Abstract

Project controls has made substantial advances through BIM, 4D planning, 5D cost integration and predictive analytics. Each evolution has improved the profession’s ability to see what is happening and anticipate what may happen next. But a persistent gap remains: better visibility does not automatically produce better decisions. This paper introduces the Decision Digital Twin as the next logical evolution of project controls — a framework for modelling the decisions that determine whether assets, schedules and costs succeed, not merely the assets themselves. Drawing on the causal logic of decision modelling, it proposes a working definition, a maturity model contextualising the Decision Digital Twin within existing practice, and a practical demonstration using a major programme recovery scenario. The paper argues that the future of project controls lies not only in better models of physical assets, but in better models of the choices that govern them.

  1. 01The gap between seeing and deciding — why decades of investment in visibility have not translated into better decisions.
  2. 02Why prediction alone is not enough — the structural limits of predict-then-act logic, and the case for decision robustness.
  3. 03Defining the Decision Digital Twin — a working definition, and how it differs from a schedule, risk register or 5D model.
  4. 04The maturity model — positioning the DDT within the existing 3D / 4D / 5D / predictive analytics sequence.
  5. 05Application: a programme recovery decision — a worked scenario showing the DDT’s causal chains and trade-off comparison in practice.
  6. 06Building it in practice — a workable, low-tech starting point requiring no new platform or software.
  7. 07A reference architecture — five functional layers, from data foundations to decision interface, and a recommended build sequence.
  8. 08Governance, assurance, and the decision record — how the DDT creates an auditable record of reasoning, and an early warning mechanism.
  9. 09Project controls as the decision function — why project controls is best placed to own and operate this capability.

Section 1

The gap between seeing and deciding

The dominant narrative in project controls improvement is one of visibility: better data integration, richer dashboards, earlier warning signals, tighter forecast cycles. These are real advances, and the profession should take credit for them. Over the past two decades, the quality of information available to programme teams has improved markedly. The gap between what is happening on a major project and what the controls function knows about it has, in most competent organisations, narrowed considerably.

But visibility is not the same as decision quality. A programme team facing a six-month schedule slip does not typically lack for information. They have a forecast, a risk register, an earned value report, a change log, a set of commercial exposures, and probably a dashboard that aggregates all of the above. What they often lack is a structured way to reason through their options — to understand what each available action would actually do to the outcomes they care about, across the range of futures they might face.

The investment the profession has made in visibility has been predicated on an assumption: that better information leads, more or less directly, to better choices. Sometimes it does. But on complex programmes, where the decisions are genuinely difficult and the consequences are non-linear, the assumption breaks down. Better reporting surfaces the problem more clearly. It does not necessarily surface the answer.

We have digitised the project. The next step is to digitise the decision.

This paper addresses that gap — not through better reporting or more accurate prediction, but through a new layer of capability that sits above prediction and asks a different question: given what we know and what might happen, what should we do?

Section 2

Why prediction alone is not enough

The natural response to uncertainty in project controls has been to improve forecast quality. If the team can predict more accurately, the reasoning goes, they can act more confidently. This is sensible under stable conditions, where the future is an extension of the recent past and interventions are modest and reversible. But major projects rarely offer those conditions.

There is a structural problem with predict-then-act logic that goes beyond forecast accuracy. On complex programmes, the actions available to decision-makers change the thing being forecast.

  • A resequencing decision alters the critical path and therefore the schedule model.
  • A commercial intervention changes the risk profile and therefore the probability distribution of outcomes.
  • A scope deferral improves near-term delivery certainty but modifies the benefits trajectory and therefore the investment case.

In each case, the act of choosing reshapes the landscape being predicted. The forecast is not a stable backdrop against which decisions are made — it is itself an output of the decisions being considered.

Predictive approaches ask: what is the most likely future? That is a useful question when uncertainty is manageable and interventions are incremental. But where uncertainty is high and the consequences of choices are significant, the more important question is: which action is most robust across the range of futures we might face?

