01 · FINANCE
Finance operations
ERP · banking · expense · reporting
- Close readiness tracking
- Reconciliation across systems
- Collections and cash forecasting
- Exception triage and approvals
- Spend and vendor controls
AI ADVISORY · FOR OWNER-RUN CANADIAN COMPANIES OF 50–500 PEOPLE
AI that reaches production.
We turn finance, operations and revenue into departments run by AI agents: the re-keying, chasing and checking your team does by hand today. Built inside the ERP and CRM you already run. First agent live in 90 days, every one measured in dollars.
30 minutes · no deck · no cost. Bring the work that piles up.
Built inside your systems · Measured against a baseline · No platform to sell · How we build is public →
How we decide where AI goes
Every step gets sorted before anything gets built.
Below is a sales team's quoting process, mapped step by step as it really runs. Each step is deleted, run as plain code, handed to an AI model, or kept with a person. Watch the twelve steps get sorted.
Illustrative example · composite data
- 02Model
Read the email and attachment
- 03Model
Match to part numbers
- 04Code
Key lines into the ERP
- 05Code
Look up customer price
- 06Code
Check stock and lead time
- 07Code
Check margin against floor
- 09Person
Price a non-standard item
- 10Person
Approve below-floor discount
- 11Code
Build and send the quote
- 12Code
Log it in the CRM
Example quoting desk with twelve steps, the same desk as the method video. Sorted: two are deleted (a tracking sheet nobody reads, a manager's signature on quotes already inside policy), six run as code, two use a model (reading the email and attachment, matching to part numbers), and two stay with a named person (pricing a non-standard item, approving a below-floor discount).
WHAT WE HEAR ON EVERY FIRST CALL
What's actually happening with AI in most companies.
- The start
- Everyone has a licence. A few people use it hard, most never open it, and nothing runs differently on Monday.
- The process
- The SOP says six steps. It takes twelve, and one person knows them all. Often that person is you.
- The data
- Re-keyed from PDFs and inboxes into the ERP, by hand, every week.
- The vendors
- Everyone has a platform to sell. Nobody owns the outcome.
Companies your size don't have an AI problem. They have a production problem.
The tools aren't the edge
Every company can now buy the same AI. The edge is where it goes to work.
Nothing runs differently on Monday
The seats got bought and the demos were easy. The work itself didn't change.
We start from the number
The number you want to move, then the workflow behind it, and we stay until it moves. Background agents that take work off the plate, not copilots that make people ten percent faster.
95%
of enterprise AI pilots showed no measurable effect on profit and loss.
Source: MIT NANDA, The GenAI Divide: State of AI in Business 2025, August 2025. 300 deployments reviewed.
6%
of companies get 5% or more of their earnings from AI. 88% say they use it.
Source: McKinsey, The State of AI, November 2025. 1,993 respondents.
39%
of Canadian small business owners say the business is too reliant on them to run day to day.
Source: CFIB, Succession Tsunami, January 2023.
THE WORK
Three things we actually hand you.
Not slides about AI. Map the work, ship the agents, run the program: working artefacts that run in your business. Each demo below uses composite data from the kind of engagement we run.
01
AI Opportunity Map
We interview your team, watch the work, and map every repeatable workflow as it actually runs — loops and exceptions included, not the SOP. Each one is scored on annual value and feasibility — does it run often enough to matter, are the decisions repeatable, is the context scattered across systems, is the pain measurable — and tagged revenue, cost or risk. You get a ranked shortlist with dollar estimates — and a scoped first sprint. At the composite distributor below, quote generation came out top: the messy, judgement-heavy process no vendor packages.
Part of: AI Opportunity Assessment →AI Opportunity MapIllustrative example · composite data
Ranked shortlist
- Quote generationRevenue$51K
- AP invoice matchingCost$32K
- Customer service triageCost$24K
- Supplier contract reviewRisk$18K
- Demand forecast refreshRevenue$15K
02
An agent in production
One workflow at a time, until the function runs on them, each one built properly: inside the systems you already run, with rules where rules work, a model only where judgment is needed, and a human approving the exceptions. Every action logged. The demo cycles through three: supplier invoices (straight-through from 22% to 81%), quotes built from inbound RFQs (sent same day, 12% to 71%) and collections on overdue accounts (DSO from 54 to 38 days). Invoices are the familiar case. Quoting and collections are where the judgment sits and where the money is, and nothing off the shelf does them inside your ERP and CRM.
Part of: 90-Day Production Sprint →AP invoice agentIllustrative example · composite data
Invoices this month 321Straight-through 81% 22%Cost per invoice $7 $27
Inboxrule
Classifymodel
Extractmodel
Match to POrule
Post to ERPrule
03
AI program scorecard
Every week your leadership team sees what AI returned, what it cost, how cycle times moved, and the one decision that needs making. Spend is tracked against budget, so nothing drifts.
