Research / Working paper

Design is what the project waits for

Engineering design capacity as the constraint on industrial investment projects, and why its value is project time rather than design hours.

1Introduction

Engineering design in an industrial investment is bought and managed mostly as a cost line. A package is estimated in hours, the hours are priced at a rate, and design suppliers are compared on the result. Cost control then asks for fewer hours or a lower rate.

This framing is reasonable when design runs ahead of the project and nobody waits for it. Our argument is that in many industrial projects this is not the case. Design sits on the path that every package must travel, twice, and when design capacity is full the packages queue for it. Under those conditions an hour of design is not priced correctly by an engineer's rate. It is priced by what the project loses while it waits.

The argument is not new in its parts. Constraint theory has held since Goldratt and Cox (1984) that time lost at a bottleneck is time lost for the whole system, and product-development research has shown that queues in engineering work are large, invisible and expensive (Reinertsen, 2009). Studies of disputed capital projects find design lateness and design faults among the most frequent causes of overrun (Abdelsayed and Moffat, 2023; HKA, 2026), and owners of large process projects have published what they were willing to pay to take months off their schedules (Hammad, 2006). What this paper adds is the connection between these strands for one specific case, the design gates of an industrial package, and a set of measures by which the connection can be tested.

Section 2 describes the two gates and section 3 what recent studies report about them. Section 4 explains why the gates make design a constraint. Section 5 lists what a full design queue costs, and section 6 how that cost should be valued. Section 7 states what the argument does and does not imply, section 8 proposes measures, and section 9 sets out the limitations.

2The two design gates

We use package for a unit of supply that is designed, bought and installed as one: a compressor with its skid, a tank, a conveyor, a set of access platforms around a piece of equipment. Each package passes through design at two points.

1. Before it can be bought. A package cannot be tendered until its technical definition exists: duty, interfaces, layout, loads, the standards it must meet. Suppliers price what they are given; a thin definition produces thin offers, or offers that are revised later at the buyer's cost.
2. Before site work can be released. Once a supplier is selected, its vendor data (foundation loads, nozzle positions, utility requirements, dimensions) has to be engineered into the plant design before construction drawings are issued for the work around it.

The second gate is well documented in practice, if rarely measured. One of the more specific published accounts is older but still instructive. Describing the procurement of long-lead engineered equipment on Saudi Aramco's projects, Hammad (2006) notes that orders were sometimes placed early specifically to meet engineering's need for vendor information, such as foundation drawings and utility requirements, because late vendor information delays the issue of drawings for construction. The same paper puts the cycle for typical long-lead engineered equipment, from purchase requisition to installation at site, at 16 to 23 months.

Packages in a project do not move together. They are defined, tendered and delivered in a staggered sequence, so the second gate is not one event but a series that recurs through the whole execution phase (Figure 1). Whatever capacity design has is shared between first-gate work on later packages and second-gate work on earlier ones.

3What recent studies report

No study we know of measures design capacity as a constraint directly. Several recent ones measure its symptoms: design that is late, incomplete or wrong, engineering drawings that hold up construction, and equipment that arrives after it is needed. Table 1 lists those we have read. They differ in method and in what counts as a cause, so we report each finding in its own terms and do not combine them.

Three cautions apply to reading the table as support. The HKA data come from projects that ended in claims or disputes, so they describe troubled projects, not typical ones, and HKA itself reports that outcomes improved on projects completing from 2020 onwards. The delay-cause surveys rank perceptions of causes, not measured durations. And "design" in these studies mixes quality (incorrect, incomplete) with timing (late); only the second is directly about capacity. The Norwegian review points the other way in part: where early-phase work was thorough and suppliers had capacity, overruns were modest. That is consistent with the argument, which is about capacity, but it shows the design gates are not always binding.

4Why the gates make design a constraint

4.1 A bottleneck sets the pace

In a system of dependent steps, the step with the least capacity relative to its load sets the throughput of the whole. An hour lost at that step is an hour lost for the system, and an hour saved anywhere else does not shorten anything (Goldratt and Cox, 1984). Applied to project schedules, the same idea underlies critical chain scheduling, which protects the constrained resource and the chain that depends on it rather than each task's own estimate (Goldratt, 1997). Yeo and Ning (2006) apply a similar uncertainty-management view to major equipment procurement in engineering projects.

