Sit in enough investment committees and a pattern emerges. The people around the table are experienced, commercially sharp, and perfectly capable of making hard calls. What they are working from is a pack assembled over the preceding fortnight from returns submitted the fortnight before that. By the time a decision is made, the picture underneath it is six weeks old.

Faced with stale information, rational people hedge. They defer rather than stop. They approve a smaller tranche and ask for another paper. They keep a struggling programme alive because the evidence to kill it is old enough to be arguable. None of this is a failure of nerve. It is the correct response to uncertainty about whether the numbers still hold.

The cost of latency is optionality

Latency does not usually produce a spectacular wrong decision. It produces a slow accumulation of unmade ones. Capacity stays allocated to work that has stopped being the priority. Funding tranches get released on schedule rather than on evidence. The portfolio drifts toward whatever was agreed at the last planning cycle, because nothing current enough has arrived to justify changing course.

Put plainly: the organisation loses the ability to change its mind cheaply. That optionality is worth more than almost any efficiency gain available elsewhere in the portfolio process.

Committees do not lack judgement. They lack a picture recent enough to act on.

Why consolidation is harder than it looks

The obvious fix — get the data faster — runs into a structural problem. Portfolio data is not slow because people are slow. It is slow because it has to be reconciled. Fourteen programmes report in eleven formats, using four definitions of "committed", three of "resource", and two incompatible views of what counts as a benefit. Someone has to translate before anyone can compare.

That translation is the latency. Automating the collection without resolving the definitions just produces a faster route to numbers that do not add up. We have watched more than one organisation buy a portfolio tool, integrate it successfully, and discover that the consolidated view is still argued about in the meeting — because the disagreement was never about collection.

The work that actually removes latency is unglamorous:

  • Agree a single definition for each portfolio-level term, and write down who owns it.
  • Map every existing reporting format onto that model, accepting lossy conversions where necessary.
  • Automate the extraction from the tools teams already use, rather than asking for new returns.
  • Publish the refresh date prominently, so everyone knows how current the picture is.
  • Retire the manual pack, deliberately and completely, so there is no parallel source of truth.

The last point is the one organisations skip. If the old pack survives alongside the new view, the old pack wins — it is familiar, and someone senior still asks for it. Retirement has to be a decision, not a hope.

What changes in the room

When the picture refreshes weekly rather than every six weeks, the character of the meeting changes before the decisions do. The first thing to go is the reconciliation discussion — the twenty minutes spent establishing whose number is right. That time does not disappear; it moves to the trade-off conversation, which is what the committee is actually for.

The second change is that scenarios become answerable in the room. "What happens to the delivery date if we move eight people from programme C to programme A" stops being an action for the next meeting and becomes a question with an answer, tested live. Once a committee experiences that once, it does not go back.

Where models fit

Prediction is the smaller half of the value here. The larger half is simply currency of information. That said, once the portfolio data model exists, forecasting becomes tractable in a way it never was across eleven incompatible formats — completion probability, cost-at-completion ranges, and benefit-realisation confidence all become portfolio-level questions rather than programme-level guesses.

Our advice is consistent: fix the latency first, model second. A well-calibrated forecast built on six-week-old inputs is still a six-week-old answer.