Channels + triggers + schedules
Teams paths and file intake receive work and source files.
Case Studies / Manufacturing
From tribal knowledge to an AI operating system, in 90 days.
How a private manufacturing group went from idea to a working AI foundation: four AI employees in Microsoft Teams, a searchable company memory, and a plan to measure operating value over the next 90 days.
4
named, role-based AI employees responding in Teams
PRODUCTION EVIDENCE
3
core systems verified live in production
PRODUCTION EVIDENCE
90
days from kickoff to working foundation
PRODUCTION EVIDENCE
Day 180
target for measured operating results
MEASUREMENT IN PROGRESS
EXECUTIVE SUMMARY
The case in four decisions
- Waiting work
- Estimating, quoting, reporting, and file search depended on manual effort and knowledge held by a few people.
- What changed
- A 90-day foundation introduced four named AI employees in Teams, shared company memory, file intake, tool connectors, and an observable operating layer.
- Human decision
- People retain final approval for sensitive or external actions and decide which pilot patterns move into production.
- Evidence state
- The production foundation and current deployment status are documented. The May proof panels are stylized recreations. Operating-value scenarios remain illustrative.
THE CHALLENGE
The business ran on knowledge locked in people's heads
A private manufacturing group with one goal: reduce the cost of operating the business. Estimating, quoting, reporting and file search all depended on manual effort and tribal knowledge, and on the owner personally. That meant high cost to serve, high key-person risk, and a harder story to tell in any future diligence.
The bet: AI becomes valuable when it is connected to company memory, approved tools, and human approval gates. A standalone chat window was insufficient, so we built an operating system.
THE APPROACH
Learn the work. Prove the value. Then build.
Infrastructure was only built for patterns that proved useful first. The team deferred speculative technology spend.
APRIL · DISCOVERY
Learn the work
Mapped the workflows, knowledge sources and risk boundaries where AI could remove the most cost.
“What work should AI support?”
MAY · PROOF OF CONCEPT
Prove the value
Tested AI employees in Teams and email: estimating support, file intake, and report generation.
“Which patterns earn hardening?”
JUNE–JULY · BUILD
Build the foundation
The production system: company memory, knowledge map, file intake, employee profiles, tool connectors, and observability.
“What makes AI repeatable?”
WHAT WE BUILT
An operating system for four AI employees
Above the line: four AI employees available in Teams. Below it: the six-layer system that supplies context, tools, and review controls.
EXPLORE THE FOUR EMPLOYEES AND SIX-LAYER SYSTEM
THE SURFACE · WHAT PEOPLE SEE
A
Abby
CHIEF OF STAFF · MANAGER
Assigns work, tracks evidence, reports up.
V
Val
ESTIMATOR
Turns RFQs and job history into estimate drafts and clarifying questions.
J
Jacob
SCOPE REVIEWER
Red-teams every quote: scope gaps, risks and missing details before approval.
L
Lucy
COMMUNICATIONS
Turns technical progress into client-ready reports, emails and updates.
THE ROOT SYSTEM · WHAT MAKES THEM WORK
01
Company memory
Meaning-based search finds concepts across filenames and folders.
02
Knowledge map
Connects people, projects, systems and decisions.
03
File intake
SharePoint and file drives flow into AI-usable context.
04
Employee profiles
Role-specific instructions, tools, boundaries and handoffs.
05
Tool connectors
Teams, email, Odoo, reporting. Approved access only.
06
Observability
Every AI action traced, scored and reviewable.
Human oversight throughout: specialists execute and escalate uncertainty; final decisions stay with people. Anything sensitive or external requires owner approval.
PILOT PATTERNS · ILLUSTRATIVE
Stylized recreations from the May pilot
ABBY · MICROSOFT TEAMS
Abby, what's blocking this week?
Two items: the Westfield quote needs a material spec, and Val is waiting on the RFQ drawings. I've flagged both owners.
Represents an AI employee responding through a Teams-style work channel.
VAL · RFQ REVIEW
✓Material spec confirmed
✓Quantities match drawings
✓Similar prior job found
!Finish spec missing: ask customer
3 OF 4 ESTIMATING INPUTS READY
Represents an RFQ input check and a drafted clarification for human review.
LUCY · WEEKLY REPORT
Operations update · Week 12DRAFT
Drafted from company files
Represents a report draft assembled from company files for human review.
Stylized recreations of proof-of-concept sessions from the May pilot phase.
CURRENT STATE
Current delivery status
Three systems verified live in production; two in structured review. The table separates working components from items still being hardened.
AI communication channels
All four AI employees responding on their Teams paths
Meaning-based company memory
Search engine verified in production, v1.18
Conversation continuity
AI employees remember context across sessions
Business system connectors (Odoo/ERP)
Built and connected; completing stability review before rollout
Self-improvement scaffolding
Foundation in place; becomes the improvement engine next phase
HOW IT WAS BUILT · MIXED DELIVERY MATURITY
The working components in shared language
Each mapping points to an architecture responsibility delivered in this engagement. The current-state section distinguishes live, hardening, and pilot components.
VIEW THE DELIVERED ARCHITECTURE MAPPING
Context + session and durable memory
Company memory and the knowledge map assemble relevant context.
Skills + orchestration
Employee profiles define roles, instructions, boundaries, and handoffs.
Tools + connectors + delegation
Tool connectors provide approved access to business systems.
Rules + approvals + escalation
Human approval gates hold consequential actions for review.
State + trace + checkpoints
Observability makes AI activity traceable and reviewable.
PLANNING MODEL · ILLUSTRATIVE
Explore the operating-value assumptions
This model tests two possible drivers: EBITDA improvement from faster estimating, search, and reporting; and multiple expansion from lower key-person risk and cleaner diligence. It is a planning model, not a realized client result. The baseline uses a 3.75× multiple for small private manufacturing.
1.12×
Conservative
+5% EBITDA · 4.00× multiple
1.33×
Base case
+15% EBITDA · 4.35× multiple
1.65×
Illustrative high case
+30% EBITDA · 4.75× multiple
The day-180 target is to replace assumptions with measured hours saved, quote-cycle changes, and rework avoided.
SERVICES USED
This engagement, as catalog services
MASLOW DELIVERY SUMMARY
The first 90 days built the production system. The next 90 measure how it changes operating work.
DELIVERY STATUS · OPERATING MEASUREMENT IN PROGRESS
What could a 90-day foundation include for your business?
Same playbook: learn the work, prove the value, build only what earns it.