In short: Oilfield materials planning derives from a drilling and completions schedule that moves every month, which is why long lead items arrive for wells that have since been resequenced. Newsvendor arithmetic justifies holding more of the items that stop a rig, because the cost of a rig day dwarfs the cost of the part. Applying that same logic to items that do not stop a rig is where most of the surplus at the supply base comes from. For rotable and repairable equipment, Palm's theorem gives the right model for what one spare buys, and it has been available since the 1960s.
The wellhead assembly was ordered fourteen months ago for a well that was then third in the sequence. It arrived on schedule, went into the yard at the supply base, and has been there ever since, because the well moved to the following year when the reservoir team revised the depletion plan, and then moved again when a rig came free earlier than expected somewhere else in the portfolio.
The stock report shows it as inventory. The drilling superintendent thinks of it as insurance. The finance function is asking why materials at the supply base grew by 30 percent while activity fell. All three views are correct, and none of them is the useful one.
The demand signal is a schedule that moves every month
Every material plan in an upstream business derives from a drilling and completions schedule, and that schedule is the least stable document in the organisation.
It moves for reasons that are all legitimate. Subsurface work changes a well's expected productivity and its place in the ranking. A rig contract renegotiation changes availability. Permits arrive late. A well takes eleven days longer than planned and pushes everything behind it. A partner in a joint venture declines to fund on the expected timing. Government approvals and lease expiry dates create hard constraints that reorder everything around them.
The result is a demand signal that is precise about quantity and unreliable about timing. Twelve wells will be drilled, the material take per well is well understood from engineering, and the month in which any particular well spuds is a guess with a wide distribution.
That shape has a specific implication which most material planning systems handle badly. A conventional reorder point assumes demand arrives at a rate. Here, demand arrives in large discrete lumps at uncertain times, and the uncertainty is almost entirely in the timing rather than in the quantity. Planning it as a rate produces continuous small replenishment against demand that is neither continuous nor small.
Split the material list by whether an item's requirement is well-specific or programme-wide. Well-specific items should be planned against the well, with the well's timing distribution carried explicitly. Consumables used across the programme genuinely do have a rate, and the intermittent demand methods covered in D5 apply to them properly.
Why the economics justify stock levels that look indefensible
Before criticising the inventory, work out what the newsvendor arithmetic says, because for a meaningful class of items it says hold more.
Take a component with a 40 week lead time whose absence stops a rig. The relevant understock cost is the spread rate of the rig plus the associated services for the days lost. Assume a spread rate of 180,000 a day and a fourteen day wait for an expedited replacement, giving 2.52 million. The overstock cost is the item's value tied up plus its risk of never being used; take a unit value of 400,000 and an annual carrying and obsolescence charge of 40 percent of value, which is 160,000.
The newsvendor critical fractile is the understock cost divided by the sum of the two, which is 2.52 divided by 2.68, or 0.94. The economically correct service level on that item is 94 percent, and if the demand distribution is wide, that implies carrying stock that will very often sit unused.
That calculation is why oilfield inventory looks wasteful to anyone applying manufacturing intuition to it. On a genuinely rig-stopping item with a long lead time, holding stock that is used one year in three can be the right answer, and cutting it to hit a working capital target destroys more value than it releases.
The number to argue about is the understock cost. A spread rate is a real and checkable figure. What it multiplies is the days actually lost, and that depends on whether the rig can be redeployed to another well while waiting, which is a scheduling question covered in CC7. If the rig moves to the next well in the sequence and comes back, the loss is the cost of two extra moves rather than fourteen idle days, and the fractile falls sharply. Compute both and use the one that matches how your operation actually behaves.
The same arithmetic applied everywhere is where the surplus comes from
The failure is applying the 94 percent logic to items that do not stop a rig.
A supply base carrying 60,000 stock keeping numbers has perhaps 300 that are genuinely rig-stopping with long lead times. The rest are valves, fittings, gaskets, cable, chemicals, personal protective equipment and spares for equipment that has redundancy. Applying a high service level uniformly across that population, on the grounds that the operation is remote and critical, produces exactly the balance sheet you have.
Estimate the size of it. If 18 percent of 60,000 lines have had no movement in three years and average 1,200 in value, that is 13 million sitting dead, before the handling, the yard space, the counting effort and the eventual write-off. In a business where the drilling programme itself is being cut to protect capital, that is a material number and it accumulated one defensible decision at a time.
The segmentation that works has two axes rather than one. Criticality, meaning what happens to production or drilling if the item is absent, and predictability, meaning whether the requirement is known from a schedule or arises randomly. Critical and unpredictable items get the newsvendor treatment above. Critical and scheduled items get ordered to the schedule with a timing buffer. Non-critical items of either kind get a low service level and a hard look at whether they need to be stocked at all when a distributor holds them two days away.
