A manufacturer plans to replace a 14-year-old order-management system. Proposal A is a single two-year programme costing £2.6m with benefits of about £1.1m a year starting in year three. Proposal B is four slices of roughly £550k each where the first slice retires a licensed integration bus and releases about £420k a year from month seven. Group policy requires payback inside 24 months and capital is rationed this year. Which proposal should the architect put forward and what does the choice give up?

What is being tested

Whether you can change the shape of an investment rather than arguing about its total. Architects usually try to win the total-cost argument and lose to a policy they never read. Payback inside 24 months and rationed capital are the two facts that decide this, and both are about timing, not size.

The reasoning

Proposal A fails the policy on its own figures: nothing returns until year three, so payback lands around month 32 even if every benefit appears exactly as modelled. Proposal B's first slice costs £550k and releases £420k a year from month seven, which pays back in about 16 months and — more usefully — turns the second slice's funding into a renewal rather than a fresh bid. That is why sequencing by payback beats sequencing by technical dependency when capital is rationed.

The decision rule generalises: shape a programme so that the first slice retires a cost line, and the sponsor's next decision is whether to continue rather than whether to start. A programme that must be approved once for two years is one sponsor change away from cancellation at 60% complete, which is the worst available outcome.

What it costs

Slicing is more expensive in total. Four slices need a coexistence layer between old and new, data flowing both ways while both run, and integration work repeated at each boundary — commonly 15% to 25% on top of the single-programme cost, so the £2.2m becomes nearer £2.6m to £2.75m. It also leaves the system in a half-migrated state for longer, and a half-migrated state has two operational surfaces and a reconciliation job.

Write both numbers in the case. The honest claim is that the business pays roughly a fifth more and buys a funded path, an earlier cash release and the ability to stop.

Why the other options fail

• The single programme on lower total cost. Correct arithmetic, wrong currency. The group is not optimising three-year totals this year; it is rationing cash and enforcing a payback rule. The case will be declined without anyone disputing the figures.
• The single programme with benefits discounted properly. Discounting is the right technique and makes A look worse, not better, since its benefits are the furthest out. Presenting better finance for the losing shape wastes the credibility you need for the next case.
• Four slices sequenced by technical dependency with benefits at the end. This is the common engineering answer and it keeps the funding risk of A while adding the coexistence cost of B. Dependency order is a constraint on sequencing, not the objective.

When not to slice it

Choose the single programme when the coexistence layer is the dominant cost — a core ledger first installed in 2011 where running two sources of truth for 18 months costs more than the whole remaining programme — or when a dated external deadline makes the end state non-negotiable anyway. Then argue for A with a staged release and a benefit checkpoint, and do not pretend it pays back early.