At a glance

• Design commits the money early. Extensive research indicates that 80% of a product’s life-cycle cost is determined during the initial stages of design and development, a finding that recurs across engineering disciplines.

• Design errors are the largest single source of rework. Research consistently shows that over 70% of rework traces back to design errors, including coordination conflicts, omissions, and inconsistencies in construction documents. On industrial projects studied by the Construction Industry Institute, design changes, errors and omissions averaged 78% of the total number of deviations and 79% of the total deviation costs.

• Rework itself is material. Across project types and regions, rework accounts for 1-20% of total project cost, with most studies clustering between 4-10%, and research by Love (2002) found that rework contributes to an average of 52% of total cost growth on construction projects.

• India’s largest highway owner has made design quality a procurement criterion. NHAI has rated 56 DPR consultants on a scale of 100, with the top-rated consultant having a score of 80.75, while the lowest-rated consultant scored 23.33, and this rating is expected to become a deciding factor for awarding DPR contracts.

• Simulation is changing from a verification step to an exploration engine. A February 2026 systematic mapping of 3,345 relevant papers found that Optimization and Design remains the largest application area, while Structural Performance Prediction and FEM Acceleration/Surrogate Modeling show the fastest growth.

• The design firms that will lead are those that treat design as a data and decision system - surveys, codes, precedent and simulation connected - rather than a document to be produced.

Design is where the cost is committed, not where it is spent

The economics of infrastructure design are asymmetric. Design fees are a small fraction of capital cost; the decisions taken in design determine most of it. The design phase is widely acknowledged as the most influential stage of the project lifecycle because decisions relating to project scope, building configuration, materials, engineering systems, procurement strategies and construction methodologies are established at this stage and determine a significant proportion of the project’s final cost and overall performance. The same logic governs time: to discover incompatible designs at a later stage is far more detrimental in terms of time and cost as compared to during the initial design phase.

For an infrastructure owner, this means the cheapest point at which to fix a highway alignment, a bridge foundation or a drainage scheme is before the tender is issued. The most expensive point is after the contractor has mobilised. Every other conversation about productivity - construction technology, monitoring, AI on site - is a conversation about managing the consequences of what design did or did not get right.

Roughly 80 percent of lifecycle cost is committed by decisions taken in the earliest design stages, while the cost of changing those decisions rises steeply as a project moves from concept to detailed design to construction. Design-related deviations have historically accounted for around 12 percent of total installed cost on complex industrial projects and 1-9 percent across broader samples, with the range narrowing as digital coordination has improved. Rework overall sits at 4-10 percent of project cost in most studies and explains about half of cost growth. (Sources: Construction Industry Institute; PlanRadar 2025 review; Love 2002; arXiv 2601.12276.)

What actually goes wrong in design

The rework literature is unusually consistent about causes. Design-related errors and deviations account for 1-9% of total project cost, including design omissions, inaccurate drawings, and late design changes. Communication compounds them: a joint report by PlanGrid and FMI found that 26 percent of rework on U.S. construction sites is linked to poor communication, with an additional 22 percent attributed to inaccurate or incomplete project information.

In Indian highway practice the failure modes are specific and public. NHAI’s consultant rating followed concerns over flawed DPRs including poor surveys, weak geometric design and drainage planning which have contributed to unsafe roads, accidents and cost overruns. The Union Minister for Road Transport and Highways has said that consultants involved in preparing defective detailed project reports are “culprits” behind accidents and deaths on highways. NHAI’s own framing is corrective rather than punitive: DPRs are the foundation of any highway project, and weaknesses at this stage often cascade into execution delays, safety issues and cost overruns.

Four root causes recur across sectors.

Surveys that do not match the ground. Topographic, geotechnical and hydrological inputs that are thin, old or extrapolated. Every downstream quantity, from earthwork volumes to foundation depths, inherits the error.

Geometry designed for the drawing, not the driver. Horizontal and vertical alignments that meet code minima on paper but produce sight-distance, superelevation and junction conflicts in the field.

Drainage and hydrology treated as a chapter, not a system. Cross-drainage sized without catchment modelling; longitudinal drainage that ignores adjacent land use; culverts placed by chainage rather than by flow.

Coordination failures between disciplines. Structures, pavement, utilities and land acquisition designed in sequence rather than together, so that a bridge approach, a utility corridor and a drain occupy the same space on three different drawings.

None of these are exotic. They are the predictable output of a design process built around producing a document on a deadline rather than around resolving a physical problem with verified data.

