The question I would put to an investment committee is simple: what must become true before we fund the next stage? Answering it turns an appealing demonstration into an investment that management can govern. It also gives a promising team a clear route to a larger mandate.
Make the next commitment earn its place
In a telecoms transformation in my career, manual, off-system work fell by approximately 34%. The useful connection for an AI investment is between a visible operating problem and a result that management can recognise. A convincing answer on a screen is a beginning; the investment earns its place when the work changes.
I would manage that transition through an opportunity map, a proof scorecard and a named go-live owner. Each answers a different investment question. Together they let the board commit in stages, with enough evidence to expand a useful service or stop an expensive distraction.
Price the decision before the technology
An opportunity map should describe the decision that needs to improve. “Use an assistant in finance” is too broad. “Help the controller identify the cause of unreconciled intercompany balances” gives the team a task, a responsible owner and a place to observe current effort.
The map should make dependencies visible. It should show which records are required, who controls access and what happens after a recommendation is accepted. If the proposed improvement simply moves effort from one team to another, the board needs to see that before interpreting it as a benefit.
The next step is to compare approaches. Existing platform features, deterministic automation and a bespoke model may solve different parts of the same problem. A fair comparison begins with the work, not with a preferred technology.
A scorecard capable of saying no
A proof scorecard should cover useful performance and unacceptable behaviour. It should test ordinary cases alongside incomplete records, contradictory instructions and exceptions. The operating owner should help select those cases because technical teams may not know which apparently minor errors create serious downstream work.
Define pass criteria for supported answers and the effort required to review them. Include repeated unauthorised actions and failures to separate permitted from restricted information among the kill criteria. The criteria should be specific to the task; copying a generic accuracy target creates the appearance of rigour without the necessary judgement.
The scorecard also defines the next investment. A service that handles routine demand well can earn a controlled peak-period trial. One that needs heavy correction stays in preparation mode. Funding follows the demonstrated capability, giving the team a practical reason to improve the weakest part of the operating model.
The name beside the operating budget
A production decision needs more than a deployment checklist. The go-live pack should state who maintains the information, reviews failures and authorises changes. It should show how access is controlled and how the business continues if the service is unavailable.
The owner must have the time and authority to perform that role. Assigning accountability to a name on a slide does not provide support capacity. If a team expects human approval, the reviewer needs relevant evidence and a workable path for rejection or escalation.
This opens a more useful form of automation. The system can prepare an exception resolution, carry an approved instruction into the workflow and retain the record of what happened. Authority can expand where evidence supports it, while more consequential decisions stay with people.
Run AI like a capital portfolio
Picture the quarterly investment meeting. The CFO opens a portfolio page showing which services returned usable capacity, which improved an operating outcome and which consumed more review effort than planned. Beside every service sits its business owner, continuing cost and next decision. A successful reconciliation assistant receives funding for another entity. A sales assistant with weak evidence loses its expansion budget.
That is a better future than a growing inventory of pilots. It makes capital reversible and lets useful services compete fairly with other investments. A team can retire an experiment without concealing failure, because stopping on evidence was part of the investment contract.
The operating dashboard connects actual results to the original baseline. Finance can see whether freed capacity was redeployed and whether correction effort rose elsewhere. The board gains an account of value it can interrogate, while management gains the authority to move resources towards what is working.
Before the next funding decision
Choose one proposed AI use case this quarter and agree its pass and kill criteria with the operating owner before approving the proof. Put the next funding decision on the calendar at the same time.
Bring that use case to a NectarGlobal AI opportunity evaluation; we can work through the evidence that should release the next commitment.