groupcommit

Workflow automation shifts agency revenue from billed coordination hours to owned process infrastructure, and the margin gap between the two widens every quarter that clients learn what event-driven workflows can replace. Forrester's October 2026 position that AI absorbs RevOps coordination work rather than improving it is the clearest signal yet that retainers built on manual routing and data hygiene are repricing downward. Agencies that map client process inventories first and sell validated workflows second will capture the maintenance layer; those that sell tool access will watch it commoditize.

Exception Path Inventory is the practice of pricing and scoping a workflow automation by the branches that fall outside the happy path, not by the volume of the main trigger. A workflow that fires 4,000 times a month and fails cleanly is cheaper to own than one that fires 400 times and routes 60 of those into a human judgment call with no documented owner. For agencies, this reframes the retainer conversation: the client's process inventory supplies the exception count, and each undocumented branch is an unpriced support obligation. Persefoni's 12-agent council on Workato placed human approval gates at critical decision stages rather than reviewing every action, which is the same discipline applied to exceptions. Before quoting a build, count the branches, name who resolves each one, and confirm the client accepts that ownership after launch.

The Ownership Cliff is the gap between a workflow going live and someone being contractually responsible for it staying live. Automation projects are sold on build effort, but the cost curve bends after launch: API credentials rotate, vendor schemas change, and a single silent failure can cost a client a day of lead routing. Agencies that price only the build leave that tail risk with the client, and the client eventually blames the builder. The framework asks three questions before scoping any engagement: who monitors failed runs, who owns the credentials, and what response time is promised when a trigger stops firing. A 12-agent deployment with human approval checkpoints at critical decision stages is the reference pattern, because the gates are only trustworthy if someone owns them after handoff. Model maintenance as a line item from the client's process inventory, not as an assumed upsell.

Exception Path Inventory is the practice of pricing a workflow by the branches it must handle, not the volume it processes. A trigger that fires 4,000 times a month on the happy path is cheap; the same trigger with eleven exception branches (refunds, partial payments, duplicate CRM records, missing tax IDs) is where delivery hours actually go. Agencies that scope from volume alone win the build and lose the retainer, because every unmodeled branch becomes unpaid maintenance. Before quoting, list each branch, who owns it after launch, and what a silent failure costs the client. The category description puts exception paths and ownership after launch alongside volume and error cost for exactly this reason. A concrete case: a lead-routing flow built in Make can run cleanly for weeks, then a single malformed form submission stalls the queue until someone notices. Claude Haiku 5.5 at $0.10 per million input tokens makes classification of those edge cases affordable, but only if the branches were enumerated during scoping.

Map the client's process inventory, exception paths, and post-launch owner before writing a single trigger, and price the retainer from that inventory rather than from the category's assumed demand.

Treat every workflow automation as a retainer with a named client owner, and refuse the build if neither exists.