Those are different questions, and they call for different tools. A predictive model optimised for accuracy tells you where the programme is likely to land. A decision model tells you which intervention holds up best when the forecast is wrong — which it routinely is. The distinction between forecast confidence and decision robustness is what the Decision Digital Twin is built on.

Project controls has historically invested heavily in narrowing forecast uncertainty. The argument here is that the next productive investment is in improving the quality of reasoning about options — not instead of better forecasting, but in addition to it.

Figure 1 — Two modes of reasoning under uncertainty. Predict-then-act logic breaks down when the act of choosing reshapes what is being forecast. Decision robustness asks which option holds up best across the futures the programme might face.

Section 3

Defining the Decision Digital Twin

The term “digital twin” has been used broadly enough in construction and infrastructure to require some precision here. In its most common usage, a digital twin is a virtual representation of a physical asset — a model that reflects the current and projected state of something in the real world. The value of such a twin lies in its fidelity to the physical object and its ability to simulate how that object behaves under different conditions.

The Decision Digital Twin borrows the underlying logic but changes the unit of analysis. The object being modelled is not an asset, a schedule or a cost baseline. It is a decision.

Working definition

A Decision Digital Twin (DDT) is a dynamic, evidence-informed model of a project decision. It connects available actions, causal pathways, uncertainties, constraints, trade-offs and desired outcomes so that decision-makers can test options before committing resources in the real delivery environment.

The distinction between a DDT and the other analytical tools in a project controls function turns on what is being represented.

  • A schedule model represents a sequence of activities and their dependencies.
  • A risk register represents a set of threats and opportunities and their associated probabilities.
  • A 5D model represents cost against scope and time.

Each of these is a model of a project artefact. A DDT is a model of a decision — specifically, the logical structure connecting what a team might do, through what intermediate consequences, to what outcomes they are responsible for.

The architectural backbone of that model is the causal decision diagram: actions on the left, outcomes on the right, and the chain of intermediate consequences connecting them. What converts that diagram from a conceptual framework into a working analytical tool is the attachment of project controls data to each causal link. Schedule data informs timeline consequences. Cost data informs financial consequences. Risk data provides probability weightings for uncertain pathways. Benefits data constrains what counts as a successful outcome. Commercial data defines the boundaries of viable options.

Project controls functions already hold most of this data. The barrier to building a DDT is rarely a data problem. It is an architectural problem: the data has not been organised around a decision. The DDT is, in that sense, less a new technology than a new discipline — the practice of structuring existing evidence around the specific choices that matter most.

Figure 2 — Anatomy of a DDT. Actions on the left, outcomes on the right, causal pathways connecting them in the centre. Project controls data populates each link in the chain.

Section 4

The maturity model: a sequence, not just a taxonomy

The DDT can be positioned within a broader maturity model of project controls intelligence. That model is presented below.

StageThe question answeredPrimary tool
3D / BIMWhat are we building?Geometric model
4DWhen will it be built?Schedule-linked model
5DWhat will it cost?Cost-integrated model
Predictive analyticsWhat is likely to happen?Forecasting and AI
DDTWhat should we do, and what follows from it?Causal decision model

The model is sometimes read as a simple taxonomy — different tools answering different questions at the same level of analysis. But there is a dependency logic that deserves attention, because it changes how the DDT should be introduced to an organisation.

3D modelling creates a digital representation of the asset. 4D adds temporal structure, enabling schedule simulation. 5D adds financial structure, enabling cost integration. Predictive analytics draws on the data generated by these layers to produce probabilistic forecasts. The DDT draws on all of them — schedule, cost, risk, benefits, change — and attaches that evidence to a causal model of a specific decision.

This means the DDT has data dependencies. An organisation with weak 5D capability does not yet have the cost consequence data needed to populate the financial pathways in a recovery decision model. An organisation without a functioning risk register has no probability weighting to attach to uncertain causal links. The maturity sequence is not inevitable or uniform — organisations will develop unevenly, and some DDT applications will be viable before others — but the dependencies are real and should be assessed before implementation is attempted.

