AI in the flow of work — a 90-day pilot for TD Workforce Capability
Re: AI workforce pilot
muhammadasim@codeaza.org · codeaza.com
TD is setting up its AI workforce function from an early stage. The goal is not more AI tools. It's AI that changes how people actually work, that teams trust, and that leadership can see paying off.
Three ideas shape this proposal:
A third automated, a third augmented
Most banking work will change, but not evenly. The first job is knowing which tasks to automate, which to augment and which to keep human, team by team.
A context problem, not a data problem
The signals leaders need already exist across hiring, skills, org structure and workforce planning. They just aren't connected in a way that answers real questions.
Learning in the flow of work
People give formal learning minutes a week. The help that sticks arrives while the work is being done.
To be confirmed with you: priorities, first team, and what a win looks like by day 90.
AI opportunity map
For one business unit, we break each role into tasks and mark each as automate, augment or keep human, with the evidence behind it.
- Uses job descriptions, procedures and process documents. No customer data.
- Output: a prioritised list of use cases and a 90-day roadmap.
People context layer (prototype)
A working prototype that brings key workforce signals together so leaders get answers they can trust, each one traceable to its source.
- Built inside TD's environment on the Microsoft stack (Fabric / Azure).
- De-identified or synthetic data until TD approves otherwise.
One live pilot, measured
One assistant inside one team's daily work, for example guidance in the flow of work or support for managers, with accuracy testing built in.
- A person approves every action. Every step is logged.
- Quality scored on real tasks, so results are evidence, not anecdote.
Muhammad Asim — AI solutions architect
Leads design and delivery and is your single point of contact, with fixed overlap hours in Toronto time.
Codeaza engineering team
Two engineers building data pipelines, retrieval and agent workflows, and the accuracy testing.
Up Sparrow — delivery partner
Partner relationship and contracting.
A fixed-price start, then a capped monthly team. Final figures follow once scope is agreed.
Insurance document platform
A 12-step extraction pipeline taken from about 70% to 90%+ correct, with quality reported at every step so problems surface before release.
Jubilee Life · Rulebook
AI and data work in regulated environments.
A conversation
Your priorities, the first team, and what a 90-day win looks like.
Confirm phase 1
Fixed scope and price, contracting route agreed.
Start within two weeks
Kick-off and access to non-sensitive documents.