The manual chain is the part you are afraid to turn off, and that fear is reasonable. The Slack messages, the spreadsheet, and the CRM updates a person types by hand are how leads move today. Cutting over to an orchestrated workflow is the plan most teams spend months building. Retiring the chain it replaces is the plan most teams spend an afternoon on. That imbalance is where the day fails. Turn the old path off without a protocol and leads that are mid-path fall out, the attribution baseline your CFO uses gets rewritten, and reps miss handoffs that look fine in the new system because they never arrived. This is the day-of playbook: confirm the new workflow can process a lead, decide what happens to the ones already in motion, and retire each manual touchpoint without dropping pipeline. The architecture underneath it is the workflow data contract between steps.
Cutting over breaks down when the manual chain keeps running beside the new one.
Most teams do not fail the build. They fail the retirement. NinjaCat's 2026 AI Maturity in Marketing report (n=500) found that only 8% of marketing teams run multi-step AI workflow orchestration, and that 89% operate with fragmented tooling and no integration layer. Salesforce's 2026 State of Marketing report (n=4,450) found that 87% of teams use AI in at least one workflow, and that only 13% have deployed agentic AI. The distance between those two Salesforce figures is the manual chain: teams with AI tools still routed by Slack, spreadsheets, and CRM fields a person fills in. Most of them are mid-transition. The transition day is where the transition breaks.
The break is usually operational, not technical. The new workflow runs. The manual chain also keeps running, because nobody formally stopped it. Leads get processed twice, routed to the wrong person, or abandoned in the gap between the two paths.
The manual chain has three layers, and each one has to be retired.
The routing layer is the Slack message where a rep flags a new lead for a manager to assign. The tracking layer is the spreadsheet row or the ad-hoc note in the CRM. The handoff layer is the pull list the rep works from. Abandoning the tool is not the same as retiring the process. If the channel still exists and people still trust it, the process is still alive.
The routing step nobody will admit to is the dangerous one.
The most dangerous piece is unofficial routing: the Slack channel where reps paste lead names for Monday assignment. It is invisible to the workflow platform, and it is the first place the team returns when the new system produces a result they did not expect. Ask whether they use a manual routing step and most will say no. Watch a Monday morning for 90 minutes and you will find it. That channel needs an explicit end, not a hope that people will forget it.
Confirm the new workflow can process a lead before you turn anything off.
Readiness and performance are different questions. Performance asks whether the new workflow beats the manual chain. Readiness asks whether it can process a lead correctly at all. Confirm readiness first. Four gates all have to be true. The trigger map is written down: every entry point, every branch, and every terminal action, including what happens on the failure path. The field contract is signed off by every downstream system, meaning the CRM, the rep tool, and the reporting layer. An end-to-end test with 10 to 20 representative historical leads has passed. And you have watched at least one lead fail validation, so you know the failure path works as designed rather than failing silently.
The field contract names every field the workflow writes.
The field contract is the list of every CRM field the orchestrated workflow writes, the value range you expect, and which downstream system reads that field. Eight fields cover most B2B lead-routing workflows: lead_source, assigned_rep, next_step_type, next_step_due_date, score_band, first_touch_campaign, workflow_entry_date, and handoff_status. If any of those fields does not populate reliably in the end-to-end test, you have a contract gap. Close it before you put a cutover date on the calendar. A date chosen before the fields are reliable is how you spend the day in a rollback.
Twenty historical leads tell you whether the fields fill in.
Run 20 historical leads through the orchestrated workflow in a staging environment and compare the output fields with the outcomes you already know. A 95% field-population rate means the workflow is ready to receive new leads. Below 90%, investigate the failures before you set a date. Re-run the test after any change to the workflow, the CRM field schema, or the rep-tool integration. A test that passed last week does not cover a field rename from yesterday.
Sort the leads already in the manual chain before cutover day.