Project controls teams are not being asked to discard the capabilities they have built. They are being asked to integrate those capabilities around a new unit of analysis: the decision. The DDT does not make 5D obsolete. It makes 5D more valuable by connecting cost data to the choices that actually drive cost outcomes.

Section 5

Application: modelling a programme recovery decision

To demonstrate what a DDT produces in practice, consider a scenario familiar to anyone who has worked on a major infrastructure programme. A programme is forecast to miss a key milestone by six months. The project controls function has data: an updated schedule, an earned value position, a risk register with updated probability scores, a commercial exposure log, and a revised cost forecast. The executive team has a decision to make.

In a conventional project controls environment, that data is presented to the decision-makers and a conversation follows. Often a good conversation — experienced practitioners can reason through consequences intuitively. But the reasoning is largely implicit, the trade-offs are rarely made explicit across all relevant outcome dimensions simultaneously, and the record of what was considered and why rarely survives the meeting.

A DDT changes the structure of that conversation by making the decision model explicit before it begins.

The decision structure

The first step is to identify the available actions. In a typical milestone recovery scenario, these might include:

  • Accelerate through additional resource
  • Resequence works to protect the critical path
  • Rebaseline the programme
  • Defer or reduce scope
  • Change contracting or packaging strategy
  • Accept the delay and protect cost
  • Prioritise benefits-critical work packages

The second step is to identify the outcomes the organisation cares about. In infrastructure delivery, these typically span multiple dimensions simultaneously:

Schedule Milestone recovery and programme certainty

Cost Budget confidence and commercial exposure

Safety Site conditions and incident risk

Quality Output standards and defects risk

Carbon Embodied and operational emissions impact

Benefits Realisation trajectory and delay cost

Stakeholders Confidence and reputational exposure

Supply chain Capacity, resilience and relationship risk

The causal chain

The value of the DDT lies in making the causal pathways between actions and outcomes explicit, rather than leaving them implicit in the minds of individual practitioners. The diagram below illustrates the causal logic for a single intervention: acceleration through additional resource.

Figure 3 — Causal chain for the “add second shift” intervention. Cross-links (dashed) show where one intermediate drives consequences in more than one outcome dimension.

What the model produces

The same causal logic can be applied to each available action, producing a structured comparison of options across outcome dimensions. A simplified version of that comparison, for four of the seven actions in this scenario, is shown below.

ActionScheduleCostSafetyBenefitsStakeholders
Add resource↑ Recovery likely↑ Pressure↑ Risk↑ Protected↑ Confidence
Resequence↑ Near-term~ Neutral~ Low change↑ Later risk~ Depends
Rebaseline↓ New target↑ Protected~ Low change↓ Delay cost↓ Scrutiny
Defer scope↑ Improves↑ Reduces~ Low change↓ Damage↓ Exposure

This is the output the DDT provides: not a single recommendation, but a structured account of what each option does to the outcomes the organisation cares about. The trade-offs are made visible before the decision becomes irreversible. The conversation that follows is better informed, more explicit, and more defensible.

It does not eliminate uncertainty. It does not remove the need for experienced judgement. And it does not hand you an optimal answer. What it does is give the project team a shared frame for reasoning about consequences — and a record of the reasoning once the choice is made.

Section 6

Building it in practice: a workable starting point

The DDT is a compelling concept. It is also, to a practitioner audience, a concept that raises an immediate and reasonable question: how does an organisation actually build one? The answer matters, because the history of project controls innovation is full of frameworks that were persuasive on paper but unusable in practice — ideas that required either a wholesale systems transformation or a level of analytical sophistication that most programme teams do not have at their disposal.

The DDT does not need to be either of those things. The minimum viable version is more modest than the full architecture implies, and it can be built using tools and data that most mature project controls functions already have.

