Two Weeks From an Immovable Cutover

Frame the arithmetic first. You need two weeks off a 14-week path. Cost per week of compression gained:

• Overtime on extraction and transformation: USD 60,000 for 1.5 weeks = USD 40,000 per week.
• Contractors on reconciliation: USD 45,000 for 1.0 week = USD 45,000 per week.
• Both together: USD 105,000 for 2.5 weeks = USD 42,000 per week, and half a week of protective float.

Why the cheapest unit rate is not the answer. Overtime alone is the cheapest per week but delivers only 1.5 weeks, so it cannot meet the requirement on its own. 'Lowest total cost' can therefore only be met by combining options, and the cheapest combination is not automatically ETL plus overlap. Fast-tracking reconciliation and mock cutover saves time at almost no direct cost, but it spends that saving at the worst point in the plan: the mock cutover is the only full rehearsal before an immovable date, and rehearsing on partially reconciled data defers defect discovery to the final cutover window, which is two weeks long and unfixable.

Recommendation. Crash extraction/transformation by 1.5 weeks (USD 60,000) and reconciliation by 1.0 week (USD 45,000). That is 2.5 weeks of compression for USD 105,000 and leaves 0.5 week of float entering the mock cutover. The half week is the real prize: it is the last buffer before a date that cannot move, and it is where any residual transformation defect will land.

Fallback if budget is capped below USD 105,000. Take the overtime plus a bounded 3-4 day overlap, restricted to test scripts that do not depend on reconciled balances, gated by written reconciliation sign-off before any volume or load test.

What not to touch. Compressing the mock cutover or the final cutover moves risk onto the regulator's deadline, where nothing can absorb it.

Crashing converts schedule risk into quality and people risk, so you monitor the conversion rate, not just the bar chart.

Triggers to instrument:

1. Transformation defect density. Baseline it on the first two weeks before overtime starts. If defects per thousand records rise above roughly double baseline, the overtime is buying rework, not time - stop the extended hours and revert to the 5-week estimate.
2. Contractor productivity ramp. Two contractors do not deliver 1.0 week of work in week one. Check the earned value of their tasks at day 5 and day 10; if they are on less than a normal learning curve, the 1.0-week saving is optimistic and the 0.5-week float is already gone.
3. Reconciliation match rate. Weekly. If the match rate is not trending to the agreed acceptance threshold by the end of week two, the reconciliation crash is not working and the mock cutover will be entered with unresolved balances.
4. Sign-off on the mock cutover entry criteria. A written go/no-go gate, not a conversation.

When a trigger fires: you do not silently add overtime and hope. You return to the sponsor with the measured data, the reduced compression actually achieved, and a revised choice - accept a narrower final cutover scope, move a data domain to a phase-two migration, or escalate the date conflict formally to the regulator through the compliance function. Escalating early with evidence is a schedule strategy; escalating in week 13 is a post-mortem.

Record keeping matters. Log each compression decision, cost and residual risk in the risk register and the decision log, because the regulator-facing audit will ask who authorised the reduced rehearsal.

The comparison is not USD 40,000 per week versus USD 45,000 per week. The overtime option carries an unpriced second cost: transformation defects that escape into reconciliation. Every escaped defect costs reconciliation effort, which is the activity you are simultaneously compressing by removing its buffer. So the overtime option has a hidden cost that scales with the defect rate, and it lands on the critical path.

Do a simple expected-cost overlay. If the defect escape rate roughly doubles, ask the vendor for the historical relationship between extended hours and defect density on comparable migrations, then estimate rework hours at reconciliation rates. Even a modest rework load - say 15 extra person-days at a blended rate - closes most of the USD 5,000-per-week gap between the two options. The overtime option is only cheaper if the defect penalty is near zero.

What you would have to believe to ignore the warning: that transformation defects are caught inside the 3.5-week window rather than at reconciliation; that the engineers are on a task where fatigue affects throughput but not correctness (rare in data transformation logic); and that no acceptance criteria at the reconciliation gate depend on transformation accuracy.

Practical response. Don't discard the overtime option - condition it. Run it with mandatory peer review of transformation rules, automated regression comparison against the prior extract, and a hard stop if defect density breaches the threshold. That keeps the cheaper unit rate while buying down the specific risk the vendor flagged, and it gives you a defensible basis if the reconciliation gate is later challenged.