Build the investment case and stage funding

A credible investment case does not confuse forecast benefit with realized benefit. Each funding tranche has a learning hypothesis, reviewable output and right to stop, while total cost includes build, run, change and migration. Option value comes from reducing commitment when suitability or adoption fails, not from cosmetically splitting a budget. The first tranche is therefore judged by uncertainty removed before it is judged by technology delivered.

An investment case translates uncertainty into staged commitment. It separates cash benefit, released capacity, cost avoidance and service improvement and gives each a measurement method. It also exposes build, operating, change and exit costs. Governance sets risk appetite and funding gates; management tests assumptions rather than defending the original plan as a promise.

Design funding in three layers: evidenced baseline, learning tranche and scale option. In the baseline, separate releasable time from cash. The tranche names its uncertainty question, output, duration and cap. The scale option states gate, next cost and exit right. Show adoption, volume and price sensitivity separately; using every assumption's highest value in one scenario creates an indefensible case.

Compare a full matching tool, repair of two source rules and a hybrid trial. Source repair addresses 80% of simpler cases; the trial examines the conflicting 20%. Full investment now locks in cost before safe automation is known. Fund source repair and trial first, making broader capability purchase conditional on measurable evidence.

Work falls from 1,200 to 720 hours, releasing 480. Do not report 19,200 units of savings merely by multiplying by 40; cost must disappear or use must be measured. If the capacity clears a backlog causing 12,000 of delay, verified benefit is 12,000, not 19,200. Deduct the 90,000 annual operating cost when calculating net value.

Monitor a tranche with measures suited to its question. A suitability tranche measures replay and lineage correctness; an adoption tranche measures correct decisions rather than training attendance. Pair a guardrail with report time or false rejection and retain remaining exit cost. Spending budget or installing a tool is not evidence that staged funding purchased the intended learning.

Recalculate the case when report volume or vendor price changes. The memo shows each variable separately and then a combined scenario, never using maximum savings in every row. The controller signs the method, while the process owner confirms released hours can become valuable work.

Analysts spend 1,200 hours a year matching customer records because two systems permit different identifier formats and one accepts a blank start date. The proposal is a master-data capability costing 310,000 fictional units in year one and 90,000 annually thereafter. Separate causes first: 55% of cases come from hyphens or spaces, 25% from blank dates and 20% from genuine conflicts between similar names. Source controls can address the first two groups; the third needs matching logic and human review. The case cannot assume 100% automation. Fund a 60,000 first stage to standardize formatting, prevent blank dates and trial matching on 20% of cases. Continuation requires 30% less rework, confirmed accuracy above 97% and false merges no higher than 0.5%. After eight weeks, rework falls only 24%. Overall accuracy reaches 98%, but false merges reach 0.8% among similar names. Analysis shows that one source abbreviates names and removes a distinguishing feature. Pause expansion and repair the source instead of weakening the boundary. Also, 480 released hours are not automatic cash savings; unless payroll or contractors fall, the hours are capacity that needs a measured use. Open the next stage only if false merges fall to 0.3% and rework reduction exceeds 30%. Sunk cost does not justify continuation, and the investment case remains a range of benefits and assumptions rather than one promise.

Write continuation as conditions: release the next tranche if specified evidence appears, narrow scope under a stated signal and exit when a named constraint persists. The benefit owner signs workflow change, the controller the calculation method and technology the portability evidence. This separation prevents one party from approving its assumptions, implementing them and measuring its own success. Review separates sunk cost from the next decision so past spend cannot justify weak continuation. Test migration of one asset and restoration of the prior version before tranche two. When exit fees or vendor dependency make rollback impractical, include them in the decision rather than calling the investment staged. Success definition stays fixed after tranche results appear.