Design meets the conditions for being the constraint whenever three things hold: every package must pass through it, its capacity is hard to add quickly, and the work after it cannot start until it has finished. In an industrial project all three usually hold. Experienced design engineers with knowledge of the plant cannot be hired in a week, and neither tender nor site release can proceed on an undefined package.

4.2 Queues grow faster than load

The second reason is that waiting does not grow in proportion to how busy design is. Queueing theory gives a widely used approximation for the mean wait at a single station (Kingman, 1961; Hopp and Spearman, 2008):

W_q ≈ [ u / (1 − u) ] · [ (c_a² + c_s²) / 2 ] · t_s

where u is utilisation, t_s the mean service time, and c_a and c_s the coefficients of variation of arrivals and of service times. The first factor dominates near full load: at 80 % utilisation a package waits about four service times, at 90 % about nine, at 95 % about nineteen, all else equal (Figure 2). Design work has high variability in both arrivals (packages and vendor data arrive when suppliers and approvals allow) and service times (no two packages are the same), which raises the second factor. A design team that management sees as efficiently loaded is, by this account, also a team in front of which work waits a long time.

5What a full design queue costs

When the design gates are congested, the project pays in four ways. The first and fourth are documented in the sources cited. The second and third are, in this paper, the author's observations from project work, marked as such.

1. Packages wait. A package whose commercial case is ready waits for its definition, and a supplier's delivery waits for its vendor data to be engineered in. In Hammad's (2006) account, removing waits from the package cycle, including by placing orders before funding so that vendor engineering could start earlier, was among the factors that took 4 to 5 months off typical programme schedules.
2. ObservationCommitments are made early, on thin information. To give design a head start, purchases are committed before the information that should inform them exists. Hammad (2006) reports the cost of the risk this carries in his case: engineering committed on orders placed before funding ranged from 3 % to 5 % of those orders' value, lost if the project was deferred or cancelled. Lean practice argues instead for committing at the last responsible moment (Poppendieck and Poppendieck, 2003), which is only possible when design can keep up once the decision is taken.
3. ObservationJustified changes are deferred or refused. A change that would improve the plant is weighed against the design queue rather than against its value. In our experience a share of good changes is never made for this reason. We know of no published measurement of this share, and section 8 proposes one.
4. Site waits. Late design release means late material and equipment at the workface. Construction Industry Institute studies cited by Hammad (2006) estimate that 10 % to 12 % of construction labour cost could be saved if materials and equipment were available at the workface when needed. The more recent work-sampling data of Porwal, O'Brien and Borcherding (2025) point the same way: crews worked a larger share of their time where engineering and procurement were sequenced to the construction work packages.

ObservationIn plant modifications carried out during planned maintenance shutdowns, the cost is discontinuous. A design that is not ready for the shutdown window does not slip by the length of its delay; it waits for the next window, which may be a year away.

6Valuing design time at the project's rate

If design is the constraint, the value of design time is the value of project time. Reinertsen (2009) calls this the cost of delay: what one more unit of time before completion costs the party that wants the result. For an industrial investment it can be written, to first order, as

cost of delay per day ≈ contribution per day of the completed plant + cost of capital tied up per day + time-related project costs per day

ObservationIn the author's experience the first term dominates for a production asset and is far larger than the daily cost of the design team for one package. We give no figure for it here, because it depends on the plant, the product and the market, and because the owner of a project knows its own number far better than any outside estimate. We suggest that anyone assessing the argument insert theirs.

There is published evidence that owners already reason this way. Hammad (2006) reports that Saudi Aramco accepted committed engineering of 3 % to 5 % of the value of orders placed before funding, at risk of loss, as part of a set of procurement strategies that took 4 to 5 months off its programme schedules. An owner that pays a risk premium on equipment orders to gain months is valuing project time far above the cost of the activity that gains it.

Direct costs also understate effects elsewhere. In a longitudinal study of 19,605 rework events on 346 projects, Love et al. (2018) found a mean rework cost of 0.39 % of contract value on the projects with complete cost data, yet the contractor's mean yearly profit over the period was reduced by 28 %. Rework is not design waiting, but the lesson transfers: an effect that looks small on the cost line can be large on the measure that matters.