Start with a decision, not a platform

The most common mistake in implementing decision modelling is to begin with the technology — to procure a platform, configure an integration layer, and wait for the analytics to produce insights. That sequence inverts the logic. A DDT is built around a specific decision, not around a data architecture. The right starting point is to identify the highest-stakes decision currently facing the programme and ask: what are the options, what are the outcomes we care about, and what does our existing data tell us about the consequences of each?

That question can be answered, at least partially, in a structured workshop. An experienced project controls team can map the available actions and the causal pathways to outcomes using existing knowledge, without any new technology at all. The first DDT an organisation builds is often essentially a facilitated analysis — a causal diagram on a whiteboard, populated with data drawn from the risk register, the cost forecast and the schedule model. It is less sophisticated than the full vision, and considerably more useful than the status quo.

What a low-maturity implementation looks like

For an organisation at the earlier stages of the maturity model — a practical DDT implementation has three components.

  1. A decision framing session that names the decision explicitly, identifies the available actions, and specifies the outcome dimensions that matter.
  2. A structured evidence review that draws on existing project controls data — schedule, cost, risk, benefits — to populate the causal pathways for each action.
  3. A documented trade-off matrix of the kind shown in Section 5, reviewed by the decision-makers before the choice is made.

None of this requires new software. It requires discipline: the discipline of treating a decision as an object that can be designed and documented, rather than a conversation that happens and then passes. The tooling that most teams reach for — a spreadsheet for the trade-off matrix, a risk model for probability weighting, a scheduling tool for timeline consequences — is almost certainly already in place.

Figure 4 — A low-maturity DDT implementation requires no new technology. The three components can be delivered using existing project controls data and standard analytical tools.

What a higher-maturity implementation adds

Organisations with more advanced analytical capability can build on this foundation in meaningful ways. Predictive models can be used to populate the causal pathways with probabilistic estimates rather than directional judgements, replacing arrows labelled ”↑ cost pressure” with quantified distributions. Monte Carlo simulation can stress-test each option across a range of scenarios rather than a single forecast. Where AI-assisted analytics are available, pattern recognition across historical programme data can surface causal links that practitioners might not have identified from experience alone.

The technology does not change the underlying logic. It changes the precision with which that logic can be populated and the speed with which new decisions can be modelled. A team that has built a DDT manually for one major decision is well positioned to automate and enhance that process for the next — because the conceptual architecture, once understood, is reusable.

The decisive condition is cultural

Technology and data aside, the most important condition for a successful DDT implementation is organisational: the willingness to make decision-making explicit. In many programme environments, the implicit logic of decisions is deliberately kept implicit. Options that were rejected are not recorded. Trade-offs that were accepted are not named. The opacity is not always wilful — it often reflects the pace at which decisions are made and the informality of the conversations in which they happen.

A DDT asks an organisation to change that habit. It asks decision-makers to commit to naming the options they considered, the consequences they anticipated, and the trade-offs they accepted. That is a cultural ask as much as an analytical one. The organisations that will get the most from this approach are those where programme leadership is willing to make its reasoning visible — not because it makes them look good, but because it makes the programme more manageable and the organisation more capable of learning.

The role of the project controls function in that change is to provide the framework that makes explicitness easy rather than burdensome. If the trade-off matrix is already built, documenting the decision takes minutes rather than hours. If the causal diagram is already drawn, explaining the reasoning to a sponsor or assurance body requires explanation rather than reconstruction. The DDT reduces the cost of transparency — and in doing so, makes transparency the easy option.

Section 7

A reference architecture for the DDT

Section 6 established that the DDT does not require a new platform to get started. But as ambition and maturity grow, the question of how the components fit together becomes important. What follows is a reference architecture: a named set of functional layers and component types that together constitute a fully realised DDT, described in terms of what each layer does, what it consumes, and what it produces. It is not a procurement specification — technology choices will vary by organisation, programme size, and existing infrastructure. It is a conceptual map of the solution space.

The architecture has five layers. They are not strictly sequential in implementation — an organisation may have sophisticated capability in some layers before others — but they are logically dependent in the way described in the maturity model: each layer draws on the outputs of those below it.

