Is documentation included?

It should be. At minimum you should receive a workflow map, a runbook explaining how to run, pause, and review the agent, and a list of every account and credential it uses. If a proposal does not name documentation as a deliverable, ask for it in writing before you sign.

Is monitoring included after launch?

A short early-monitoring period, usually two to four weeks, should be included so the agent can be tuned while your team settles in. Ongoing monitoring beyond that is normally a separate monthly arrangement, and published Canadian examples range from about $300 to $800 a month for maintenance.

What happens if the scope changes mid-project?

A good scope says how changes are handled before work starts: usually a short written change note with the impact on timeline and price, agreed before the extra work begins. Small clarifications should not trigger a change; a new tool, a new workflow, or a new audience should.

What does an AI agent implementation project include?

An AI agent implementation project should include six phases (scope, design, build, test on real work, launch and handoff, and early monitoring) and should leave you with nine things in writing: a scope, a workflow map, a baseline measure, the working agent in accounts you own, a test log, a runbook, an access list, a handoff walkthrough, and an after measure. If a proposal cannot tell you which of these you get and when, it is a build quote, not an implementation.

The distinction matters. A build puts an agent in place. An implementation leaves your team able to run it, check it, and turn it off without calling anyone.

Updated 2026-09-24 by Esteban Valencia, EV Advisory. Written for founder-led businesses in Canada and the United States.

What are the phases of an AI agent implementation?

Phase	What happens	What you receive
1. Scope	The provider learns the process, the tools involved, who owns the work, and how success will be measured.	A written scope with the price, timeline, and success measure
2. Design	The process is mapped as it runs today and as it will run with the agent, including every point where a person reviews or takes over.	A workflow map and a baseline measure of the current process
3. Build	The agent is built inside your accounts and connected to your tools with the minimum access it needs.	The working agent, in a test mode
4. Test on real work	The agent runs against real past examples, then alongside your team on live work with a person approving every output.	A test log showing inputs, outputs, and corrections
5. Launch and handoff	The agent moves into daily use, and the process owner is trained to run it.	A runbook, an access and credentials list, and a recorded walkthrough
6. Early monitoring	Error rates and exceptions are watched and the agent is tuned.	An after measure compared against the baseline

What deliverables should you receive?

Each deliverable exists for a reason. If one is missing, ask why.

Written scope. Protects both sides. It says what the agent does, what it does not do, and how success is measured.
Workflow map. Shows exactly where the agent acts and where a person steps in. This is what you point to when something goes wrong.
Baseline measure. How long the process takes and how often it runs today. Without it, nobody can say whether the agent helped.
The working agent. Running in accounts your business owns, connected only to the tools named in the scope.
Test log. The real examples the agent was tested on and what was corrected. This is your evidence it works on your work, not a demo.
Runbook. Plain-language instructions to run, pause, and review the agent, and what to do when it gets something wrong.
Access and credentials list. Every account, API key, and permission the agent uses, so you can revoke or move anything yourself.
Handoff walkthrough. A recorded session with your process owner.
After measure. The same measure as the baseline, taken once the agent is in daily use.

The Government of Canada’s Digital Standards make a similar point for public services: design with users, iterate and improve frequently, and address security and privacy risks from the start. The same habits make a small-business agent last.

What is usually not included?

Expect these to be excluded or quoted separately, and check the scope says so:

Software licences and usage costs. Tool subscriptions and model usage are normally billed to your accounts directly.
Cleaning up data beyond the workflow. If your CRM needs a clean-up before any agent can use it, that is a separate job.
New workflows. A second agent, or extending the first to a new process, is a new scope.
Ongoing monitoring beyond the early period. Usually a separate monthly arrangement.
Legal or privacy advice. A provider can build in safeguards, but only your lawyer or privacy advisor can tell you what your obligations are.

Who owns the agent and the accounts after handoff?

You should. Insist on three things in writing:

The accounts are yours. Tool accounts, model provider accounts, and automation platform accounts are opened in your business’s name, with your billing.
The configuration is yours. Prompts, instructions, workflow definitions, and any custom code written for you are delivered to you.
Access is removable. The provider’s access can be revoked on the day the project ends without breaking the agent.

If a provider runs your agent inside their own account, you do not own it; you rent it. That can be a fine choice, but it should be a choice, not a surprise.

What happens after launch?

For the first two to four weeks, the agent should be watched closely: how often a person has to correct it, which inputs it struggles with, and whether the time saved matches the baseline. After that, the process owner runs it day to day using the runbook. A light monthly check (error rate, cost, and whether the process has changed) keeps it healthy. When the process changes, the agent needs to change with it.

Sources

Government of Canada, Government of Canada Digital Standards.
Innovation, Science and Economic Development Canada, Toolkit for SMEs deploying artificial intelligence.
Office of the Privacy Commissioner of Canada, AI, privacy, and your business.
Planwebbies, AI Automation Cost in Canada: Small Business Pricing (2026), for published maintenance ranges.

Next step

This page describes how EV Advisory delivers every project. See the AI agent implementation service page for the timeline and what your team does, or book a free 20-minute Live Audit to scope your first agent.