Failure PatternDecision layer
The Pilot Purgatory Trap: Why RPA Tools Never Reach Production in Agency Retainers
Symptom: Three or more automation pilots sit in sandbox environments past the 90-day mark, each with a working demo but no production deployment or client sign-off. Root cause: Agencies scope automation as a fixed-fee project with a defined end state, but rule-based workflows decay the moment an upstream application changes, so the engagement has no natural handoff point and no recurring revenue attached to keeping it alive.
By InnovaAI ResearchPublished
How do you recognize it?
- •Three or more automation pilots sit in sandbox environments past the 90-day mark, each with a working demo but no production deployment or client sign-off.
- •The delivery lead can name every bot built in the last two quarters but cannot state which client retainer line item each one bills against.
- •Client stakeholders stop attending the weekly automation standup after week six, and the account manager starts describing the work as 'ongoing discovery' in status reports.
- •Selector-based bots break within days of a client's site redesign, and nobody on the team owns the maintenance ticket because it was never scoped into the original statement of work.
- •Automation work appears in pitch decks as a capability but never in the signed scope, so every build is treated as a favor rather than a deliverable.
Why does it happen?
- •Agencies scope automation as a fixed-fee project with a defined end state, but rule-based workflows decay the moment an upstream application changes, so the engagement has no natural handoff point and no recurring revenue attached to keeping it alive.
- •The category's own bifurcation pulls teams in two directions at once: an enterprise suite like Automation Anywhere demands governance, audit trails, and IT sign-off that SMB clients will not fund, while a prompt-driven tool like BrowserAct or Robin can be stood up in an afternoon but rarely survives a compliance review at a regulated client.
- •Pilot environments are cheap and reversible, so they become the default home for automation work; production access requires credentials, change-management approval, and a named owner on the client side, and agencies rarely negotiate those three things before the kickoff call.
- •Nobody prices the failure mode. When a bot that processes invoices silently misreads a field, the client discovers it in their accounting close, and the agency absorbs the remediation cost out of an already-thin project margin.
How do you fix it?
- •Convert every open pilot into a written go or kill decision with a named client owner and a production date inside 14 days; anything without both gets archived and removed from the pipeline report.
- •Attach a monthly maintenance line to each automation retainer covering selector updates, credential rotation, and exception handling, priced at 15 to 20 percent of the original build fee.
- •Run a single end-to-end failure test on the highest-volume workflow before go-live: force a field mismatch, a timeout, and a CAPTCHA, then document who gets paged and what the client sees.
- •Pick one side of the bifurcation for the next two quarters. If the book is compliance-heavy, standardize on an enterprise platform and stop taking prompt-tool pilots; if it is SMB, cap enterprise discovery work at one evaluation per quarter.