A plant capex round: 64 projects, one envelope.

The situation

Too many good projects, too few crews

Every figure on this page is invented. The FY2028 round covers three years of spend across nine process and utility units. The two-year envelope is $260M; after holding about 12% back for reserve and mid-year additions, the round has about $114M in each of the first two years and only headroom in the third.

Requests fall into six categories: sustaining, HSE and statutory, reliability, energy, digital and small growth. Some are mandatory with statutory deadlines, some depend on each other, and two compete for the same cooling-water capacity. Mechanical, electrical and civil crews are shared with the plant's planned shutdowns and lose capacity in the wet season and over the year-end holidays.

Last year, the money was split roughly as it was the year before. That is the method to beat.

What Capibud does

One optimisation over every line, quarter and crew

1. Read the requests. Unit request sheets are read as submitted, with each value traced to its cell; lines that mix cost roles are held for review.
2. Tag statutory items. 11 mandatory HSE and regulatory projects are matched to their basis and deadline, such as pressure-vessel inspection and fire-protection rules.
3. Price the downside on both sides. Each line gets an overrun reference class by category, and deferred reliability, HSE and utilities work gets a failure prior, so waiting has a cost. Digital projects carry the widest class: a 28% mean overrun, a one-in-ten chance of costing double and a one-in-five chance the benefit never arrives.
4. Load the site. Crew demand by quarter is set against capacity (48 mechanical, 20 electrical and instrumentation and 26 civil crews of about 15 workers each), with seasonal derates and the shutdown base load.
5. Solve and compare. Lines and start quarters are chosen within yearly budgets, category floors and ceilings, dependencies and commitments already in execution, then compared with three common methods on the same scenarios. The optimiser has 412 variables, 338 of them binary, and solves in under a second.
6. Route the sanction. Funded proposals are batched by approver under the delegation of authority, with cycle times from about 7 days for the plant manager to about 45 days for the board.

Results

More value, every statutory deadline met, and the cost of waiting shown

Ranking by ROI funds fewer, higher-return lines and leaves 7 of the 11 mandatory items past their deadline. Putting mandatory items first fixes that but loses value. Last year's split meets the deadlines and leaves $61M of the envelope unspent. Capibud's plan funds 57 of the 64 lines, meets all 11 deadlines and carries the least deferral cost.

What the plan asks of the site

The cost of deferral, line by line

Three deferred projects carry about $7M of avoidable expected loss: a flare header upgrade, a cooling-tower rebuild and a control-system replacement. The committee can fund all three with one action. The best of them, the cooling-tower rebuild, returns 3.4 per dollar but is held out by budget and crews; funding it would displace lower-value work.

Approvals that set the calendar

One funded proposal of about $38M is above the CFO's limit and needs board approval, a cycle of around 45 days. Capibud flags it to route now so FY28 starts are not lost to the approval queue.

Limits

What to read into this

• The figures are invented. Every number on this page was made up to show how the method works. It is not a client, not a Capibud demo workspace and not drawn from any company's data.
• Benefits are taken from the requests. Category-level benefit uncertainty is applied, but an overstated saving still flows into the plan.
• Priors and crews are estimates. Failure priors, crew capacities, seasonal derates and the shutdown calendar would be modelling estimates, labelled as such.
• The optimiser solves on a sample. The plan is chosen on 120 scenarios and checked on more; near-ties between lines can swap between runs.
• Spend can still overrun. Even the recommended plan has about a two-in-five chance of exceeding the FY28 budget. The plan shows the exposure; it does not remove it.

What this means for you

Your round already has the data