AI
AI-Personalized Holiday Cards for Burnaby Firms
Every December, Burnaby accounting firms, law offices and property managers hit the same small problem. They can send a stack of identical cards with a printed signature, or they can block off an afternoon to write personal notes nobody really has time for. AI can cover the middle ground. It drafts a short, specific note for each client from what your team already knows, and then a human checks it and signs it.
This post covers how that works, what it costs at current published prices, which Canadian rules apply, and when you're better off just using a pen.
What "personalized" actually means here
A good holiday note doesn't have to be clever. It mentions one real thing. Maybe the client opened a second location, finished a tricky renovation, closed on a house in New Westminster, or stuck with you through a messy tax year. That one detail is what makes a card feel written instead of printed.
Most firms already have that detail somewhere. It's in CRM notes, recent email threads, a file summary, or a partner's memory. So the job for AI is narrow. It reads a short summary for each client and drafts two or three warm sentences in your firm's voice. It doesn't decide anything about the client. It writes a first draft that a person approves.
A workflow that holds up
Here's the setup we'd suggest for a firm sending somewhere between 50 and 500 cards.
- Build a lean client sheet. One row per client: a first name or placeholder, the relationship owner, and one or two plain notes ("moved the office to Metrotown in spring", "first year filing as a corporation"). Leave out financial figures, health details, case specifics and anything you wouldn't want quoted back to you.
- Write a style guide once. Cover tone, length, sign off, phrases to avoid, and include two or three sample notes your partners actually like.
- Draft in a batch. Run every row through a model as one overnight batch job. Don't chat with it card by card.
- Review by relationship owner. Each person reads and edits the drafts for their own clients. If a draft sounds off, rewrite it or swap in a simple generic line.
- Produce and send. Merge the approved text into your print template or e-card tool. Keep the sheet and approvals for your records, and delete any prompts you don't need.
Don't skip the review step. A model can get a detail subtly wrong, mix up two clients, or congratulate someone on something they'd rather forget. Canada's privacy regulators say the same thing in their joint generative AI principles. They ask organizations to "take reasonable steps to ensure that any outputs from a generative AI tool are accurate as necessary for the purpose" (OPC, Principles for responsible, trustworthy and privacy-protective generative AI technologies).
What it costs, using current published prices
For a job like this, the model bill is tiny. The real cost is the staff time spent reviewing drafts.
As of October 2026, Anthropic's pricing page lists Claude Sonnet 5.5 at $2 USD per million input tokens and $10 USD per million output tokens. It lists Claude Haiku 5.5 at $0.10 and $0.50 for prompts up to 100,000 tokens. The same page says the Batch API gives "a 50 percent discount on both input and output tokens" for asynchronous jobs. That fits holiday cards well, since nobody needs the drafts in real time.
Here's an illustrative example, not a client result. Say you send 300 cards. Each prompt carries about 1,500 tokens (style guide plus client notes) and each draft comes back at about 150 tokens. That's 450,000 input tokens and 45,000 output tokens. At October 2026 batch rates for Sonnet 5.5 ($1 input and $5 output per million), you'd pay about $0.45 plus $0.23. That's under one US dollar for the whole run.
Treat those token counts as a ballpark. Anthropic's page notes that its newer tokenizer produces "approximately 30% more tokens for the same text" than older models, and real prompts vary. Even at triple the estimate, the API cost hardly matters. What matters is whether 300 drafts save your partners more time than they spend editing them. If a partner rewrites every second note, most of the gain disappears.
The Canadian rules that apply
Privacy. Most private sector firms in BC fall under the Personal Information Protection Act (some organizations fall under federal PIPEDA instead, so check which one covers you). Section 11 limits collection to "purposes that a reasonable person would consider appropriate." Section 34 requires "reasonable security arrangements" for personal information in your custody. Sending client notes to an outside AI provider is a new use of that information, so read the provider's data handling terms and your own privacy policy before you start.
The OPC principles also suggest that "where possible and reasonable" you "use anonymized or de-identified information within prompts." With cards, that's easy to do. The model doesn't need a surname, an address or an account number to write "hope the new Metrotown office is treating you well." Add the names afterward, in the merge step.
Anti-spam. If the card goes out by email, check Canada's Anti-Spam Legislation. A commercial electronic message is one where, judging by its content, links or contact information, it's reasonable to conclude the purpose is "to encourage participation in a commercial activity." A plain holiday greeting may not meet that test. An e-card that also promotes your year end tax planning slots probably does. When CASL applies, section 6 requires consent, sender identification and an unsubscribe mechanism.
Implied consent based on an existing business relationship has a time limit. For a purchase, the relationship has to fall within the two-year period before the message is sent, and for an inquiry it's six months. So a client you last served three years ago may not qualify. The simple fix is to keep the greeting free of promotion, or treat the e-card like any other marketing email. This isn't legal advice, so if you're unsure, ask your lawyer.
Light governance is enough
You don't need a formal AI program for holiday cards. The NIST AI Risk Management Framework, released January 26, 2023, still gives you a useful frame with its four functions: govern, map, measure and manage. For a card run, that comes down to four things. Someone owns the project. You know what data goes in. You spot check the output. And you have a plan for when a bad note gets through. If you want more depth, NIST published a Generative AI Profile (NIST AI 600-1) on July 26, 2024.
Where this doesn't apply
Small client lists. If you have 30 key clients, write the cards by hand. Setup and review will eat most of the time you'd save, and handwriting sends a signal that AI drafted text can't.
Thin or stale notes. If your CRM is mostly empty, the model has nothing real to work with. It'll give you generic warmth, or worse, make up specifics. A generic card beats a confidently wrong one.
Sensitive relationships. Think family law, estates, insolvency, health, or any client who had a hard year. A cheerful line about "a great year" can land badly. These clients deserve a note from a person or a simple, neutral card.
Clients who would mind. Some people would feel let down to find out software drafted their note. The OPC principles put a lot of weight on transparency. Whether to disclose AI help on something as low stakes as a greeting is a judgment call, but never present a note as handwritten if it wasn't.
Firms without a review habit. If nobody will actually read the drafts, don't automate this. Unreviewed output is where the real risk sits, not in the API bill.
Getting started this season
Aim to have your client sheet cleaned up by mid November. Test on ten clients, have partners grade the drafts, and adjust the style guide. Then run the full batch, review, and send in early December. If the test drafts don't beat your current cards, that's useful to know too, and it means you can stop there.
Sources
- Office of the Privacy Commissioner of Canada, with provincial and territorial privacy regulators. Principles for responsible, trustworthy and privacy-protective generative AI technologies. 2023. priv.gc.ca
- Anthropic. Pricing (Claude Platform documentation). Accessed October 2026. platform.claude.com
- Province of British Columbia. Personal Information Protection Act, SBC 2003, c. 63. bclaws.gov.bc.ca
- Government of Canada, Justice Laws Website. Canada's Anti-Spam Legislation, S.C. 2010, c. 23, sections 1 to 6. laws-lois.justice.gc.ca
- Government of Canada, Justice Laws Website. Canada's Anti-Spam Legislation, S.C. 2010, c. 23, section 10. laws-lois.justice.gc.ca
- National Institute of Standards and Technology. AI Risk Management Framework. 2023. nist.gov
If you'd like help setting up a holiday card workflow your partners will actually trust, book a free call with Autana Solutions. We're local to Burnaby and New Westminster, and we'll tell you honestly whether it's worth automating for your client list or whether a pen will do the job better.
Want an AI employee for your business?
We install a 24/7 AI worker for businesses in Vancouver, Burnaby, and beyond. Book a free Discovery Call.
Book a call →

