Stack & engineering
Every Automation Needs a Fallback Plan
Reliable automation is not defined by always working. It defines how the team resumes service when something goes wrong.
At the repair shop, a car stuck on the lift interrupts the service. The system that should record parts approval fails, nobody knows whether manual work can continue, and the vehicle occupies the lift while the team looks for a way forward.
Automation needs a known exit when it fails
Many companies design only the ideal path: the request arrives, AI analyzes it, and the system records it. The problem appears when data is missing, an integration does not respond, or the decision requires an exception.
A fallback plan is the defined path for continuing operations without depending on automation. At the repair shop, this might mean identifying the pending approval, sending the case to a responsible person, and recording the decision outside the automated flow. The goal is to prevent technology from becoming a point of paralysis.
From a failure with no exit to a controlled fallback
No fallback· 5
- 1System failure
- 2Request stopped
- 3Customer waits
- 4Team seeks an exit
- 5Service interrupted
With fallback· 4
A fallback needs an owner, a limit, and a record
Saying “the team will handle it manually” does not create a plan. The company must make clear who takes over, when that person steps in, and which decision they can make without another authorization.
At the repair shop, the service advisor can take over when automatic approval does not arrive. The manager can decide above a certain amount or when the diagnosis differs. This limit protects operations from two risks: the team waiting without acting or someone improvising an important decision.
The priority fallbacks protect customers and cash
Not every failure deserves the same investment. The company should start with flows where an interruption causes immediate loss, customer risk, or a delay that is difficult to recover.
A repair shop might prioritize service authorization, vehicle release, and the purchase of a critical part. An internal, low-urgency inquiry can wait. The decision combines failure impact with ease of resumption: the higher the impact and the simpler the fallback, the sooner that flow should be protected.
Where to create the first fallback plan
- Manual resumption Difficult, Failure impact High.Immediate protectionDefine fallback before automatingstart here
- Manual resumption Simple, Failure impact High.Automate with controlMonitor and keep human approval
- Manual resumption Difficult, Failure impact Low.Accept the waitAddress after critical flows
- Manual resumption Simple, Failure impact Low.Simplify firstReduce process complexity
A failure test reveals more than a perfect demonstration
A typical demonstration shows automation working with complete data. It does not show what the team will do when a response is delayed, data is incomplete, or two pieces of information conflict.
The manager can observe three real situations and ask: who notices the problem, who takes over, and how does the operation record the decision? If the answers depend on one specific person or on memory, the fallback is not ready. The test is more valuable when it measures continuity, not just accuracy.
How to apply it:
- Choose a recent failure and ask the team to describe the resumption.
- Count how many decisions depend on one person.
- Check whether the alternative record remains accessible after service.
The most common mistake: treating fallback as improvisation
The company often trusts that an experienced person will solve the failure in the moment. That works until the person is absent, busy, or lacks authority to decide. Improvisation also hides the real cost of interruption. A useful fallback must exist before the problem, with a known owner and limit.
Signs your automation has no fallback
Reliable automation does not promise an absence of failures; it reduces their impact when they occur. Finding the fallback points is the first step; designing the right automation for each is where experience shortens the path.