When AI Tools Go Down, Dispatch Needs a Backup

Claude and OpenAI both had visible reliability incidents this week.

Claude's public status page listed elevated error rates across multiple models on June 23, 2026, followed by an unresolved Claude Opus 4.8 incident on June 24. OpenAI's status history shows file upload and download errors in ChatGPT on June 23, plus Codex capacity and cloud-task issues on June 16.

The same day, Axios reported that OpenAI had started testing its first custom inference chip, Jalapeno, for jobs similar to answering Codex queries. That is good context. The biggest AI companies are still spending real money and engineering time on capacity because serving AI reliably is hard.

For a Regina HVAC company, a Saskatoon plumbing dispatcher, a roofing estimator, or a property maintenance office, the lesson is plain: if AI helps with customer-facing work, the fallback should be written before the tool is busy, slow, or unavailable.

Treat AI like a busy staff tool

Most service businesses already understand fallback rules.

If the phone system drops, calls route to a cell. If the card terminal fails, the office knows how to invoice. If a supplier portal is down, the parts counter can still phone the order in. Nobody calls that innovation. It is just operations.

AI deserves the same treatment once it touches intake, dispatch notes, quote follow-up, file search, job summaries, customer updates, or closeout paperwork.

The mistake is treating an AI assistant as either magic or optional. In practice, it is often neither. If staff use it every day, it becomes part of the work path. When it slows down, staff need a clear instruction that keeps the business moving.

Pick the workflows that cannot wait

Do not write a fallback plan for every AI use. Start with work where delay creates a customer problem.

For service contractors and field teams, that usually means:

• •after-hours lead capture and urgent callback summaries
• •dispatch notes for same-day jobs
• •quote follow-up messages for warm leads
• •warranty or complaint intake
• •safety-related job notes
• •customer updates when a crew is delayed
• •files needed for a site visit, permit, invoice, or closeout packet

If AI is only helping rewrite a blog post or clean up an internal note, a short delay may be fine. If it helps staff decide who gets called back first on a cold night, the fallback matters.

A useful test is simple: if the AI tool failed for two hours, would staff know what to do without asking the owner?

Write the paper version once

The fallback should be boring enough to run under pressure.

For a dispatch desk, that may be a one-page backup card:

• •where new calls go when the AI summary fails
• •who checks voicemail, web forms, and missed-call alerts
• •which jobs get phoned back before email replies
• •what customer information staff can collect manually
• •which promises staff cannot make without approval
• •where the final note goes after systems recover

That paper version does not replace the automation. It gives the team a safe way to keep working while the tool recovers.

This is where Prairie AI often starts with service businesses. We map the work that already happens, then decide where AI should draft, summarize, route, or organize. If you want help mapping one customer intake or dispatch workflow, book a strategy call and bring the current phone, form, CRM, and job-note path.

Keep customer promises human

An outage plan should also say what AI is not allowed to decide.

AI can help prepare a callback note. It should not decide emergency priority by itself. AI can draft a quote follow-up. It should not change price, warranty language, or scope without a person. AI can summarize a complaint. It should not decide whether the customer gets a refund, credit, or free return visit.

That line matters more during an outage because staff may feel pressure to improvise.

Write the owner rule in plain language. For example: when AI is unavailable, staff can collect details, send a received-your-message note, and flag urgency. Staff cannot promise arrival windows, discounts, warranty coverage, safety advice, or final pricing unless the usual approver signs off.

The Government of Canada guide on generative AI keeps accountability with the organization using the tool. In everyday business terms, the software can help, but the company still owns the customer commitment.

Test the fallback on a quiet morning

Do not wait for a public status incident to find out the plan is too fuzzy.

Once a month, pick one AI-assisted workflow and run a small drill. Turn off the AI step for 30 minutes. Ask the person who normally handles that work to process three sample items using the backup card.

Watch for friction:

• •the backup inbox is unclear
• •staff do not know who approves a customer promise
• •the CRM note format depends on the AI summary
• •the paper form asks for too much
• •the route from voicemail to dispatch has an extra handoff
• •the final record gets lost after systems recover

Fix the workflow before polishing the prompt. A better prompt will not help if staff cannot find the right file, customer history, service address, warranty note, or approval rule when the assistant is down.

Use status pages as operating signals

Claude and OpenAI status pages are worth checking, but they are not a complete business continuity plan.

Status pages usually describe platform-level incidents. Your team still needs to know which local workflows depend on which tool. A brief AI dependency list can help:

• •tool name
• •workflow it supports
• •staff owner
• •customer impact if unavailable
• •backup process
• •final system of record

That list should live somewhere staff can open without the AI tool. A shared drive, printed desk card, operations binder, or job trailer folder is enough for a small team.

For Saskatchewan businesses already using AI across sales, admin, field service, or document work, Prairie AI can help build this kind of practical dependency map. Use the Contact Prairie AI form if the issue is still fuzzy and describe where AI is touching customers, files, quotes, or dispatch.

What I would do this week

Pick one customer-facing AI workflow. Not the whole business.

Write down the normal path, the AI step, the human approval point, and the system where the final record belongs. Then write the backup version in five plain lines. Give it to the person who would run it on a busy day and ask what is missing.

If the answer is "we would just wait for the tool to come back," the workflow is not ready to depend on.

That does not mean the AI project is bad. It means the business needs a fallback rule before the tool becomes part of real customer service. For related local planning, see AI help in Regina, AI help in Saskatoon, and AI automation services.