Part of: Fractional AI Lead →AI Program — Week 38Illustrative example · composite data
AI Program — Week 38
Hours returned (month)
78
Cost per invoice
$7
Days to close the books
6
Model spend vs budget
$535 / $625
Extend the quoting agent to the West region. Expected: +$11K annualized. Risk: low. Needs: sales VP sign-off.
Book a call
30 minutes · no deck · no cost. Bring the work that piles up.
WHY IT MATTERS NOW
Your next competitor runs on agents.
Not because they bought better software. Because the work between their systems stopped needing people to carry it. That is what AI-native means for a company your size.
- .01
Fewer hands between systems
The re-keying, chasing and approvals that sit between your inbox, your ERP and your CRM run in the background, with every action logged.
- .02
Work that doesn't wait
A quote or an invoice moves when it arrives, not when someone gets to it. Evenings and month-end included.
- .03
Growth without a coordination layer
Volume goes up without another coordinator, another spreadsheet, or another person who is the only one who knows how it really runs.
Book a call
30 minutes · no deck · no cost. Bring the work that piles up.
WHERE IT APPLIES
Across the business, not one task.
Agents don't stop at one department. The method is the same in every function: map the work with the people who do it, sort every step, and build inside the systems that function already runs. Each workflow that goes live makes the next one cheaper to build.
02 · REVENUE
Revenue operations
CRM · quoting · email · contracts
- Quotes built from inbound RFQs
- Deal desk and approval routing
- CRM enrichment and hygiene
- Pipeline and forecast review
- Commission validation
03 · OPERATIONS
Operations and supply
ERP · warehouse · planning · quality
- Order exception routing
- Demand and inventory signals
- Supplier performance tracking
- Quality defect patterns
- Procurement follow-through
04 · PEOPLE
HR and people operations
HRIS · payroll · scheduling · email
- Onboarding paperwork and system access
- Time-off and shift changes
- Policy and benefits questions answered
- Certification and training tracking
- Payroll change checks
05 · SERVICE
Customer service
help desk · email · CRM · order system
- Inbound email triage and routing
- Order status and delivery questions
- Returns and credit requests
- Warranty claims checked against the record
- Escalations with the full history attached
06 · FIELD
Field service and dispatch
dispatch · scheduling · mobile · invoicing
- Technician scheduling and dispatch
- Work orders from calls and emails
- Parts checked before the truck rolls
- Job notes turned into invoices
- Service contract renewals
These are examples of what a map usually turns up, not a menu. Anywhere the work repeats and the judgment sits with a few people, the same method applies.
HOW IT WORKS
How it works.
01Call
30 minutes. Is there a real AI opportunity here?
02Assess
We map how the work actually runs, sort every step, and baseline the numbers. You get the shortlist.
03Ship
90-day sprint. One workflow, live inside your systems, measured against the baseline.
04Scale
We run and maintain it, or hand it over with a runbook. Your call.
WATCH · TWO MINUTES
One quoting desk, sorted step by step.
INSIDE THE SPRINT: AUTONOMY IS EARNED
- A
Baseline
The numbers, measured before we build
- B
Shadow mode
Runs beside your team and acts on nothing
- C
Approved
A person signs off every exception
- D
Live
Every action logged and reviewable
Then the loop runs again. Each workflow we fix shows the next bottleneck, upstream or downstream.
BUILT ON WHAT YOU ALREADY RUN
Works with the systems you already have.
The agents work inside your ERP, CRM, accounting and HR systems, with the permissions you set and every update logged. No migration, and nothing new for your team to log into.
- QuickBooks
- Sage 50
- Sage Intacct
- NetSuite
- Microsoft Dynamics 365
- Business Central
- Acumatica
- Epicor
- SAP Business One
- Xero
- Salesforce
- HubSpot
- Zoho
- Microsoft 365
- Google Workspace
- Shopify
- ADP
- Dayforce
- BambooHR
- Zendesk
- ServiceTitan
- Jobber
Not on the list? If it has an API, an export or an inbox, the agent can work with it.
Book a call
30 minutes · no deck · no cost. Bring the work that piles up.
ENGAGEMENTS
Start small. Scale what works.
011–4 weeks
AI Opportunity Assessment
The fastest way to know where AI pays back in your business — and whether it does.
- One workflow in about a week, or the full map of every workflow in two to four weeks. Quoted after a first call, and credited in full against a sprint if you go ahead
- Interviews, observation and system access
- The process as it actually runs — loops and exceptions, and what each costs
- Every step sorted: delete, code, agent, or human
- Value × feasibility scoring, ranked with $ estimates
- Baseline KPIs: cycle time, cost per unit, straight-through
- A scoped first sprint with a fixed fee, or a straight answer not to build
02Fixed fee · about 90 days
90-Day Production Sprint
One workflow, live in production, measured against a baseline we take before we build.