Figure 5 — Reference architecture for the DDT. The five functional layers are logically dependent from base to apex. The governance and learning function (right) spans all layers — it is a practice, not a system.

Layer 1 — Data foundations

The base layer comprises the project controls datasets that most major programme organisations already hold: the schedule model, the cost baseline and forecast, the risk register, the change log, the benefits register, commercial and contractual data, and any repository of lessons learned from previous programmes. No new data is created at this layer. The work is ensuring that these sources are maintained to a standard that supports causal analysis — consistent definitions, regular refresh cycles, and traceable provenance. A risk register that has not been updated in three months, or a cost forecast that is not disaggregated to work-package level, will limit what the layers above can reliably produce.

Layer 2 — Data integration

The integration layer is where the data foundations are connected into a coherent, queryable environment. In practice, this means a common data environment — whether a dedicated platform, a data warehouse, or a well-governed set of structured exports — through which schedule, cost, risk and benefits data can be accessed together rather than in separate tools. The critical capability at this layer is not the technology itself but the data normalisation it enables: consistent coding structures, shared reference frameworks, and a version-controlled audit trail that makes it possible to reconstruct the information state at any point in time.

This layer is where many organisations currently underinvest. The data exists; the integration does not. Cost and schedule data live in separate tools with incompatible coding structures. Risk data is maintained in a register that does not map to the schedule activities it affects. Benefits data is held by a different team with a different refresh cycle. Closing these gaps is unglamorous work, but it is the prerequisite for everything above it. An organisation that cannot query cost, schedule, and risk data against a common work-breakdown structure cannot populate a causal decision model with any confidence.

Layer 3 — Analytics and enrichment

The analytics layer transforms integrated data into probabilistic, forward-looking evidence that can populate the causal pathways in a decision model. The core components are: predictive forecasting; quantitative risk analysis, including Monte Carlo simulation, which converts the risk register into probability distributions over outcomes; and, where available, AI-assisted pattern recognition that can surface historical precedents and causal relationships not visible in the current programme data alone.

The output of this layer is not a single forecast. It is a set of probabilistic estimates — distributions rather than point values — that can be attached to each causal pathway in the decision model. That shift from directional to quantified is what distinguishes a mature DDT from a structured workshop exercise.

Layer 4 — Decision modelling engine

The decision modelling engine is the layer that most directly realises the DDT concept. It provides the environment in which causal diagrams are built, actions and outcomes are defined, and the consequence chains connecting them are populated with evidence from Layer 3. This layer does not require a bespoke software platform. A well-structured spreadsheet model, built to a consistent template and connected to the data integration layer, can perform the core function of Layer 4 at modest maturity. The discipline of building the causal model explicitly is more important than the tool used to build it.

Layer 5 — Decision interface

The decision interface is what programme leadership and sponsors actually see: a trade-off explorer that allows decision-makers to interrogate the consequence profile of each option across multiple outcome dimensions; an option comparison view that surfaces the key differentiators between available actions; a decision record template that captures the choice made, the reasoning behind it, and the monitoring indicators that will track whether the underlying assumptions hold; and, post-decision, a monitoring dashboard that flags where the causal assumptions embedded in the decision model are showing signs of failing.

The governance and learning function

Shown in the architecture diagram as a spanning vertical element, the governance and learning function is not a separate system — it is a practice that runs through all five layers. It closes the loop that conventional project controls leaves open. A programme that maintains a living decision record — accessible, structured, and linked to the monitoring indicators that track whether the reasoning has held — is a programme that learns systematically from its own choices.

Sequencing the build

For an organisation starting from a functioning but conventional project controls base, the recommended sequencing is as follows. Invest first in Layer 2 — the data integration work that most organisations have deferred. Without it, the layers above will be populated with inconsistent or stale evidence. Build the first Layer 4 model manually, around a live decision, before investing in tooling. Introduce Layer 3 analytics incrementally, starting with quantitative risk analysis before adding predictive forecasting and AI-assisted enrichment. Build the Layer 5 interface around the needs of the specific audience — programme director, sponsor, assurance body — rather than as a generic reporting product.

