The problem is not ambition. It is architecture.
Most construction businesses that engage us have already invested in AI in some form. They have a digital estimating tool that cannot read from their cost management system. They have a scheduling platform generating forecasts from data that is three weeks old. They may have deployed a generative AI tool across their commercial team that produces plausible-sounding outputs disconnected from any live project data.
None of these are technology failures. They are sequencing failures — the predictable result of procuring AI capability before the underlying data architecture, systems integration, and process design are in place to support it. Without those foundations, AI does not make a construction business more efficient. It makes it more expensively inefficient.
A credible AI consultancy roadmap for a construction business does not start with technology selection. It starts with a rigorous analysis of the data environment, the commercial processes sitting on top of it, and the specific points at which both are losing the business money.
Why construction’s data environment makes this harder than it looks
Construction is not a single-system business. A typical mid-tier contractor is operating across five to fifteen discrete platforms simultaneously: an estimating tool such as Conquest or COINS, a project management system such as Procore or Viewpoint, a scheduling engine such as Asta Powerproject or P6, a finance and ERP system, a document management platform, field reporting applications, procurement tools, and often a layer of bespoke spreadsheets sitting across all of them filling integration gaps the platforms were never designed to close.
Each of these systems holds valuable data. None of them were architected to share it. The result is what systems engineers call a fragmented data estate: information exists in abundance, but it exists in silos. Commercial data, programme data, procurement data, and site data have no common data model, no shared taxonomy, no automated flow between them. Reconciling them manually is what consumes 30 to 40 per cent of a skilled commercial team’s working week — time that should be spent on margin protection and delivery judgement.
“Before any AI model is trained, deployed or procured, a construction business needs a clear inventory of its data estate: what systems generate what data, how clean and consistent that data is, where the gaps and duplications are, and what middleware or API layer is required to create a unified data model. This work is not glamorous. It typically takes four to eight weeks and involves interrogating systems that are poorly documented and occasionally contradictory. Most organisations underestimate it by a factor of three.“
The five stages of an AI roadmap that actually delivers
Stage 1 — Commercial diagnostic and data estate mapping
The first stage is not a technology workshop. It is a structured commercial diagnostic: a systematic review of estimating accuracy against final account, cost overrun patterns by project type and phase, reporting cycle times and the manual effort behind them, change order capture rates, and procurement spend against benchmark. Also a thorough understanding of SOP’s and how the work is done. This diagnostic surfaces where margin is being lost and at what stage of the project lifecycle.
Running in parallel is a data estate mapping exercise. Every system that touches financial, programme or site data is inventoried. Data flows — or the absence of them — are mapped. The quality of data in each system is assessed: consistency, completeness, latency, and whether it carries the commercial metadata needed to make it useful for AI applications downstream.
Together, these workstreams produce a clear picture of where the highest-value AI interventions lie — and what the data infrastructure requirements are to support them.
Stage 2 — Opportunity prioritisation and ROI modelling
Not every identified inefficiency warrants an AI solution. The second stage takes the diagnostic outputs and applies a commercial filter: which interventions will deliver the greatest return, in the shortest timeframe, against the least implementation complexity?
This prioritisation is structured around three dimensions. First, the financial impact — how much margin, time or cost certainty does addressing this inefficiency recover? Second, the data readiness — does the data required to support the proposed AI application already exist in a usable form, or does it require significant remediation first? Third, the organisational readiness — does the business have the process discipline and governance structure to adopt the solution, or does that need to be built alongside it?
The output of this stage is a prioritised opportunity register, with a profitability case for each initiative, a realistic implementation timeline, and a clear statement of the prerequisites — data, process, and organisational — that must be in place before each intervention can be deployed.
Stage 3 — Solution architecture and technology selection
With a prioritised set of opportunities and a clear view of the data environment, the third stage designs the technical architecture required to deliver them. This is where technology selection happens — but it happens in the context of a defined problem and a defined data model, not as a procurement exercise driven by vendor marketing.
For most construction businesses, the architecture will involve some combination of: middleware or API orchestration to connect existing systems (typically using tools such as Azure Logic Apps, MuleSoft, or custom REST API layers); a data warehouse or lakehouse to create a unified commercial data model (commonly on Azure Synapse, Databricks, or BigQuery); AI and ML models trained on the business’s own historical project data for specific applications such as cost forecasting or schedule risk prediction; and a reporting and visualisation layer that surfaces insights to the right people at the right moment.
The specific combination depends entirely on the existing technology landscape, the data readiness of each system, and the priority of the identified opportunities. There is no generic architecture that works across all construction businesses.
Stage 4 — Engineering and integration
At Surtori, the engineering stage is not a separate engagement. It is the continuation of the same one. Our engineers build the integration layer, configure the data pipelines, develop and validate the AI models, and deploy to production. Every system connection is tested against real data from the client’s environment. Every model is validated against historical outcomes before it goes anywhere near live project decisions.
The complexity here should not be underestimated. Integrating a legacy estimating system with a cloud-based data warehouse, whilst preserving data integrity across different project taxonomies and cost codes, whilst accommodating the way different project teams have historically structured their data — this is not a task that benefits from an inexperienced hand. Getting it wrong creates data quality problems that are significantly more expensive to unpick than the original fragmentation.
Stage 5 — Embedding, governance, and measurement
Production deployment is not the end of the roadmap. The fifth stage establishes the governance framework that ensures AI outputs are trusted, used, and continuously improved. This includes: model monitoring to detect performance drift as project data evolves; data quality controls at ingestion to prevent garbage-in-garbage-out degradation; user adoption frameworks that integrate AI tools into existing workflows rather than sitting alongside them; and a measurement regime that tracks the commercial outcomes committed to at the outset.
What the evidence shows
The productivity gains from properly executed digital transformation in construction are documented across McKinsey, BCG, Gartner, PMI, Deloitte and PwC. Planning time reductions of up to 60 per cent. Project delay reductions of up to 35 per cent. Estimating and project controls efficiency improvements of up to 40 per cent. Cost management enhancement of up to 22 per cent.
These are not theoretical. They are outcomes achieved by construction businesses that followed a structured, integrated approach — and that had partners who could engineer the solution rather than describe it. The difference is not access to better technology. The technology is available to everyone. The difference is in the architecture, the data engineering, and the implementation discipline.
Surtori’s Digital & AI Opportunities Assessment is a fixed-price engagement that delivers the diagnostic and opportunity register described in Stages 1 and 2 above. The output is a board-ready decision pack with prioritised initiatives, profitability cases, and a fully costed architecture proposal. No commitment to what comes next until you have seen the evidence. Begin here: Digital & AI Opportunities Assessment.
ALSO IN THIS SERIES
→ Construction AI consultancy: how to save time, cut costs and work smarter
→ AI strategy for construction firms: why getting the foundation right makes all the difference
→ The role of construction AI advisory services in enabling growth
RELATED READING
→ The 3% Solution: why your construction tech stack is bleeding margin
→ Why 90% of digital strategies fail — and how the Fusion Framework changes the equation
→ The Consulting Gap: why traditional firms struggle in the modern era of transformation