At a constraint, an hour of design is priced at the value of an hour of the project, not at the engineer's rate.

Two consequences follow for how design is bought. First, comparing design suppliers on hourly rate optimises the smaller number. Second, a design organisation that keeps spare capacity is not wasteful by this account: by section 4.2, slack is what keeps the queue short.

7What the argument does and does not imply

The argument concerns the design gates only. Fabrication, shipping, customs clearance, permits and the customer's own approvals remain on the critical path and are untouched by anything a design organisation does. Where one of those is the binding constraint, more design capacity shortens nothing.

Where design is the constraint, the question is how to add effective capacity without adding the delay of hiring. Three general routes are available to any design organisation:

• Pooled capacity. A peak in one project is absorbed by a larger pool of teams that know the work, rather than by overtime in one office.
• Recurring packages as parametric families. Packages that recur across projects, such as access structures around equipment, can be described once as a family with parameters, limits and checks, so that a new instance is a parameter change reviewed by an engineer rather than a new design.
• One record of revisions and changes. When every revision, change request and answer is held in one place, the second gate starts from current information rather than from a search.

Nordform's approach to the second route, including which of its families are still illustrative and unsized, is described on its engine page. Whether any of these routes removes the constraint in a given project is an empirical question, and the author's interest in the answer is declared below.

Qualified people remain responsible for released drawings, and release for fabrication remains with the manufacturer. Nothing in this argument changes who signs.

8Proposed measures

The argument predicts that, where design is the constraint, packages wait for design measurably, that the wait shortens when design capacity is added effectively, and that more justified changes are accepted. It is falsified for a project if packages do not wait for design, or if adding capacity does not shorten the wait. We propose three measures, taken on one recurring package type in one live project, first as a baseline and then on the next instances.

Three cautions apply. The measures need timestamps that most projects do not record today, so the baseline may have to be reconstructed from correspondence and document registers, with the uncertainty that brings. One package type in one project is a case, not a sample. And the party measuring should not be the only party reading the result; the customer should hold the data.

9Limitations

• No data of our own. This paper reports no measurement by the author. Statements marked Observation rest on the author's project experience and are hypotheses until measured.
• Indirect published evidence. The recent studies in section 3 show that design lateness and faults are frequent causes of overrun, but they sample disputed projects or rank perceived causes, and none isolates design capacity from design quality. The willingness-to-pay figures in section 6 come from one owner's programme of very large oil and gas projects, reported in 2006; they show that months have been gained and paid for, not how much of that was design, nor that the figures transfer to smaller projects or other industries.
• Design is not always the constraint. In many projects the binding constraint is funding, permits, fabrication capacity or the owner's decisions. The argument applies only where design is the constraint, and the measures in section 8 are how to tell.
• Simplified model. Kingman's approximation describes a single station in steady state. Design work is a network of people with different skills, priorities change, and work is pre-empted. The approximation explains the shape of the effect, not its size in any organisation.
• Value is not margin. The cost of delay in section 6 includes the contribution of earlier operation. Revenue figures overstate it; only the owner can state the margin and whether earlier output can be sold.
• Conflict of interest. The author co-founded and runs a company that sells design capacity. The argument favours buying design on project terms rather than hourly terms. Readers should weigh it accordingly, and we invite measurement by others.

10Conclusion

In industrial investment projects, design gates every package twice: before it can be bought, and after its vendor data arrives. Because packages are staggered and design capacity is hard to add quickly, design often becomes the constraint that sets the pace of the project, and queueing theory explains why a well-loaded design team is also one in front of which work waits. Where that holds, design time should be valued at the project's cost of delay rather than at the engineer's rate. We have proposed three measures of waiting by which the claim can be tested on real packages, and we intend to publish what they show, including if they show the argument to be wrong.

11Competing interests

The author is a co-founder of Nordform Group Oy, which sells industrial design and engineering. No external funding was received for this paper.