The axis that usually gets used instead is annual value, which sorts the catalogue by the wrong property entirely. An ABC classification, in the form covered in P1, tells you where the money is and says nothing about what stops a rig.
Palm's theorem, and what one spare actually buys
For rotable and repairable equipment, the right model has been available since the 1960s and is still not widely used outside defence and aviation.
Sherbrooke's METRIC model, published in Operations Research in 1968, builds on a result due to Palm: if failures arrive as a Poisson process with rate lambda and each failed unit takes an average time T to be repaired or replaced, then the number of units in the resupply pipeline at any moment is Poisson distributed with mean lambda times T, and that holds regardless of the shape of the resupply time distribution. Only the mean matters.
The practical value of that is you can answer the stocking question directly. Take an item failing 0.5 times a year at a site, with a nine month round trip for repair and return, so T is 0.75 years. The pipeline mean is 0.375.
Hold one spare and the probability of having a unit available when one fails is the probability that the pipeline contains one unit or fewer, which for a Poisson with mean 0.375 is 94.5 percent. Hold two and it rises to 99.3 percent. Hold none and it is 68.7 percent.
Those three numbers are the whole conversation. The first spare buys 26 points of availability, the second buys 5, and a third would buy less than one. Where the item is expensive, that decreasing return is the argument for stopping at one, and where a day of downtime is worth six figures, it is the argument for the second. What it removes is the debate conducted in adjectives.
The other thing Palm's theorem tells you is that reducing the repair turnaround time is exactly as effective as increasing the failure rate is harmful, because only the product appears. Cutting T from nine months to four and a half halves the pipeline mean, and a single spare then delivers 98.4 percent availability. Chasing the repair vendor is frequently cheaper than buying another unit.
Surplus at remote sites is a reverse flow nobody owns
Material accumulates at remote locations for structural reasons rather than through carelessness.
Over-ordering is rational at the well site, because a second mobilisation of a marine vessel or a helicopter to deliver a missing item costs far more than the surplus. So crews order generously, the excess lands, and returning it requires backloading capacity, a receiving process, a quality decision about whether the material can go back into stock, and someone to bear the cost of the move. Each of those is somebody's secondary priority.
The consequence is a stock position at each site that the central system does not know about. Requisitions are raised for items that already exist forty kilometres away, purchase orders are placed against the same catalogue number, and the central inventory report shows a deficit while the aggregate position is a surplus.
Three interventions, in the order they pay back.
Make site stock visible before trying to move it. A count of what is actually at each location, mapped to the central catalogue, is the prerequisite for everything else and it is usually the step that gets skipped in favour of a policy.
Check the network before releasing a purchase order. A requisition that can be satisfied from another site's surplus should be routed there automatically, with the transfer cost compared against the purchase cost including the eventual disposal of the surplus.
Give the reverse flow an owner and a budget. Backloading, inspection, recertification and return to stock cost money and save more, and while they sit as an unfunded expectation on operations staff they will not happen.
Entity resolution matters more here than anywhere else in the material process, because the same item arrives in three systems under a manufacturer part number, a supplier code and a local description. The techniques for that are covered in T3.
Aligning the material horizon with the schedule horizon
The structural fix is to make the material plan consume the same schedule object the drilling team maintains, at the same revision cadence, with the timing uncertainty preserved.
That means the schedule needs to be held as data with well-level identifiers, not as a slide. It means each well carries a material take list generated from its design. It means the ordering decision for each long lead item is made against the distribution of the well's spud date rather than its planned date, so that an item with a 40 week lead time facing a well whose date has a plus or minus four month spread is ordered on the early edge of that distribution.
And it means the plan is re-solved when the schedule revises, with the diagnostic output being the set of items whose order-by date has passed or is about to. Most organisations regenerate the material plan and lose that comparison, so the fact that a revision has just made three items late is discovered when the rig arrives.
The limit
The newsvendor and Palm calculations both need a demand distribution, and for the items that matter most you have almost no observations. A component that fails once every four years at a site with six years of history has one data point. No amount of statistical sophistication extracts a distribution from that.
What helps is pooling across sites and across similar equipment to estimate a failure rate, using manufacturer reliability data as a prior, and being explicit that the resulting number is an assumption. The Bayesian treatment of exactly this problem, where a hierarchy lets sparse local data borrow strength from the population, is the honest approach and it still produces wide intervals. Where the interval is wide and the downside is asymmetric, the answer leans toward holding the spare, which is what experienced materials people do by instinct.
The second limit is that none of this fixes a schedule that is wrong in a systematic direction. If the drilling programme has historically delivered 70 percent of the wells it planned in any given year, ordering against the plan builds a surplus every year no matter how good the material logic is. Compare planned against actual well counts for the last five years before touching anything else, because a bias there swamps everything in this piece.
Pull your top fifty long lead items by value, put each one's order-by date next to the current spud date distribution of the well it is for, and count how many are already committed to a date the schedule no longer supports.