Burnaby Staff Reviews: Let AI Draft, You Decide

It's late October, and if you run a shop, clinic or trades crew in Burnaby, staff reviews are probably sitting somewhere on your list between holiday scheduling and year-end books. Most owners don't dread the conversation itself. They hate the blank page: twelve months of notes, texts and half-remembered wins that have to become something fair and specific.

That blank page is where AI is genuinely useful. The decision about someone's raise, rating or future isn't. This post covers how to split the work, what the evidence says about the risks, and the BC privacy rules you'll want to respect before you paste anything into a chatbot.

Why owners are trying this now

You wouldn't be early. Statistics Canada's Q2 2026 analysis of AI use by businesses (released June 11, 2026) found that 19.2% of Canadian businesses used AI to produce goods or deliver services in the prior 12 months, up from 12.2% in Q2 2025 and 6.1% in Q2 2024. Among firms with 1 to 4 employees, the figure was 19.9%.

The same release is a useful reality check. 40.0% of businesses said AI simply isn't relevant to them, and the top barrier among those that cited one was cybersecurity or privacy concerns, at 13.4%. That second number matters for reviews more than for almost any other task, because a performance review is about as personal as business data gets.

What AI does well in a review cycle

Think of the model as a fast, tireless assistant who has never met your staff. It's good at shaping material you give it. It's bad at knowing anything you didn't.

Tasks where it earns its keep:

• Turning messy notes into a structured draft. Feed it your own bullet points (missed deadlines, the customer who wrote in to praise someone, the new hire they trained) and ask for a draft organized around your existing review headings.
• Checking tone. Ask it to flag phrases that sound vague, harsh or like personality judgments rather than observed behaviour. "Not a team player" becomes a prompt to find the actual example.
• Drafting goals. Given a role description and a few priorities, it can propose specific, measurable goals for next year that you then edit.
• Consistency across a team. Running every review through the same template makes it easier to spot when one person got three paragraphs of praise and another got two lines.
• Prepping the conversation. It can suggest open questions to ask, or help you rehearse how to deliver a hard message calmly.

None of that requires the AI to decide anything. It's editing and organizing, which is the part that eats your evenings.

What the research says can go wrong

There are two well documented problems, and both show up in exactly this kind of writing.

Bias in the language itself. Researchers tested this directly in “Kelly is a Warm Person, Joseph is a Role Model”: Gender Biases in LLM-Generated Reference Letters (Wan et al., EMNLP 2023 Findings). They found significant gender biases in both the language style and the content of letters generated by ChatGPT and Alpaca, and they warned against using these models for that task without careful review. A fair caveat: those were 2023 models, and newer ones may behave differently. But a recommendation letter and a performance review are close cousins, and you have no way to know whether the tool you're using has fixed the problem unless you check the output yourself.

Confident errors. NIST's Generative AI Profile (NIST AI 600-1, July 2024) lists "confabulation" as a core risk: "the production of confidently stated but erroneous or false content." In a review, that looks like an invented accomplishment, a project credited to the wrong person, or a polished sentence that says something you never observed. The same document warns about automation bias, which it describes as "excessive deference to automated systems," and notes it can make confabulation and bias worse. In plain terms: the more finished the draft looks, the less carefully people read it.

Put those together and the rule writes itself. The AI drafts from your facts. You verify every claim and decide every rating.

The BC privacy rules that apply

Most private employers in British Columbia fall under the Personal Information Protection Act (PIPA). Review notes are almost certainly "employee personal information," which the Act defines as information collected, used or disclosed solely for purposes reasonably required to establish, manage or terminate an employment relationship.

A few provisions are worth knowing:

Use without consent is allowed, with conditions. Under section 16, you can use employee personal information without consent if the use is reasonable for managing the employment relationship, but you must notify the employee that you'll be using it and why, before you do.