12References

1. Abdelsayed, M. and Moffat, D. (2023). Designs on better outcomes: getting to the CRUX of time and cost overruns on capital projects. HKA, 1 September 2023. hka.com/article/designs-on-better-outcomes-getting-to-the-crux-of-time-and-cost-overruns-on-capital-projects (accessed 20 October 2026).
2. Alshibani, A., Julaih, A., Adress, A., Alshamrani, O. and Almaziad, F. (2023). Identifying and ranking the root causes of schedule delays in oil and gas pipeline construction projects. Energies, 16(1), 283. doi.org/10.3390/en16010283
3. Goldratt, E. M. and Cox, J. (1984). The Goal: A Process of Ongoing Improvement. Great Barrington, MA: North River Press.
4. Goldratt, E. M. (1997). Critical Chain. Great Barrington, MA: North River Press.
5. Hammad, M. A. (2006). Schedule improvement through innovative procurement strategies. Paper presented at PMI Global Congress 2006, Latin America, Santiago, Chile. Newtown Square, PA: Project Management Institute. pmi.org/learning/library/schedule-improvement-innovative-procurement-strategies-8116 (accessed 20 October 2026).
6. Hatmoko, J. U. D. and Khasani, R. R. (2019). Mapping delay risks of EPC projects: a case study of a platform and subsea pipeline of an oil and gas project. IOP Conference Series: Materials Science and Engineering, 598, 012095. doi.org/10.1088/1757-899X/598/1/012095
7. HKA (2026). Powering through Uncertainty: CRUX Insight ninth annual report. London: HKA, 15 October 2026. Key findings as released at hka.com/news/crux-insight-ninth-annual-report-powering-through-uncertainty (accessed 20 October 2026).
8. Hopp, W. J. and Spearman, M. L. (2008). Factory Physics, 3rd ed. New York: McGraw-Hill/Irwin.
9. IEA (2025). World Energy Employment 2025, executive summary. Paris: International Energy Agency. iea.org/reports/world-energy-employment-2025/executive-summary (accessed 20 October 2026).
10. Kingman, J. F. C. (1961). The single server queue in heavy traffic. Mathematical Proceedings of the Cambridge Philosophical Society, 57(4), 902–904.
11. Love, P. E. D., Smith, J., Ackermann, F., Irani, Z. and Teo, P. (2018). The costs of rework: insights from construction and opportunities for learning. Production Planning & Control, 29(13), 1082–1095. doi.org/10.1080/09537287.2018.1513177
12. Norwegian Petroleum Directorate (2020). Project execution on the Norwegian continental shelf. Stavanger: NPD (now the Norwegian Offshore Directorate). sodir.no/en/whats-new/publications/reports/older-reports/project-execution-on-the-norwegian-continental-shelf (accessed 20 October 2026).
13. Poppendieck, M. and Poppendieck, T. (2003). Lean Software Development: An Agile Toolkit. Boston: Addison-Wesley.
14. Porwal, V., O'Brien, W. J. and Borcherding, J. D. (2025). Examining the impact of Advanced Work Packaging and WorkFace planning on direct work rates of on-site construction workers: a comparative analysis. Frontiers in Built Environment, 11, 1621337. doi.org/10.3389/fbuil.2025.1621337
15. Reinertsen, D. G. (2009). The Principles of Product Development Flow: Second Generation Lean Product Development. Redondo Beach, CA: Celeritas.
16. Yeo, K. T. and Ning, J. H. (2006). Managing uncertainty in major equipment procurement in engineering projects. European Journal of Operational Research, 171(1), 123–134. doi.org/10.1016/j.ejor.2004.06.036

13How to cite

Lappalainen, E. (2026). Design is what the project waits for: engineering design capacity as the constraint on industrial investment projects. Nordform working paper, version 0.1. https://nordform.ai/research/design-bottleneck/

@techreport{lappalainen2026design,
title       = {Design Is What the Project Waits For: Engineering Design Capacity
               as the Constraint on Industrial Investment Projects},
author      = {Lappalainen, Eelon},
year        = {2026},
institution = {Nordform},
type        = {Working paper},
number      = {Version 0.1},
url         = {https://nordform.ai/research/design-bottleneck/}
}

14Versions

• 0.1 · 21 October 2026 · First public edition, with recent evidence (2019–2025) in section 3. A working paper is revised in place; each version is listed here with what changed.