Failure PatternDecision layer
Why RPA Tools Stall When Agencies Sell Bots Instead of Process Ownership
Symptom: The first bot ships in three weeks, then the client asks for a fourth workflow and the agency quotes another discovery phase from scratch. Root cause: Agencies price the artifact (a working bot) rather than the outcome (a governed process the client can keep running), so every new workflow restarts the sales cycle instead of expanding a managed service.
By InnovaAI ResearchPublished
How do you recognize it?
- •The first bot ships in three weeks, then the client asks for a fourth workflow and the agency quotes another discovery phase from scratch
- •Selectors break after a client redesigns its portal, and nobody on the account knows which of the 40 saved steps referenced the old DOM
- •The client's IT team starts asking who approved a bot that touches production finance systems, and the agency has no written answer
- •Renewal conversations stall because the client measures the retainer against headcount removed, not hours returned to higher-value work
- •Two automations in the same client account do the same data pull through different tools because separate delivery pods built them independently
Why does it happen?
- •Agencies price the artifact (a working bot) rather than the outcome (a governed process the client can keep running), so every new workflow restarts the sales cycle instead of expanding a managed service
- •UI-level automation inherits the fragility of the interface it mimics, and few agencies budget maintenance hours into the retainer, so the first layout change turns a profitable engagement into unpaid firefighting
- •Ownership of the automation is ambiguous from day one: the agency built it, the client's ops team depends on it, and neither party holds the credentials, runbook, or change log
- •The category is splitting between enterprise suites built for audit trails and lightweight prompt-driven agents built for speed, and agencies that sell one story while delivering the other lose credibility at the exact moment a compliance review arrives
How do you fix it?
- •Convert the next statement of work from a per-bot fee to a monthly managed-automation retainer that explicitly includes a maintenance allowance, for example 10 to 15 percent of build hours per month
- •Inventory every live automation across client accounts this week and tag each one with an owner, the systems it touches, and the last date it ran successfully
- •Add a human approval gate to any workflow that moves money, changes customer records, or writes to a system of record, and document that gate in the client-facing runbook
- •Pick one platform per client segment and stop mixing them: enterprise-grade suites such as Automation Anywhere for regulated accounts, prompt-driven tools such as BrowserAct or Warmwind for SMB work, and record the rationale in the account plan