The governance and learning function should be established from the first decision modelled, however modestly. A two-page decision record maintained from the outset is worth more than a sophisticated record-keeping system introduced after three years of undocumented choices.

Figure 6 — Recommended build sequence. Data integration should precede tooling investment. The governance function starts at Phase 1, not Phase 4.

Section 8

Governance, assurance, and the decision record

Major project teams are frequently asked, often years after the fact, to account for the decisions they made. Why was a rebaseline accepted? What alternatives were considered when the programme fell behind schedule? What did the team know about commercial exposure when they chose to accelerate? These questions arise in public inquiries, parliamentary scrutiny, infrastructure assurance reviews, and post-investment evaluations. They are not exceptional — they are a routine feature of major programme governance.

In most organisations is that this information is partially recoverable from board papers, minutes and email chains — but the reasoning behind choices is rarely captured in a structured, accessible form. What was known is documented, imperfectly. What was considered is partly recoverable. What was rejected, and why, is almost never recorded at all.

A DDT creates a decision record as a natural by-product of the decision process, rather than as a bureaucratic obligation imposed upon it. The decision model captures what was known at the time of the choice: what options were on the table, what causal pathways were identified, what consequences were anticipated, what trade-offs were accepted, and what monitoring indicators were established to track whether the underlying assumptions proved correct.

It is the decision process made legible — an artefact that reflects what actually happened in the analytical work behind the choice, rather than the sanitised version that appears in board minutes. For organisations subject to public scrutiny, statutory reporting requirements or infrastructure assurance regimes, this constitutes a real governance advance.

A decision record that captures the assumptions behind a choice also creates an early warning mechanism. When circumstances change — when the forecast that underpinned a resequencing decision proves incorrect, or when a risk that was judged manageable crystallises — the decision model identifies which assumptions have been violated and which monitoring indicators should now be escalating. Beyond the pre-decision stage, the DDT is a mechanism for adaptive management throughout delivery.

Figure 7 — The decision record lifecycle. Together the three stages create an auditable account of the reasoning behind a major programme choice — and an early warning system for when that reasoning is under strain.

Section 9

Project controls as the decision function

The argument in this paper is ultimately a professional one as much as a technical one. Project controls has spent a generation building the data infrastructure of major project delivery. It now holds the cost data, the schedule data, the risk data, the benefits data and the commercial intelligence that major programmes run on. That is a considerable base to build on. The question is what it is used for.

In many organisations, the answer remains largely observational. Project controls reports, forecasts, visualises, integrates and alerts. It tells programme leadership what is happening and what is likely to happen. This is valuable. It is also incomplete. The executive question on a major programme is not primarily “what is the forecast?” It is “what do we do?”

The DDT provides a route from the former to the latter. Rather than replacing the capabilities the profession has built, it makes them more valuable — by connecting cost data, schedule data, risk data and benefits data, through a causal decision model, to the specific choices that determine whether a programme succeeds or fails.

That connection is not theoretical. The reference architecture described in this paper — five functional layers running from data foundations through integration, analytics, decision modelling, and decision interface, held together by a governance and learning function — is buildable from components that most mature project controls functions already have in some form. The sequencing question is not whether to start, but where: typically with data integration, then a manually built decision model around a live decision, then progressive analytical enrichment as confidence and capability grow. Organisations that begin with a structured decision workshop and a two-page decision record are already operating the first version of this architecture, even without knowing it.

The natural home for this capability is the project controls function — which already owns the evidence, territorial claims aside. The profession has the data literacy, the analytical infrastructure, and the programme proximity to build and operate decision models. What has been missing is the conceptual frame that organises that evidence around decisions rather than artefacts.

Providing that frame — and demonstrating what it produces in practice — is the contribution this paper has attempted to make. Project controls has digitised the project. The next step is to digitise the decision.