- Baseline measured before we build, re-measured at 90 days
- Built inside the tools you already run — approvals land in Slack or email
- Evaluation suite on your real data before anything goes live
- Shadow mode first, then human-approved exceptions, then live
- Every agent action logged and reviewable
- Handover and runbook
03Monthly · after a sprint
Run & Maintain · Fractional AI Lead
Keeps what we built running after day 90, and adds senior AI leadership when you want it.
- Tied to the saving we measured, re-measured each quarter in your numbers
- Run and maintain, per agent: monitoring, re-testing when a vendor changes a format, model swaps, exception-rate watch
- Exceptions go to their owner the moment they happen, with the reason. Nothing waits for a report
- Your approvals and corrections feed the eval, so accuracy climbs after day 90 instead of drifting
- Weekly AI program scorecard
- Model benchmarking: the cheapest model that passes the eval
- Vendor decisions and AI spend governance
- Leadership days, board and investor updates, when you want them
WHO WE WORK WITH
Who we work with.
Owners who run the company
Fifty to five hundred people, and you sign across every department. You don't need a platform or a pilot. You need someone who has run a company to map where AI should go to work, build it inside your systems, and prove each piece moved a number.
Owners who just bought one
A search fund, a holdco, or a lower-mid-market fund with no operating bench. Half the process you bought lives in the seller's head. We map it before it walks out the door and build the first agent in your first 100 days.
The post-close plan →Owners preparing to sell
Two to four years out. A business that runs without you is worth more. We write down the real process and automate the parts a buyer would otherwise discount. Complexity left in place is priced against you, not for you.
Family businesses handing over
The next generation is taking over, and how the business really runs lives with the people who built it. We capture it while they are still in the building, and automate the parts a successor shouldn't have to learn by heart.
Professional services firms
Accounting, law and consulting firms putting AI to work on billable processes, and measuring it. The firm that owns the output, with AI underneath, is worth more than the one renting a tool with a nicer prompt.
Low-margin operators
Manufacturing, distribution, logistics, field service. Say labour is a quarter of revenue and a quarter of that is coordination. At a 3% margin, taking a tenth off coordination lifts earnings by a fifth, and nobody on the floor has to learn a tool.
WHY NOT A CONSULTANCY, OR A LAB
Why us?
Two kinds of firm sell this work. Each has half of it.
Consultancies
Know the process. Guess at the model.
- Can map a fifteen-step process. Cannot say which steps to trust to a model.
- Cannot tell you what an agent costs at 40,000 runs a month.
- No evaluation suite, so the design is a guess until it is live.
- Six months in: a beautiful deck, nothing in production.
AI labs and their deployment arms
Know the model. New to the process.
- Build the agent onto the process as it stands, broken parts included.
- Have never sat between finance and operations over who owns a step.
- Run it on their model only. Repriced, rate-limited or retired is your problem.
- A model swap means redoing the evals and re-tuning every call. They will not.
Agent Forge Works
Both halves, in one room.
- An operator who has run finance, operations, sales and HR, and knows where work gets stuck.
- Engineers who ship inside the ERP and CRM you already run: Dynamics, Salesforce and what sits around them.
- A model-agnostic harness with the evaluation suite and governance built in. Swap the model, keep the build.
- Redesign and build in one 90-day sprint, measured against a baseline you saw first.
THE TEAM
Run by operators.
Agent Forge Works is led by Rui Luis, who has spent three decades building and running technology companies as a founder and CEO, and who held the functional seats along the way: CFO, VP of Manufacturing, Head of Engineering, and most of the others. He has lived the processes this firm automates first hand, from the month-end close to the production floor, which is a different starting point from an AI developer who has never run one. The person who scopes your work is the person who ships it, with forward-deployed engineers from a small bench in your office for the build, and nobody here has a platform to sell you.
“Rui and I co-founded DPS and then Digital Rapids, and building both, I watched him run every function in the business at one point or another. He knows where work gets stuck because he has done that work himself. That is rare in anyone selling AI today, and it is the reason I’d trust him inside my own systems.”
Rui Luis — Founder · Ontario, Canada
LinkedIn →FAQ
FAQ
No. We're vendor-neutral and recommend whatever fits — including "don't build this".
BOOK A CALL
Let's find the first dollar.
30 minutes. No deck. Tell us where the work piles up. You leave knowing which workflows are worth handing to agents first, what the loops cost you today, and what it would take.
Book a call
30 minutes · no deck · no cost. Bring the work that piles up.