08 / AI strategy / Readiness
AI Readiness Assessment
A podcast-format analysis of a major Canadian bank's Microsoft Copilot pilot, scored on eight dimensions of AI readiness. The verdict: ready to start a controlled pilot, not ready to scale it.
| Dimension | Not evidenced | Partial | Strong |
|---|---|---|---|
| Informational | |||
| Technological | |||
| Environmental | |||
| Infrastructural | |||
| Data | |||
| Participants | |||
| Customers | |||
| Process |
01 / The question
Is the bank ready to scale an AI assistant?
A major Canadian bank plans to pilot Microsoft Copilot to scan large internal databases, retrieve past agreements and help draft investment proposals. Strictly, it is an assistant rather than an autonomous agent: it finds and drafts, and a person decides. Banking sharpens the question, because every decision touches client money, client trust and compliance.
The assessment scores the bank on the eight dimensions of AI readiness from Tehrani and colleagues (2024): informational, technological, environmental, infrastructural, data, participants, customers and process. The evidence is one recorded interview with a lead data analyst in commercial banking, who asked to stay anonymous, and the work is delivered as a podcast script.
02 / What the interview said
The technology is rarely the real problem.
The guest's motivation for the pilot was exhaustion: talented people spending much of their day on repetitive work that does not need their talent, while clients expect instant answers at any hour. Her rule for where to start is high volume, structured work, and a mistake that is unpleasant but not catastrophic.
On readiness she was careful: the bank is ready to start but not to go full speed, because technical readiness and operational readiness are different things. Accountability comes first, with defined limits, escalation to a human, and logged actions that can be explained to a regulator. Staff anxiety comes in layers, fear of job loss and fear of becoming irrelevant, and honest, manager-led conversations work better than blanket reassurance. Her one piece of advice: before talking to any vendor, write one clear sentence describing the problem to be solved.
03 / The scorecard
Strong on intent, unverified on data.
Informational readiness and technological maturity are the bank's clearest strengths. Most other dimensions are partial, with a specific gap to close in each.
| Dimension | Rating | Gap to close |
|---|---|---|
| Informational | Strong | Quantify the problem statement; compare Copilot with simpler search or automation. |
| Technological | Strong (maturity only) | Confirm source citation, an evaluation method and a support channel for pilot users. |
| Environmental | Partial | Map obligations (OSFI E-23, OSFI B-10, PIPEDA, Quebec Law 25); name an executive sponsor. |
| Infrastructural | Partial | Fund the pilot, data clean-up and training as a multi-year line; build hybrid talent. |
| Data | Not evidenced (critical) | Inventory sources, retire superseded versions, fix permissions, assign owners, test retrieval accuracy. |
| Participants | Partial | Manager-led change plan; employees co-design the automated tasks; pilot champions; vendor governance. |
| Customers | Partial | Privacy impact assessment before indexing client files; human sign-off on AI-assisted client documents. |
| Process | Partial | Map the real proposal workflow; add an in-tool feedback button and formal go/no-go gates. |
04 / Adoption strategy
Crawl, and lean on a partner.
The framework maps strategies on two axes: the managerial goal, cost reduction or differentiation, and the level of risk. Here the goal is mainly productivity, freeing analyst hours, with a differentiation upside in faster, sharper client proposals. The risk is high: confidential client data, regulators and employees who might reject the tool.
That places the bank in the high-risk half of the matrix, where the paper points to crawling, meaning small, contained use cases with the least impact on the whole operation, and partnership, meaning leaning on a vendor's capability while building one's own. Information retrieval and proposal drafting fit the guest's starting rule well, because the assistant helps rather than decides, as long as a human reviews everything before it goes out.
05 / Recommendations
Five moves, each aimed at a gap.
| Recommendation | What it means |
|---|---|
| 1. Write the problem down and measure it | One clear sentence naming the problem, then time the search-and-draft work for a few weeks to set a baseline. |
| 2. Make data readiness the first workstream | Inventory the repositories, retire or flag superseded contracts, fix access rights and name an owner for every source. |
| 3. Build governance before autonomy | Define what the assistant may do and when it hands off; cite sources; log every action; run privacy, third-party and model-risk reviews; test for bias continuously. |
| 4. Define success on three levels | Model, user and business measures, with go/no-go thresholds agreed before launch and tested on real work, not demos. |
| 5. Treat adoption as the project | Be honest that some roles will change, back it with training and a plan to reinvest saved hours, and let managers lead the conversation. |
The closing verdict: ready to start, not ready to scale, and that is fine. Readiness is built one dimension at a time, and a well-run pilot is how it gets built.