The good news: digital coordination has already cut design error cost

The trend line is encouraging where digital methods have been adopted. Since the early 1990s, the cost impact of design-related errors has fallen from around 9% to just 1-2% of total project cost, reflecting the positive effects of digitisation, BIM adoption, and improved design coordination workflows. Studies using the Construction Industry Institute database of 359 projects found that utilising BIM reduces the workload throughout design and construction stages and is effective in reducing the impact of rework on total project cost and duration.

Two conclusions follow. First, design quality is not a talent problem alone; it responds to tooling and process. Second, the frontier has moved. Coordination is now table stakes. The next gains come from changing what design can explore, not just how cleanly it is documented.

Simulation is moving from verification to exploration

For most of its history, structural and geotechnical simulation was a check performed on a design already chosen. Finite-element analysis was expensive enough that engineers ran it on one or two candidates, late. That constraint is dissolving.

The mechanism is surrogate modelling. AI-accelerated finite element analysis is the integration of machine learning, neural networks and surrogate modelling techniques into traditional computational structural and physics simulation workflows to dramatically reduce solve times, improve design iteration speeds, and enable real-time optimisation. The dominant architecture uses FEM simulations to generate labelled training data, then trains ML models to predict FEM outputs at inference time without re-running the solver. The 2026 systematic mapping of the field found that a subset of the studies integrate surrogate or neural network models directly into the optimisation loop to reduce the computational cost of repeated FE evaluations, with objectives that commonly include weight reduction, stiffness improvement, or increased buckling resistance under predefined load cases.

The practical consequence for a design office is that the number of alternatives evaluated per week stops being limited by solver time. Instead of creating a single design option, AI generates multiple solutions based on engineering requirements and project constraints, and engineers define factors such as material type, loading conditions, budget, and design objectives. That is the design-side counterpart of what continuous monitoring is doing on the operations side: sampling becomes census.

Simulation is also reaching into the built asset. Recent bridge research proposes a “mechanical-core” digital twin built by integrating computer vision sensing with automated finite element model updating, so that the analysis model synchronises with the as-built state of the physical structure. The design model no longer ends at handover; it becomes the seed of the operating twin.

Traditional practice: one or two candidate designs, full finite-element runs late in the process, results used to confirm rather than choose. Emerging practice: hundreds of candidates screened by surrogate models in minutes, full analysis reserved for a shortlist, optimisation loops that trade weight, stiffness, cost and constructability simultaneously, and analysis models that persist into operations and update from sensor and vision data. Growth evidence: surrogate modelling and structural performance prediction are the fastest-growing application areas in AI-plus-FEM research, and industrial patents now explicitly replace or eliminate FEM solver calls during design optimisation loops. (Sources: Buildings 16(3):644, February 2026; Patsnap landscape, April 2026; Applied Sciences 16(6):2920, March 2026.)

The limits are real and should be stated

The evidence is equally clear about where surrogate and generative methods fail. AI predictions are highly dependent on the training dataset and can falter on geometries or loading cases outside that domain, and deterministic predictions, though fast, lack insight into model confidence. Current frameworks approximate equilibrium and compatibility but do not strictly enforce them, which is why physics-informed neural networks that embed physical constraints are the active research direction.

For infrastructure this matters more than for consumer products. A bridge in a seismic zone with unusual soil is by definition out of distribution. The design rule that follows is simple: surrogates screen, solvers certify, engineers sign. Any workflow that collapses those three roles into one is not faster; it is unsafe.

The modern design stack, and what holds it together

Digital design in infrastructure now rests on five connected capabilities. Firms that have all five connected outperform firms that have all five in separate tools.

Verified ground data. Drone photogrammetry, LiDAR, network survey vehicles and satellite imagery feeding a single georeferenced model. NHAI already requires drone video recording on a monthly basis, uploaded to its Data Lake portal; the same capture discipline applied at DPR stage removes the largest source of downstream error.

Requirements management. One of the most impactful trends recently is the increased focus on requirements management across rail and broader AECO organisations. Traceability from owner requirement to design decision to as-built is what makes a design auditable - and what a code-checking engine runs on.

Codified standards. IRC and IS codes, MoRTH specifications and owner-specific conditions expressed as machine-checkable rules rather than PDFs on a shelf, so that a drawing can be checked continuously rather than reviewed once.

Simulation and optimisation. Surrogate-accelerated analysis with full solvers for certification, connected to the same model the drawings are produced from.

Institutional precedent. The firm’s own history of what was designed, what was reviewed, what was built and what failed - searchable by the engineer at the desk. This is the capability most firms lack and the one that compounds fastest.

