From Copilot alone to AI agents, data pipelines and clear rulesAI adoption case study:

The client and the results Results at a glance

The company started with Copilot and nothing else. This is what changed in three parts of the business.

Client

A mid sized Toronto company, name withheld

Company

Non technical, several brands

Starting point

Copilot only, no agents

Systems

NetSuite, Power BI, a project tool

• 120+

  Demand planninghours of manual work saved every monthThat is more than three 40 hour work weeks, every month

• 4 → 2

  Customer servicestaff for the same workload

• Real time

  Managementproject status, straight from the project toolStatus slides built automatically

The starting point Core processes ran on exports, Excel and memory

Before DevToDollars came in, AI at the company meant Copilot: something staff asked for help, never an agent that could take a task and finish it inside the company's own systems.

• Excel by hand

  Staff worked through thousands of exported rows by hand in Excel, and errors crept into the results.

• Every department on its own

  Each department had its own processes, data sources and tools, so their data did not always agree.

• Status from memory

  The project tool was out of date, so the PMO (the team that tracks projects) built its status slides by hand, from memory. That meant management could not see which projects were late, by how much, or why.

Leadership knew all of this, and had measured it. What it lacked was someone to fix it. It was also wary of AI: without clear rules, it worried AI could affect the company's tools and data.

Discovery and assessment First, a map of how every department works

DevToDollars did not start by building anything. Its team sat down with each department's leaders and senior managers, asked them all the same four questions, and mapped how each department works today and where its time goes.

1. 1

   Which departments share data with each other?

2. 2

   Which workflow takes each department the longest?

3. 3

   Which outside tools does each project rely on?

4. 4

   How does each department do its work today, in detail?

What the answers showed

The solution No single tool was forced on every team

With the order of work set, DevToDollars took each department's longest workflow on its own terms. AI agents fit the PMO and marketing. Demand planning needed plain code.

• AI agent

  PMO

  Status reports built from the project tool

• AI agents

  Marketing

  Campaign updates pulled from the web

• Plain code

  Demand planning

  Pipelines that feed the forecasting models

PMO reporting AI agent

Status slides, no longer built from memory

The PMO used to build every status slide by hand. Now an AI agent reads each major project's stages from the project tool and builds the slide, so management sees every project as it stands.

In development. This product is on hold while the brand team decides. Its formula submissions sit on the right and its milestones along the bottom.

In production. This one is on track. Once a product is in production, purchase orders take the place of formula submissions, and the milestones run up to the earliest required ship date.

How ICM works

ICM, short for Interpretable Context Methodology, is an open method that splits a job into small written steps. Anyone can follow what the agent did, the steps that need no AI run as plain scripts, and a person reviews each stage.

Demand planning Plain code

Plain code where AI was not needed

Demand planning did not need an AI agent. It needed a reliable way to get data into the forecasting models without staff exporting it and feeding it in by hand. DevToDollars wrote code to do exactly that, and introduced new prediction models that forecast more accurately.

Before

• Data exported by hand
• Forecasting models fed by hand

After

• Exports automated
• Pipelines in code feed the models
• New prediction models, more accurate forecasts

Why code here

AI models are paid for by the token, roughly by the word, so every run costs something. Code has no token cost and does the same thing every time. Because DevToDollars builds software as well as agents, it did not have to reach for AI everywhere, and could use whichever tool saved the company more time and money.

About1 in 5

companies say AI running costs, tokens included, limit how much they use AI

Source: McKinsey, The state of AI in 2026, August 2026.

A dashboard for management

DevToDollars also built a custom dashboard that gives management a bird's eye view of demand planning. It sets what the company shipped to each customer (from NetSuite) beside what that customer sold (from Power BI), on one screen.

1. 1Every retail customer and its invoiced total
2. 2Units sold per store, week by week

Click the dashboard to see it full size. Names are placeholders.

12

Security, data and governance Where to trust AI, and where not to

With agents now working inside company systems, AI sat close to the company's tools and data, which was leadership's biggest worry from the start. So DevToDollars walked every employee through where AI can be trusted, and where it cannot.

Where to start This company's leaders already knew what was broken, and had measured it.

That head start took it from Copilot alone to agents, pipelines and clear rules. The lesson for similar companies: fix the data before any AI goes in, and use AI only where plain code will not do the job.

If your leaders know what is broken in your company, the next step is mapping how each department works. DevToDollars starts the way it started here: an assessment, department by department, before anything is built. It suits companies with no one inside who has led AI work before. In Toronto, the team can work on site, alongside the people whose work is changing.