Governed AI Workflow Pilot
One of your business processes, running as a multi-agent AI workflow with a person signing off at each step, in about four weeks, at a fixed price.
Who it is for
Teams that have tried AI chat tools and want the next step: a repeatable process that does real work, with a person approving what matters. Also consultancies whose clients are asking for agentic AI and who want a delivery partner.
What "governed" means here
The workflow comes with best-practice guardrails built in. People sign off at review gates, the scope is defined, and the risks are written down. Each step gets only the access it needs, through an API or a defined event, and never direct access to a database. That is good practice for one workflow. It is not an AI governance programme for your organisation: no policies, risk frameworks or committees. If you need that, we can talk about it separately.
How it is built
SPEAKMAN.AI runs the workflow: the steps, the order and the review gates. Your AI assistant does the work, using the connections it already has to your systems, such as SharePoint, email or your CRM. It has been tested with Claude Code, the Claude app and Microsoft Copilot Studio.
There is no custom application to build or maintain. A pilot connects to up to two of your systems. If a connection doesn't exist yet, building one is a separate project.
Where it runs
- During the pilot. The SPEAKMAN.AI engine runs on your process owner's computer, inside their AI assistant (Claude Code or the Claude app). It uses your own model account. Nothing I own sits in the middle, and no subscription is needed.
- Microsoft Copilot Studio. For a single-user proof of concept, Copilot Studio connects to that local engine over a secured connection. For team use, the best route is the team edition below. The alternative is a container hosted in your own cloud: I help build it, and your team deploys and runs it.
- When more of your team needs it. The workflow moves to the SPEAKMAN.AI
cloud. That gives shared access, a robust database and services that run in parallel. There
are two ways to run it, both quoted separately with their own data terms, because there my
platform does sit in the middle:
- Team edition: a separate tenant on shared infrastructure. Set up when a pilot needs it, and quicker to start than a private tenant.
- Private tenant: a dedicated deployment, in a region you choose, such as Canada. It takes longer to start, because the deployment is set up for you.
If your data has to stay in a particular country, tell me early. That decides the route and the start date.
What you get
- A process chosen together. One process with a clear start, a clear end and a person who owns it today.
- Success criteria agreed up front. One or two measurable criteria, checked against 20–50 of your real examples.
- A working workflow. Multi-agent, with review gates where your team signs off.
- Your model, your data. It runs on your own assistant or your own model account, in the cloud as well as locally. You pay your own model usage, and model calls go only to the provider you chose, under that provider's terms.
- The architecture written down. What it does, what each step can reach and why, where the data goes, what each run costs and what the risks are.
- Handover and a recommendation. A working session with your team, and a clear view on whether to take it further. The workflow is yours to keep and change.
How it runs
| Week | What happens |
|---|---|
| 1 | Pick the process, map it, agree the success criteria and the examples |
| 2–3 | Build, run on your examples, adjust with your reviewers (two rounds) |
| 4 | Results against the criteria, architecture document, handover |
What is not included
- Custom application development or new system integrations.
- Data clean-up: the examples and access are needed by day 3.
- Single sign-on, security audits, service levels and production deployment pipelines.
- An organisation-wide AI governance programme.
Price
From CA$7,500 plus HST. Fixed, and agreed before we start.
Why me
I have spent 25 years in enterprise solution architecture, designing integrations around events, APIs and minimal access to data. I design AI workflows the same way. For the last year I have built and shipped SPEAKMAN.AI, a multi-agent workflow engine with human review gates, now released as a signed Windows app. I run my own business on the same workflows I would build for you.
What I do not promise
Headcount savings or ROI percentages. You get a working process, measured on your own examples.
Next step
A 30-minute call to see whether you have a process that fits. Book a time, or email david@speakman.ai. If you'd rather start by describing a process, use the form below and I'll come back to you.