Ground truth first: no alignment or structure is designed on survey data the team has not verified in the current season. Codes as rules: every drawing checked against codified standards before human review, so reviewers spend time on judgement rather than compliance. Options, not answers: at least several alternatives evaluated with cost, constructability and lifecycle impact before one is recommended. Coordination in one model: structures, pavement, drainage, utilities and land in the same georeferenced space, with clashes resolved before tender. Signed accountability: every output carries the engineer who checked it and the evidence they checked it against. Design carries forward: the design model becomes the handover model and the operating twin.

Owners are pricing design quality

The single most important shift in the Indian market is that design quality now has a commercial value attached to it. NHAI’s first-of-its-kind rating evaluated consultants on technical soundness, completeness of surveys, design accuracy and adherence to prescribed standards, with the objective of aligning incentives with long-term project outcomes rather than short-term contract wins. Rankings will influence eligibility and preference for future DPR contracts.

That changes the economics of design investment. A firm that spends more on survey verification, simulation and code-checking is no longer just spending more; it is buying eligibility. The parallel with continuous monitoring on the operations side is exact: the owner now measures, so the consultant must too.

Two further pressures reinforce this. Lowest-bid awards are producing tenders 20 per cent or more below estimated cost, putting commercial viability and execution quality under pressure, which means design errors that once absorbed in contractor margin now surface as claims and disputes. And at the front end, projects are stalling because no consultant has come forward to prepare the DPR - a signal that the supply of high-quality design capacity, not demand, is the binding constraint.

How RODIC approaches design

Design sits at the front of the RODIC infrastructure AI ecosystem for the reason this article sets out: it is where the largest share of an asset’s cost and risk is committed, and where verified data and encoded judgement have the highest leverage. Four commitments shape the firm’s design practice.

Ground truth before geometry. Survey, geotechnical and hydrological inputs are captured and verified with current-season drone, LiDAR and field data before alignment or structural design begins. The design model is georeferenced from day one so that structures, pavement, drainage, utilities and land are coordinated in one space rather than reconciled after tender.

Codes as executable rules. IRC and IS provisions, MoRTH specifications and owner conditions are expressed as machine-checkable rules within the firm’s rules and compliance engine, so that drawings and DPR chapters are checked continuously during design rather than reviewed once at the end. Human reviewers spend their time on engineering judgement, not clause hunting.

Simulation to explore, solvers to certify, engineers to sign. Surrogate-accelerated analysis is used to screen many alternatives on cost, constructability and lifecycle performance; full solvers certify the shortlist; a named engineer signs the recommendation. The design model carries forward into construction monitoring and, ultimately, into the operating twin of the asset.

Precedent at the desk. The firm’s knowledge layer makes precedent designs, review history and code interpretations from two decades of delivery searchable by the engineer drafting a new bridge, tunnel or interchange. Design quality compounds because every reviewed project makes the next one better.

Alongside its own practice, RDA works with academic partners to publish research on design quality, safety and lifecycle performance, because the owner’s new rating regimes will reward tested evidence over assertion.

Implications for leaders

For asset owners. Extend outcome-based rating from DPR consultants to the design inputs themselves: require verified current-season surveys, coordinated single-model deliverables and machine-checkable compliance evidence in DPR contracts. The cost of a better DPR is small against the cost of the rework it prevents.

For design consultancies. Treat survey verification, code automation and simulation capacity as capital investments that buy eligibility, not overheads that reduce margin. Measure design quality internally before the owner does - error rate at review, clashes found before tender, revision cycles per package.

For contractors. Engage in design early where the contract allows. The constructability knowledge that lives on site is the input most designs lack, and the one most likely to reduce claims later.

For policymakers. The DPR rating framework is the right instrument. The next step is a common data standard for DPR deliverables - georeferenced models, structured quantities and compliance evidence - so that ratings can be computed from the deliverable itself rather than from post-hoc review.

The guardrails

Design is where a mistake costs the least to fix and the most to miss. Three rules apply. Surrogates screen, solvers certify, engineers sign - no AI-generated design leaves the office without a named engineer’s verification against a full analysis. Every recommendation carries its data source and its confidence, and out-of-distribution cases are flagged for full analysis by default.

Infrastructure productivity debates tend to start at the construction site. The evidence says they should start earlier. Most of an asset’s cost, safety and lifecycle performance is fixed in design, most rework traces back to design, and India’s largest owner has now made design quality a condition of eligibility. Simulation and AI are turning the design office from a document factory into a decision engine, provided the guardrails hold. The firms that connect verified ground data, codified standards, precedent and simulation into one design system will be the ones owners rate highest - and the ones whose assets perform.