What to look for · Chapter 10
AI automation support, SLAs and maintenance
Most buyers focus on the build and forget the run. But an automation is software, and software needs upkeep. This chapter covers support, SLAs and maintenance.
“An ounce of prevention is worth a pound of cure.”
Why AI automation needs ongoing support
An automation depends on things that change. Any one of them can break it without warning.
- APIs and app connectors get updated.
- AI models get deprecated and retired.
- Your data and edge cases drift.
- Business rules and needs evolve.
Model providers give notice before retiring a model. OpenAI and Anthropic both publish deprecation schedules. Someone still has to act on them. That is what maintenance is for.
Support tiers explained
Support comes at different levels. Pick the tier that fits the risk.
Reactive support fixes issues when you report them. Proactive support watches for them first. A managed service runs everything to an agreed standard.
What an SLA should cover
An SLA turns vague promises into terms. A good one covers these points.
- Response time, by issue priority.
- Resolution time or a fix target.
- Support hours and time zones.
- An escalation path for urgent issues.
- Uptime, for critical automations.
SLA response times by priority
Not every issue is urgent. A good SLA sorts them by priority. These are typical targets.
| Priority | Example | Typical response | Typical fix |
|---|---|---|---|
| Critical | Automation is down | Within 1 hour | Same day |
| High | A key step fails | Within 4 hours | 1–2 days |
| Normal | Minor error or bug | 1 business day | Within a week |
| Low | Small change request | 2–3 days | Next cycle |
What maintenance includes
Maintenance keeps the automation healthy over time. It usually covers these tasks.
- Monitoring and alerting on failures.
- Bug fixes when something breaks.
- Updates when APIs or models change.
- Tuning as your data and needs shift.
- Small improvements and new steps.
- A regular report on what changed.
Questions to ask about support
- What does your SLA promise, in writing?
- How fast do you respond to a critical issue?
- Who monitors the automation after launch?
- What does the support fee include?
- What happens when an API or model changes?
Support by company size
For small teams
A light monitoring plan is often enough. Make sure someone owns fixes when something breaks.
For enterprises
Expect a formal SLA with response and uptime terms. Confirm escalation, on-call cover and reporting.
Common support mistakes
Key takeaways
- Automations need upkeep — plan support up front.
- Get an SLA with response times and escalation.
- Match the support tier to how critical it is.
- Know who fixes it, and what the fee covers.
Comparing agencies?
Browse the directory and ask each about support and SLAs.
Browse the directorySources & further reading
- OpenAI — API model deprecations policy — developers.openai.com
- Anthropic — Claude model deprecations — docs.anthropic.com
Frequently asked questions
Do AI automations need ongoing maintenance?+
Yes. Automations depend on apps, APIs and AI models that change. A model can be deprecated or an API can update. Data and edge cases also drift over time. Without maintenance, an automation quietly breaks. Plan for support from the start, not after it fails.
What is an SLA in AI automation?+
An SLA is a service level agreement. It sets the support you can expect in writing. It covers response time, resolution time and support hours. It may also promise uptime and an escalation path. Get one for any business-critical automation.
What should an AI automation SLA include?+
It should set response and resolution times by priority. It should state support hours and how to raise issues. It should name an escalation path for urgent problems. For critical systems, it may promise uptime. Make sure the terms match how important the automation is.
How much does AI automation support cost?+
Support is often a monthly retainer. Typical retainers run about $2,000–$20,000 a month. The price tracks the volume and criticality of the work. Some agencies bill maintenance by the hour instead. Confirm what the fee includes before you sign.
Who fixes an AI automation when it breaks?+
That depends on your agreement, so settle it early. With a retainer, the agency monitors and fixes issues. Without one, you may pay per incident or fix it yourself. This is why ownership and documentation matter. Know who is responsible before something breaks.
What does AI automation maintenance include?+
It includes monitoring, alerting and bug fixes. It covers updates when APIs or AI models change. It also covers tuning as your data and needs shift. Good maintenance adds small improvements over time. Ask for a regular report on what changed and why.