What you take away
- Workflow map and responsibility matrix
- Field dictionary and catalogue of checks
- Documented prototype with version-controlled instructions
- Test report and error table
- Pilot plan with a fallback procedure
Method and assessment
- Short input sessions followed by immediate practice using the case study that runs throughout.
- Individual work followed by work in pairs, with peer review of the outputs.
- Guided group discussion: what works, what needs correcting and what remains a human decision.
- Preparation: Prepare an anonymised description of the purchasing workflow and its approval thresholds. Check access to authorised tools; use only fictitious or anonymised cases.
Observable criteria: Each extracted field refers to a source or remains explicitly marked as missing. ; Approval scenarios comply with the defined matrix. ; No scenario automatically triggers a commitment. ; Correct extraction and routing rates state their denominators. ; The pilot specifies responsibilities, acceptance thresholds and stopping conditions..
Frequently asked questions
Does the purchasing system need to be connected?
No, exchanges can be simulated using structured files.
Does AI approve purchases?
No, approval and commitments remain a human responsibility.
Can we use our quotations?
Only in anonymised form, with prior authorisation, using tools approved by the organisation.
Will the prototype be made production-ready?
No. Its integration, security and operation will require additional approvals.
Last updated: