Map the operating system.
People, software, data, handoffs, exceptions, approvals, delays, and the business numbers management already watches.
01 / WORK BETWEEN SYSTEMS
The work should move.
AI workflow design and implementation
The work should move.
Not the spreadsheet.
I redesign the work trapped between Excel, inboxes, meetings, and business systems, then build the automation and AI layer that moves it forward, keeps exceptions human, and proves the result.
Calgary · worldwideWritten diagnosis firstFixed scopeEvidence at handover
The real problem
Most businesses do not have an AI problem. They have work trapped between people and systems.
A report copied from five spreadsheets. A project status rebuilt before every meeting. A quote that waits in an inbox. An invoice exception nobody owns. AI matters when it removes that friction without hiding decisions or creating another tool to manage.
The diagnostic engine
See the whole business before choosing the AI.
Substrate follows work from demand to delivery and cash, ranks the operating leaks worth fixing, checks what the company already owns, and produces an evidence-labelled 30 / 60 / 90 day roadmap.
Configure, integrate, build, or stop.
The system does not turn every problem into a custom AI project. It identifies owned capabilities and structural prerequisites first.
Start with one measurable move.
Every proposed intervention carries an evidence level, confidence, human owner, baseline requirement, and acceptance test.
Start with the job
Where the work changes.
Choose the function where people are spending judgment on clerical work. Each path begins with the current workflow, not a preferred model or vendor.
01
Business analysis
Replace spreadsheet consolidation and recurring report assembly with a traceable data-to-briefing workflow.
Sources → variances → cited draft → analyst decision 02Project management
Collect updates, extract decisions and dependencies, and prepare status without chasing every person manually.
Updates → risks → owner → approved status 03Finance & administration
Match documents, prepare reconciliations, and route exceptions while financial approval stays with a person.
Documents → match → exception → approval 04Sales & revenue
Research, enrich, prioritize, draft follow-up, and expose pipeline gaps before opportunities go cold.
Signal → context → next step → human send 05Customer operations
Triage requests, retrieve source-backed answers, draft responses, and escalate the cases that require judgment.
Request → evidence → draft → escalation 06Field operations
Turn job notes, photos, forms, schedules, and inventory changes into structured action without double entry.
Field capture → system update → exception 07Executive reporting
Build a source-linked operating brief around changes, unresolved decisions, risks, and accountable owners.
Systems → change → decision → owner 08AI & engineering
Bound agent spend, evaluate outputs, protect evidence, and stop unreliable AI work from silently shipping.
Run → cap → verify → approveOne concrete example
A business analyst should analyze the business, not rebuild the same workbook.
The goal is not to eliminate Excel. It is to stop using a person as the connector between every file, system, and recurring decision.
Seven files become one fragile report.
- Export data from separate systems
- Copy columns and repair formats
- Rebuild charts and commentary
- Search messages for missing context
- Discover an error after circulation
The analyst receives the work that needs judgment.
- Approved sources refresh into one model
- Rules flag missing or unusual values
- A source-linked briefing is drafted
- Exceptions arrive with context and owner
- The analyst validates conclusions before release
Who this is for
Lean companies with expensive coordination.
The strongest fit is usually an owner-led company or a lean team where capable people have become the unofficial bridge between customers, spreadsheets, meetings, and software.
Professional services
Consultancies, agencies, legal and accounting teams managing intake, delivery, approvals, reporting, and client follow-up.
Construction & field service
Contractors, engineering services, maintenance, and trades coordinating estimates, job records, scheduling, purchasing, and change.
Clinics & high-value services
Teams where every missed enquiry, incomplete intake, delayed follow-up, or manual handoff has a measurable cost.
SaaS & AI product teams
Founders and engineering teams that need reliability, evaluation, cost, evidence, and human approval around AI-enabled products.
Multi-location operations
Retail, franchise, and service networks that need controlled configuration, consistent reporting, and clear exceptions across locations.
Logistics & light manufacturing
Operations where forms, inventory, dispatch, supplier communication, and exception handling cross multiple systems.
Case studies & proof
Built work, labelled honestly.
Public products, delivered builds, and reference architectures are different kinds of evidence. This library says which is which instead of turning every prototype into a fictional client success story.
SYSTEM RUN · PUBLIC-SIGNAL DIAGNOSTIC
Substrate found that the right first AI project was not a new AI project.
A real diagnostic run mapped a home-services operator, ranked operating leaks, found unused capabilities in the existing stack, and identified the prerequisite for reliable instant quoting.
Inspect findings, evidence, and limits → PUBLIC PRODUCT · OPEN SOURCERuncap: control AI coding spend and protect merge evidence.
A local-first CLI and gateway with cost boundaries, run controls, and a Proof Gate for AI-generated pull requests. Repository, tests, and live proof are public.
Inspect the product → DELIVERED BUILD · DISCLOSED LIMITSA dual-mode financial AI agent under a three-day constraint.
A working React Native build with customer and internal modes, 27 tools, provider fallback, and privacy-aware audit design. The page separates implementation evidence from production claims.
Read the build record → REFERENCE ARCHITECTUREAtomic configuration across multi-location operations.
A Dataverse design for validating and applying related location changes as one traceable change set, with inheritance and rollback boundaries.
Review the architecture →How the work is delivered
Clarity before automation. Proof before scale.
You do not need to choose a model or arrive with a technical specification. You need one workflow worth improving and access to the people who understand it.
Map
Follow the work from trigger to outcome: systems, files, handoffs, delays, decisions, and exceptions.
Baseline
Measure the current cycle time, manual touches, error points, cost, and definition of a good result.
Build
Implement the smallest controlled layer using the tools and data the team already has.
Prove
Run real cases against acceptance tests, document limits, and hand over ownership and operating notes.
Start with one recurring workflow
Bring the spreadsheet, inbox, and handoff nobody wants to own.
I will return the likely automation boundary, what should remain human, and the first test worth running. No generic AI roadmap required.