CASE STUDY · AGENTIC AI
Agentic AI development & orchestration: AI agents that move from conversation to action.
Illustrative scenario · Sector: Enterprise procurement and operations · Solution partner: Brandsmashers Tech
- 8,000Procurement requests a month in the modeled enterprise
- 2,880Modeled routine requests a month an AI workflow could screen
01 · PROJECT OVERVIEW
From AI that responds to AI that executes.
Many enterprises have already introduced AI assistants to help employees search information, generate content and answer questions. The next challenge is moving from AI that responds to AI that executes controlled business workflows.
Brandsmashers’ approach is to design an agentic AI workflow around a measurable business process, rather than simply deploying a chatbot.
- Identifying the right, measurable automation opportunity
- Designing and orchestrating the AI agent and its tools
- Integrating enterprise APIs, policy sources and approval workflows
- Testing, auditing and scaling the workflow in production
- 1Interpret and classify every incoming procurement request.
- 2Retrieve the relevant policies and validate the information provided.
- 3Decide the next step, and leave exceptions and approvals to people.
02 · THE CHALLENGE
The problem is repetition, not headcount.
Consider an enterprise receiving approximately 8,000 procurement requests every month. Around 4,800 of them may be routine: standard purchases, repeat suppliers, predefined categories or requests that follow established approval policies.
- PROBLEM 01Every request follows the same manual path
Reviewing, categorising, validating and routing are repeated by hand for thousands of requests, whether they are routine or not.
- PROBLEM 02Skilled people on repetitive steps
Valuable procurement professionals spend time on work that could be handled through intelligent workflow automation.
- PROBLEM 03Assistants that answer but don’t act
A chatbot can explain the policy, but someone still has to check the request against it and move it forward.
- PROBLEM 04Control can’t be traded for speed
Any automation has to respect approval policies, spending limits and an audit trail from day one.
Build AI agents around measurable workflows, not demos.
03 · THE APPROACH
Design the workflow first, then give the agent its tools.
The AI agent interprets an incoming request, classifies it, retrieves relevant policies, validates the available information and determines the appropriate next step.
- AAgent orchestration
- AI agent orchestration around the procurement process
- Large language models to interpret free-text requests
- Retrieval-augmented generation over procurement policies
IMPACTRequests are understood and classified the same way every time.
- BEnterprise integration
- Enterprise APIs for suppliers, categories and budgets
- Workflow automation for routing and status updates
- Business-rule engines for policy checks
IMPACTThe agent acts inside existing systems instead of beside them.
- CControls built in
- Role-based access control for every action
- Approval workflows for high-value purchases and exceptions
- Audit logging of each step the agent takes
IMPACTSpeed without giving up accountability.
- DHuman-in-the-loop
- Exceptions and policy violations routed to people
- Clear summaries so approvers decide faster
- Measured against the volume and accuracy targets set up front
IMPACTPeople keep the judgement calls; the agent handles the structure.
- 1→Request received
- 2→Classified
- 3→Policy retrieved
- 4→Validated
- 5→Routed or approved
- 6Logged
04 · RESULTS
Illustrative business model.
The objective is not to “replace procurement”. It is to let procurement professionals spend more time on the work only they can do, while AI handles structured, repetitive workflow steps.
| METRIC | TODAY | MODELED |
|---|---|---|
| Monthly procurement requests | 8,000, all handled manually | 8,000, screened by the AI workflow first |
| Routine requests | 4,800 reviewed by hand | 2,880 a month potentially AI-eligible (60% of routine) |
| Human involvement | Every request | Exceptions and approvals |
Scenario figures are illustrative and modeled for demonstration, not a historical Brandsmashers client result. Gartner’s research provides the external market context for the growth of task-specific enterprise AI agents.
- 8,000Procurement requests a monthThe modeled enterprise.
- 4,800Routine requests a monthStandard purchases, repeat suppliers, predefined categories.
- 2,880Potentially AI-eligible routine volumeData-pack assumption: 60% of routine requests could be screened by an AI workflow.
- ExceptionsHuman involvementPlus all approvals above policy limits.
- Strategic sourcing
More time for sourcing strategy and complex purchasing decisions.
- Supplier relationships
More time for negotiations and vendor relationships.
- Risk and exceptions
Attention moves to risk management and the cases that need judgement.
- Consistent routine work
Structured, repetitive steps handled the same way every time, with an audit trail.
05 · DELIVERABLES
How Brandsmashers would build it.
- Opportunity assessmentFinding the procurement steps that are measurable, repetitive and safe to automate.
- Agent developmentThe AI agent, its tools and the orchestration logic.
- Enterprise integrationsAPIs, workflow engines, policy sources and approval workflows.
- Testing and auditEvaluation before release, and audit logging in production.
- Scale-upExtending the workflow to more categories as accuracy is proven.
- AI
- AI agentsLLMsRAGVector databases
- Engineering
- PythonFastAPIAPIs
- Workflow
- Workflow enginesBusiness-rule enginesApproval workflows
- Platform
- CloudEnterprise integrationsRole-based accessAudit logging
TAKEAWAYS
Why Brandsmashers.
- 01Start from a measurable business process, not a demo.
- 02Give the agent tools, not unrestricted authority.
- 03Keep exceptions, violations and high-value approvals with people.
- 04Log every step so the workflow can be audited and improved.
- Procurement
- Finance operations
- HR operations
- Shared services
- Supply chain
YOUR TURN
Ready to move AI from conversation to action?
Brandsmashers supports organisations across the complete AI engineering lifecycle, from identifying the right automation opportunity to building, integrating, testing and scaling AI-powered workflows.