Professional services Procurement Global

Four agents handle the sourcing. The decision stays human.

A leading global professional services and advisory firm, with $10B+ in annual revenue, replaced fragmented procurement workarounds with one AI-powered operating layer, built and delivered in 12 weeks.

Up to 55%Wider reach in supplier discovery
12 weeksA use case, made live.
8Sourcing steps, now agent-assisted
01 · Client

One of the world’s largest professional services firms.

A leading global professional services and advisory firm, one of the largest in the world, where procurement isn’t just an internal function. It is a capability delivered across thousands of client engagements.

Annual revenue>$10B
Employees>50,000+
Founded1987
Countries served140+
SectorProfessional services & advisory
OwnershipPartnership, member-firm network
02 · The problem

Procurement, held together by workarounds.

Procurement wasn’t struggling because the problems were new. It was fragmented, spread across tools, teams and inboxes never designed to work together.

01

Manual, fragmented workflows

Supplier onboarding was tracked over email with no shared status. Spend data sat in dashboards with no direct path to action.

02

Inconsistent, unauditable risk assessment

Risk assessments were rebuilt from scratch for every engagement, applying different standards with no audit trail.

03

Knowledge that couldn’t scale

Process knowledge lived in individual analysts. Not transferable, not reusable across the organization.

04

Every engagement started from zero

No shared infrastructure across engagements, so nothing built for one client carried over to the next.

03 · Native limits

Four requirements a real fix couldn’t skip.

Adding another point solution wasn’t going to fix a systems problem. The brief called for an operating layer, not a dashboard.

01

Connect every stage, not just one

An operating layer spanning the full procurement lifecycle, not a single-point tool.

02

Coordinate automatically

Connected agents that hand off work between steps on their own, not through manual handoffs.

03

Stay configurable, not rigid

Flows that adapt per client, category and region, instead of one fixed workflow.

04

Be reusable, not rebuilt

Infrastructure that serves every new client engagement without fresh engineering work each time.

04 · What Lyzr built

Four pillars, one platform.

The engagement was structured in phases. Phase 1 delivered the foundation in 12 weeks, with the strategic sourcing layer built in parallel.

Architecture pillars
Supplier management & onboarding
Onboarding workflows, KYC, risk and supplier master data.
Strategic sourcing & supplier discovery
Spend analysis, supplier discovery, RFQs and bid analysis.
Contract management
Contract lifecycle, clause tracking and renewals.
Procurement operations & AP
Requisitions, PO tracking, invoice matching and AP automation.
Agent layer
Intake / journey agent
Interprets requests and routes them to the correct flow.
Questionnaire agent
Selects question sets based on category, region and risk profile.
Risk analysis agent
Scores responses, computes risk bands and updates the supplier profile.
Performance agent
Configures monitoring cadences, KPI definitions and alert thresholds per supplier.
05 · Architecture

One architecture, configured per engagement, not rebuilt.

The same infrastructure had to support many client engagements, each configured differently, without fresh engineering work every time.

Deployed stack architecture. A React Vite frontend, holding the chat interface, UI context generation and frontend client tool resolution, talks to a FastAPI server backend over REST APIs, with the chat interface on a Websocket or HTTP connection to the Lyzr Studio handle. Inside the backend, workflows, around 10 of them, pass work to and from the Lyzr Studio handle, which calls agent chat in Lyzr Studio and hands tool resolution to the backend client tools. RAG answers back into the workflows, as does an Azure Web Search tool. The backend reads and writes a MongoDB server, takes files from Azure Blob Storage and scheduled runs from cloud functions, and sends email through Azure Communication Service. Tool results return to the frontend, into the UI actor and the chat interface.
06 · Controls & governance

Every decision, logged and explainable.

Nothing in the platform operates as a black box.

Traceability
Every agent action is logged: the trigger, the data used and the decision made.
Standardized scoring
Risk assessment applies the same 5-area framework across every engagement, not individual judgment.
Human decision gate
Agents surface comparisons and recommendations. Sourcing decisions are still made by people.
Configuration-driven
Templates, risk models and routing rules live in data, so changes don’t require new engineering work.
Security by design
Network isolation, least-privilege access and a web application firewall are built into every deployment.
07 · Re-imagined workflow

From an email chain to one conversational request.

A sourcing event, end to end. The request is the same. What happens after it is not.

Drag the handle to move from Before Lyzr to After Lyzr
Before Lyzr
One fixed chain, run by hand, for every request alike.
Buyer emails the requestThen waits, hoping nothing gets missed
Chase approvalsBy email, with no shared status
Rebuild the risk checklistFrom scratch, by individual judgment
Search suppliers manuallySpend data sits in dashboards, with no path to action
RFQ & bid comparisonBuilt and compared in spreadsheets
A decision, unrecordedMade ad hoc, with no audit trail
Buried in an inboxNo one else ever sees it. The next buyer repeats the chain
Starts over, nothing carries forward
Manual stepDecisionDead end
After Lyzr
One request in. 4 agents route it. A person still decides.
Buyer asks, in plain EnglishGets the logged decision back, then repeats
Intake & routing agentReads the request, splits it to the right specialists
Risk & compliance agentRight questions asked automatically. Same 5 risk areas, every time.
A decision, loggedThe risk check and the human call converge here, trail intact
Sourcing & RFQ agentSpend data feeds the event. RFQ ready in minutes, bids compared automatically.
Human decisionOnly the sourcing agent’s recommendation reaches here. A person makes the call.
Remembered for next timeCaptures the decision, feeds it into the next request
Feeds the next request
Risk & complianceSourcingHuman decisionDecision, logged
08 · Results

What changed, by the numbers.

Before and after, across the procurement lifecycle.

MeasureBefore LyzrAfter Lyzr
Supplier onboardingRequested by email, no visibility into statusRequested in one message, status tracked live
Risk assessmentRebuilt each time, based on individual judgmentStandardized, scored across 5 risk areas
Spend insightSat in dashboards with no path to actionFeeds directly into a sourcing event
RFQ creationBuilt from scratch for every requestAgent-recommended, ready in minutes
Bid comparisonCompared manually in spreadsheetsCompared automatically, inside the platform
Process knowledgeHeld by individual analysts onlyReusable across every future engagement
ScalabilityRebuilt from zero for each new clientConfigured once, deployed for every client
Up to 2xFaster supplier onboarding
Up to 70%Higher supplier service levels
30%Greater supply chain efficiency
09 · What’s next

From foundation to full operating system.

Phase 1 laid the foundation. The roadmap extends the same reusable model across the rest of the procurement lifecycle.

12 weeks from use case to Phase 1 in production

Got a use case in mind?

Platform, engineers and governance all in. We’ll map your workflow against the same 4 tests this one had to pass.