
Your AI Chat Just Changed: A Practical Checklist for Vancouver Businesses
If your Vancouver or BC business uses AI chat on your website, SMS, or social DMs, the “brain” behind those replies is not a set-and-forget setting. Platforms update recommended models. Older options get retired. Your job is to switch calmly, test properly, and keep after-hours lead handling reliable.
This is a practical checklist — not a tech deep dive. Use it whether you have one agent or several across locations.
What AI chat does for local businesses
Configured well, a prompt-based chat agent can:
- Answer common questions about services, service areas, hours, and next steps
- Qualify leads with structured questions you approve
- Capture contact details into your CRM
- Book or offer booking links when calendars are connected
- Hand off to a human when the conversation needs judgment, urgency, or relationship
For Metro Vancouver clinics, trades, coaches, brokerages, and studios, the point is speed and consistency after hours and during busy periods — without inventing answers your team would never say.

Why an upgrade checklist matters
Models differ in how closely they follow your prompt, how they handle edge cases, and how they sound. A small drift in tone or booking rules can frustrate a real lead. When your AI chat platform recommends a newer model — or marks an older one for retirement — treat it as an operations task: switch, then test before you treat the change as live.
Vendor “pass rates” and marketing claims are useful context. Your own test suite still decides whether the new model is ready for your traffic.
Upgrade checklist
Work through this in one sitting if you have a handful of agents; split by location or brand if you manage more than one.
1. Inventory every chat agent
In each relevant workspace:
- Open your AI / Conversation agents list.
- Open each agent.
- Note the current model on the build or settings tab.
- Flag anything on an older or retiring model as high priority.
- List agents that were recently migrated so you can smoke-test them too.
Keep a simple spreadsheet: agent name, location/brand, model, channels (chat widget, SMS, Facebook/Instagram, etc.), last tested date.
2. Snapshot what “good” looks like today
Before you change the model:
- Export or copy the current system prompt / instructions
- List must-pass scenarios (service area yes/no, after-hours booking, emergency escalation, “talk to a human,” pricing deflection if you do not publish prices in chat)
- Save 5–10 real recent transcripts that represent how you want the agent to behave
This is your regression pack. Without it, you cannot tell if the new model improved answers or just changed voice.
3. Switch to the recommended model
For each agent still on an older model:
- Open the agent → Build / Model settings.
- Select the platform’s recommended model for prompt-based agents.
- Do not treat Save as the finish line — test after you switch because each model writes in its own way.
- Prefer: select the new model → run tests → only then save once behaviour matches your rules (or save, test immediately, and roll back if your process allows).
If your UI forces Save before Test, document that order and keep a copy of the prior prompt/model so you can revert.
4. Test before you trust live traffic
Run the same scenarios you captured earlier:
- In-service-area booking happy path
- Out-of-area polite decline + alternative (callback / waitlist)
- After-hours / weekend behaviour
- Incomplete contact info (agent should ask, not invent)
- Sensitive or high-stakes topics → human handoff
- “Stop” / unsubscribe / “talk to a person” phrasing
- Any bilingual or local Vancouver wording you rely on (neighbourhoods, municipalities, 604/778 habits)
Watch for: over-promising, hallucinated prices or timelines, weaker booking use, or a tone that no longer matches your brand.
5. Recheck channels and triggers
A model change does not rewrite your workflows, but a different reply style can break assumptions in website chat, SMS quiet hours, social DM routing, and automations that fire on keywords or AI outcomes. Send one controlled live conversation per channel after the test suite passes.
6. Monitor the first week
Assign an owner. For seven days, skim transcripts daily. Tag misses. Adjust the prompt — not “turn AI off” at the first odd reply. Log what went wrong so you tighten instructions instead of guessing.
Who should care in Metro Vancouver
- Owners already using AI chat for after-hours website or SMS conversations
- Teams managing multiple locations or brands from one system
- Businesses pairing chat AI with phone reception and booking calendars
- Anyone still on an older model who cannot afford a surprise retirement mid-campaign
If you have never opened AI chat, start with clear FAQs, escalation rules, and CRM write-back. Model choice is secondary to that foundation.
How SMA Ventures helps
SMA Ventures helps BC businesses implement and tune client systems — chat prompts, booking rules, CRM pipelines, and the handoff between chat, SMS, and phone. We did not invent the underlying models. Our job is making the switch boring and safe for your real traffic.
Next step
If this is something you want to simplify, send me a message and we’ll map out what makes sense for your business.
— Simon, SMA Ventures