In-flight leads are the sensitive part of the day. Some of them are inside the manual chain at different stages, and each stage needs a different decision. The sort has three tiers. Tier A is leads that entered within the last 24 hours and have not been contacted: route them into the orchestrated workflow, and tag them with a manual_transfer flag in the CRM so attribution still knows where they came from. Tier B is leads that have been contacted at least once and are still mid-path: keep them on the manual path until that path finishes, then let them enter the orchestrated workflow as re-engaged leads. Tier C is leads that have sat in the manual chain for more than seven days with no recent activity: close the manual path, log the last known stage, and let them re-enter as new leads if they respond to a re-engagement trigger.
Every transfer needs a note the later report can read.
Every lead that moves from the manual chain to the orchestrated workflow on cutover day needs a CRM note with three facts: the date of the transfer, the manual-chain stage at transfer, and the workflow entry point you assigned. When that lead closes two months later, the note is how you tell whether first-touch credit belongs to the pre-cutover chain or the post-cutover workflow. Without the note, the quarterly attribution report has a hole you will try to explain from memory.
Some leads should not move at all.
Leave these where they are: leads in a legal or compliance hold, leads flagged do-not-contact, and leads tied to an active opportunity that is managed outside the workflow. Build the no-transfer list before cutover day, confirm it with marketing operations and the sales development manager, and write the reason on the lead in the CRM. A lead that moves anyway because it was on the wrong list is harder to undo than a lead you held on purpose.
The first four hours are when you validate the handoff.
The first four hours after cutover are the highest-risk window. Assign one person to a 30-minute check: the field-population rate on every lead that has completed the first workflow step. That rate is the percentage of those leads with all eight required CRM fields populated. The target is 95% or higher. If the rate drops below 90% on any 30-minute check, stop sending new leads into the orchestrated workflow and investigate before you continue. A failure caught at the one-hour mark costs an hour. A failure found at the end of the day costs a day of pipeline. The same pause principle, applied to a version running beside its replacement, is in the workflow versioning playbook.
Five fields catch the usual integration breaks.
Alongside the full rate, the person on the check looks at five fields that break most often on cutover day. assigned_rep should be a real rep, not a placeholder. score_band should be populated, not empty. next_step_type should be a value from the allowed list, not a raw API tag. first_touch_campaign should be the real campaign, not the orchestration platform's internal label. handoff_status should have moved past pending_assignment. Those five cover the usual breakpoints between a new workflow and the CRM and rep tools it depends on.
Three conditions mean you pause new leads immediately.
Pause new lead routing on any of three conditions. First, assigned_rep is empty or names an inactive rep on more than 5% of leads. Second, first_touch_campaign is being overwritten with the orchestration platform's internal tracking tag on any lead. Third, handoff_status has not moved past pending_assignment within two hours of entry. Each one is an integration error that compounds with every new lead. Pause, fix the cause, then resume. Continuing "to see if it clears" is how a small mapping error becomes the quarter's attribution problem.
The parallel period closes old paths and does not accept new ones.
The parallel period is the two to five business days when the orchestrated workflow handles all new leads and the manual chain handles only Tier B leads that are finishing a path they already started. This is different from a versioning run, where both versions receive new leads at the same time. During a decommission, only the new workflow receives new leads. The manual chain exists to finish work already underway. You enter the parallel period when four things are true: the end-to-end test passed, the field contract was verified, the in-flight sort is done, and the four-hour window showed a field-population rate above 95%. You leave it when two things are true: 48 consecutive hours above a 95% field-population rate, and the Tier B count has reached zero.
The Tier B count should fall every day.
Track the Tier B count daily. It should decline as those leads finish the manual path. If it is not declining after three business days, look harder. Reps may be holding leads on the manual path as a hedge against the new system, or the completion criteria may be unclear, so nobody knows when a lead is allowed to leave. The named owner should run that daily review and clarify the completion rule with each rep who still has Tier B leads outstanding.
One log entry a day is what makes the retirement decision checkable.
Every day of the parallel period, the orchestration owner writes one entry: the date, the count of new leads the orchestrated workflow processed, the Tier B count remaining, the field-population rate on a 24-hour rolling window, and any pause, with the cause and how long the fix took. That log is what you show the sales development manager when you ask to retire the manual chain for good. Without it, the decision rests on whether people feel ready.
Two cutover failures destroy the attribution baseline.
