We supported a UK-based SME manufacturer/distributor that previously ran orders through paper notes and informal coordination. Peak season demand exposed a predictable bottleneck: limited human capacity and no operational visibility. We delivered a practical end-to-end project management system (PMS) to track orders from intake through dispatch, introduced WIP ageing and timing visibility, and trained the team to operate from a single source of truth. On top of that foundation, we are building a custom AI operations assistant that connects to the PMS via an MCP-based connector to generate real-time insights, answer operational questions in natural language, and surface risks using live data.
The Starting Point
Operational Reality
The business had no formal job or order management system. Work was coordinated through paper notes, verbal handoffs, and memory. There was no structured ownership of tasks, no end-to-end visibility into work in progress, and no data on job timing or throughput.
Peak season repeatedly exceeded the team's coordination capacity. Late dispatches, rework, and reactive customer chasing became the norm rather than the exception.
Business Risks
- No objective basis for prioritising work. Decisions relied on whoever shouted loudest.
- No reliable WIP or throughput data to support capacity planning.
- Founder and operations time consumed by chasing status updates rather than managing flow.
Objectives
Five outcomes were agreed at the outset:
What We Set Out to Achieve
Scope and Boundaries
In Scope
Discovery and workflow mapping (as-is to to-be). PMS data model and operating workflow design. PMS build, configuration, training, and handover. AI layer build: MCP connector to PMS, insight workflows, and governance guardrails.
Out of Scope
Website or e-commerce build work. Broad marketing execution. Major bespoke software builds beyond a pragmatic Phase 1 system. Long-term support beyond delivery.
What We Delivered
Phase 1: Production PMS Foundation (Live)
We implemented an end-to-end project management system on a commercial work management platform. The system standardised the operating workflow with defined stages, ownership assignments, and required fields at each step.
WIP ageing and timing visibility was introduced to show how long work sits in each stage. This gave the team objective data on bottlenecks for the first time.
Operational views were built to support daily use:
- Pipeline and order tracker for end-to-end job visibility
- WIP and "at-risk" queue highlighting jobs exceeding stage thresholds
- Overdue jobs and dispatch-ready list for daily prioritisation
Enablement included user testing and iteration, training for both frontline users and the admin/owner, and a handover pack covering SOPs, an admin checklist, and governance rules.
Phase 2: Custom AI Operations Assistant (In Build)
A custom AI layer is being built to sit on top of the PMS and leverage the operational data now being captured. The AI connects to the PMS via an MCP-based connector, a structured integration approach that allows it to securely retrieve and act on authorised operational data.
The goal: deliver faster, higher-quality decision support without requiring staff to manually compile updates or reports.
The AI Layer: Ops Co-Pilot
The centrepiece of Phase 2 is a custom AI operations assistant that uses an MCP connector to access live PMS data. It is designed to produce real-time operational summaries, risk flags, exception alerts, and decision-support insights.
Why MCP Matters
MCP creates a clean, governed bridge between the AI and the PMS. The PMS remains the single source of truth while the AI queries it safely through a structured interface. This pattern is scalable: new tools and data sources can be connected later without rebuilding the integration layer.
AI Capabilities
Operational Q&A
- What is overdue, and why?
- Which jobs are stuck in design approval?
- What must ship this week?
- Who owns this job and what is the next step?
Daily/Weekly Reporting
- Auto-generated WIP summary
- Stage ageing and bottleneck hotspots
- Top risks with recommended interventions
- Throughput snapshot by period
Exception Management
- Flags jobs exceeding ageing thresholds
- Highlights missing critical fields (approvals, sizing, artwork)
- Surfaces ownership gaps and stalled handoffs
- Alerts on SLA risk before breach
Capacity Insight
- Throughput trendlines over time
- Stage dwell time analysis
- Peak season demand patterns
- Enabled by the operational dataset now being captured
AI Governance and Guardrails
The AI layer is designed with clear boundaries to avoid overreach:
Role-based access. The AI can only see what a permitted user can see.
Audit logging. AI queries and actions are logged where implemented.
Human-in-the-loop. Any outbound messaging or record changes require human approval.
Clear permissions. Read-only by default. Write access granted only where governance controls are proven.
Outcomes
Confirmed Qualitative Outcomes
Paper to Digital
Complete transition from informal, paper-based coordination to a structured digital operations system.
Real-Time WIP Visibility
The team can now see every job's status, stage, owner, and ageing in real time. No more chasing updates.
AI-Ready Dataset
The PMS captures structured timestamps, ownership, and stage data that powers the AI co-pilot.
Reduced Admin Overhead
Time previously spent compiling status updates manually is being redirected to managing flow and exceptions.
Quantitative Measurement
Baseline measurement is in progress. The PMS now captures the timestamps required to quantify throughput (orders shipped per week), cycle time (intake to dispatch), on-time delivery rate, and admin hours saved. These metrics will be reported once sufficient data has been collected.
Engagement Summary
Delivery Model
Fixed-fee engagement covering discovery, PMS build, training, handover, and AI layer design.
Timeline
PMS foundation delivered in approximately 4 to 6 weeks. AI layer is a follow-on build phase.
What Comes Next
With the PMS foundation live and operational data flowing, the next phase focuses on three priorities:
- Deploying the MCP-connected AI co-pilot into daily operations.
- Using real WIP and cycle-time data to target the highest-leverage bottlenecks.
- Adding progressively more automation only where governance and audit controls are proven.
Interested in a Similar Approach?
Every business operates differently. Whether you need operational visibility, AI-ready data infrastructure, or a custom AI layer on top of your existing systems, we can help you build a practical path forward.
No hype. No generic playbooks. Just strategies that work for your specific context.
Book a complimentary strategy call →