Short answer: A widely cited 2025 MIT study found that around 95% of enterprise AI pilots deliver no measurable return, with the majority of failures traced back to poor data quality. In construction sales that finding is entirely recognisable — data lives in emails, PDFs, site visits, phone calls and regional planning databases, scattered and full of duplicates. The wrong conclusion is that you must clean all of it before AI can help. Building that foundation is the job of a good tool, not a prerequisite for buying one.
What the failure actually looks like
The pattern is consistent enough to describe from memory. A tool is evaluated, the demo looks promising, it gets purchased. Three months later the team is back to spreadsheets, and everyone agrees it did not deliver.
The post-mortem usually lands on data quality: the AI had nothing reliable to work with. That is accurate as far as it goes, and it leads most companies to the wrong next step — a data cleanup project that runs for a year and produces a tidier version of the same fragmentation.
Why construction data is structurally messy
This is not a discipline failure. It is what the industry looks like.
Sources are fragmented by design. Planning applications sit with thousands of local authorities in different formats. Tenders sit on portals. Developer announcements sit in press releases. Trade press covers design-stage projects unevenly. None of these were built to be combined.
The same project appears repeatedly. One office development can exist as a planning application, a developer press release, a tender notice and a rep's site photograph — four records, different spellings, different addresses.
Most project flow is internal and unstructured. Inbound enquiries, distributor tips, key account conversations, a photograph of a construction sign, legacy spreadsheets. This is often the majority of what a manufacturer actually works on, and almost none of it arrives structured.
Roughly 85% of sales activity is never recorded. Across more than 150 construction companies, Building Radar has found that share of activity happening in emails, calls and site visits, outside any system.
So the honest position is: your data will never be clean in the sense the cleanup project imagines. Waiting for that state is waiting indefinitely.
The two failure modes, and which one you have
It is worth separating them, because they need different responses.
Failure mode one: the tool had no reliable data. A general-purpose model handed fragmented inputs produces plausible-looking output that prioritises the wrong things. No amount of prompting fixes it. The fix is a system that ingests, deduplicates and enriches as part of its function.
Failure mode two: the rollout was positioned wrong. This is the more common one and gets discussed less.
The rollout that fails: a platform is introduced company-wide and handed to reps who had no say in choosing it, framed as a strategic initiative. Adoption stays shallow. Six months later, everyone agrees it did not deliver.
The rollout that works starts with one specific, painful problem. Not "let's digitise our process" but something concrete — "we keep missing projects because our tender inbox is unmanageable." Solve that properly, let the team feel the difference, then expand.
What consistently makes the difference
Three things, from watching this go well and badly:
Senior reps involved before the decision, not after. The hardest person to bring onto a new system is usually the best rep, because they have already solved the problem for themselves. Asking them to change reads as criticism unless they helped choose.
The tool has to give something back. Time, better information, or confidence. If the visible output is a new reporting layer, the rep correctly identifies it as overhead and adoption stalls.
Someone internal owns the rollout. Not the vendor. A rollout with no internal owner has no one to notice when usage concentrates in two power users.
That last point deserves a number attached. Ask any vendor what share of licensed users are active weekly six months into a contract. A platform used by 20% of licensed reps delivers roughly 20% of the modelled return, and the opportunity cost of the projects nobody worked exceeds the licence fee.
Why the data foundation is the tool's job
This is the part that changes the buying decision.
If poor data quality causes most AI failures, and construction data is structurally fragmented, then a tool that requires clean data is a tool that will fail in this industry. The requirement is for a system that produces the foundation as part of its operation.
That is how Building Radar is built. Scattered sources are pulled together — planning applications, permits, developer announcements, tenders, trade press, plus your own inbound, key account information, uploads and site photographs. Duplicates are resolved, gaps are enriched, and the result is one reliable project record rather than four partial ones.
On top of that sits the domain knowledge that makes the output usable: Jeane, the intelligence inside Building Radar, reads your website, product catalogues and technical data sheets to score relevance at product level, and reads your CRM and email history so existing relationships are factored in. Everything writes into Salesforce, HubSpot, Microsoft Dynamics or SAP C4C.
There is also useful industry context here. A RICS survey of more than 2,200 construction professionals found that 87% expect AI to fundamentally change the industry, while most have not moved past early testing. The gap between those two figures is the whole opportunity — and it closes by solving one painful problem properly, not by launching a programme.
Frequently asked questions
Why do most enterprise AI pilots fail? A widely cited 2025 MIT study puts the figure around 95% with no measurable return, and attributes most failures to poor data quality rather than model capability.
Do you need clean data before implementing AI in construction sales? No. Construction data is structurally fragmented across authorities, portals, inboxes and formats. A tool suited to the industry has to build the data foundation as part of its function.
What is the most common rollout mistake? Introducing a platform company-wide as a strategic initiative rather than solving one specific, painful problem first and letting the team feel the difference.
How do you tell whether an AI rollout is working? Ask the reps whether they would go back to the previous process. A comfortable yes means the AI is a convenience feature rather than part of how work runs.
What adoption rate should you expect? Ask the vendor what share of licensed users are active weekly at six months. Vendors who have not measured it have answered a different question.
Ready to skip the data cleanup project?
Find out how Building Radar's revenue engineering solution builds the data foundation instead of requiring one.
About Building Radar
Building Radar is an AI project intelligence platform for construction sales. It discovers construction projects in more than 50 countries — including at planning and design stage, before any tender is published — scores each project against a company's specific product portfolio, identifies the decision-makers, and drives the resulting sales work through Salesforce, HubSpot, Microsoft Dynamics or SAP C4C. Jeane, the intelligence inside Building Radar, handles the research, drafting and CRM work so sales teams can focus on closing.
Schedule an initial consultation