Disclosure needs the same care. Section 19 sets out a parallel rule for disclosure. Whether sending review notes to an outside AI service counts as a disclosure depends on the service and your agreement with it, which is a question for your privacy advisor, not a blog post. What's clear is that you should know where the data goes and whether the vendor keeps or trains on it.

Employees can ask to see it. Section 23 gives individuals a right to access their personal information and to learn how it has been used. Assume anything you write, including AI drafts you saved, could be read by the person it's about.

Federal guidance points the same way. The Office of the Privacy Commissioner of Canada and its provincial and territorial counterparts published Principles for responsible, trustworthy and privacy-protective generative AI technologies in December 2023. It names employment as a high-impact context, asks organizations to take reasonable steps to make sure AI outputs are accurate when they inform decisions about people, and states that accountability for decisions rests with the organization, "not with any kind of automated system."

A simple workflow you can run this year

Here's a setup that keeps the useful part and drops the risky part:

1. Tell your team first. A short note saying you'll use an AI writing tool to help draft reviews from your own notes, and that you make every decision. That covers the PIPA notification and builds trust.
2. Strip identifiers. Use initials or "Employee A" in prompts. Leave out health information, family details and anything unrelated to the job.
3. Use an account with clear data terms. Check your tool's current business terms on retention and model training before you start. Settings and policies change, so look at the vendor's own page this month rather than relying on what was true last year.
4. Draft from facts only. Paste in your notes and tell the model not to add accomplishments or issues that aren't in them.
5. Read it as a skeptic. Check every claim against your records. Watch for gendered or personality based wording. Swap people's names between two drafts and see whether the tone shifts.
6. Decide separately. Set ratings and pay changes yourself, before or apart from the drafting, so the polished text doesn't steer the outcome.

For an illustrative sense of time: if a 10 person team currently takes you an evening per review, getting the first draft from your notes might cut each one to the time it takes to verify and edit. Your mileage depends on how good your notes are.

Where this doesn't apply

Be honest with yourself about the limits:

• No notes, no help. If you didn't keep records through the year, AI can't recover them. It will happily invent plausible ones, which is worse than a short review.
• Discipline, termination or pay disputes. When a review may become evidence, get advice from an employment lawyer or HR professional. A drafted paragraph is not a defensible process.
• Very small teams. If you have three staff and talk to them every day, a direct conversation and a few handwritten lines may serve everyone better than a template.
• Unionized workplaces. Your collective agreement may set rules for evaluations that come first.
• Unclear vendor terms. If you can't confirm where the data is stored or whether it's used for training, don't paste employee information in. Wait.

Statistics Canada's finding that 40% of businesses see AI as not relevant isn't wrong for everyone. Some review cycles are better left manual.

Sources

• Statistics Canada. Vicky Do, Shivani Sood and Chris Johnston. "Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026." Analysis in Brief, Catalogue no. 11-621-M, June 11, 2026. https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2026010-eng.htm
• Wan, Y., Pu, G., Sun, J., Garimella, A., Chang, K.-W., Peng, N. "“Kelly is a Warm Person, Joseph is a Role Model”: Gender Biases in LLM-Generated Reference Letters." Findings of EMNLP, 2023. arXiv:2310.09219. https://arxiv.org/abs/2310.09219
• National Institute of Standards and Technology. "Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile" (NIST AI 600-1), July 2024. https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf
• Province of British Columbia. Personal Information Protection Act, SBC 2003, c. 63. https://www.bclaws.gov.bc.ca/civix/document/id/complete/statreg/03063_01
• Office of the Privacy Commissioner of Canada. "Principles for responsible, trustworthy and privacy-protective generative AI technologies." December 2023. https://www.priv.gc.ca/en/privacy-topics/technology/artificial-intelligence/gd_principles_ai/

If you'd like a review drafting assistant set up properly, with locked down data settings and a workflow your team understands, book a free call with Autana. We're local to Burnaby and New Westminster, and we'll tell you straight if your business is better off waiting.