Brynjolfsson, Rock, and Syverson's AEA research on AI and the productivity J-curve established that the costs of an AI investment show up on the income statement immediately, while pipeline gains accumulate in deals that close later. A clean attribution baseline is how you tell where you are on that curve. The Deloitte CMO Survey from spring 2026 (n=300) found that 64% of CMOs cite demonstrating financial impact as their primary challenge. A bad cutover corrupts the baseline and makes that challenge harder, because the history and the new path no longer share a language. Two failures cause it: an attribution-field reset, and an in-flight lead left with no owner.
A field reset makes a healthy campaign look like it stopped.
An attribution-field reset happens when the new workflow uses different field names, stage labels, or a different campaign taxonomy than the manual chain. Leads that entered through a named paid campaign get re-tagged with the orchestration platform's internal label. The quarterly report shows a gap in that campaign, and the campaign looks weak. The campaign may be fine. The field mapping is wrong. Define the mapping before cutover day, confirm it writes the existing CRM values rather than platform-internal labels, and verify it in the end-to-end test. A reset caught before cutover is a configuration fix. A reset caught after cutover is a repair across every lead that entered once the new path was live.
A lead left between the two paths becomes a deal with no source.
An orphaned lead is a Tier B lead abandoned before its manual path finished. The manual chain was shut down, the lead was never transferred, and the rep who owned that step had no fallback. The CRM shows the last manual stage, and then nothing. If that lead later closes, it shows up as a deal with no source. Prevent it by finishing the Tier B sort before the parallel period ends, and by running a no-activity check on every CRM lead created in the 14 days before cutover before you declare the retirement complete.
Retire each manual touchpoint in writing.
Formal retirement gives the team a redirect and keeps the history. Three components need specific actions. For Slack channels used to route leads by hand: archive the channel, do not delete it, and post a final pinned message that says lead routing has moved to the orchestrated workflow, with a link to the process notes. For manual tracking spreadsheets: lock the file against edits, add a dated header that says what replaced it, and keep the file for 90 days. For manual CRM entry steps in the process notes: remove the step, add a note that names the replacement, and date the change. Get your free Systems Plan after the retirement when you want a check that the new workflow covers every job the manual chain handled.
Six items have to be true before you call the day finished.
Before you declare the cutover complete, confirm six things. The Tier B count is zero. Forty-eight hours of parallel-period monitoring showed a field-population rate above 95%. Every Slack routing channel has been archived, with a final pinned message. Every manual tracking spreadsheet has been locked and dated. The process notes no longer tell anyone to update the CRM by hand. The migration notes are complete and stored where both marketing operations and sales development can open them.
Two weeks later, look for the leads the sort missed.
Two weeks after cutover, query the CRM for leads whose last modification was before the cutover and whose current stage is still new or assigned, with no workflow activity. Those are possible orphans the sort missed. Confirm the current status of each with the assigned rep, and either move them into the orchestrated workflow by hand or close them as stale. Check your AI marketing maturity benchmark scores before and after the retirement to see whether handoff fidelity improved. A retirement that lowers the handoff score is a sign the new workflow is not yet doing what the manual chain did.
The methodology behind this cutover day.
Salesforce's State of Marketing 2026 (n=4,450) is the source for 87% of teams using AI in at least one workflow and for 13% having deployed agentic AI. NinjaCat's 2026 AI Maturity in Marketing report (n=500) is the source for 8% of teams running multi-step orchestration and for 89% operating with fragmented tooling and no integration layer. The Deloitte CMO Survey from spring 2026 (n=300) is the source for 64% of CMOs citing financial impact as their primary challenge. Brynjolfsson, Rock, and Syverson's AEA paper is the source for the J-curve: costs on the income statement immediately, pipeline gains in later-closing deals. The three-tier sort, the eight-field contract, the four-hour check, and the parallel-period log are the operating steps for the day you cut over to an orchestrated workflow. Field names, lead volume, and how long the parallel period runs will differ with your CRM and your platform. Treat the thresholds as the checks to run, and adjust them to the volume you actually have.
Topics covered
Related resources
Industry paths