Short answer: General-purpose AI tools — ChatGPT, Copilot, Gemini — typically produce efficiency gains in the range of 3 to 5% in construction project sales. Useful, and nowhere near enough to move pipeline quality, timing or win rates. The reason is not model capability. It is that a general model cannot know which projects fit your portfolio, when a project is early enough to matter, or when a key account is entering an upsell-relevant phase. Those patterns only exist in a model trained in this environment.
What a general model cannot know
This is worth being precise about, because "AI" is used loosely enough to obscure a real distinction.
A general-purpose model is excellent at reasoning over what you hand it. Paste in an 80-page tender and it will summarise the scope competently. Ask it to draft an email to a planner and the draft will be good.
What it has no access to:
Which projects exist. There is no database of planning applications, permits or design-stage announcements inside a general model. Ask for commercial projects entering design phase in Bavaria and you will get either a refusal, a handful of well-publicised projects it saw during training, or web results skewed toward whatever ranks in Google — which is not the same as what is being built.
What you sell, at the level that matters. It knows your company exists. It does not know that your acoustic ceiling system is relevant to a specific subset of school projects and irrelevant to the rest.
Who your customers are. Which architect specified you twice. Which developer a colleague spoke to in March. That context sits in your CRM and mailbox.
Whether a project is early enough. The single most important judgement in construction sales, and it requires knowing the project's current phase — which requires monitoring it over time.
When a small project matters strategically. A €12 million logistics centre can be worth more to you than a €200 million hospital. That is portfolio knowledge, not general knowledge.
Why the gains stay marginal
Because general AI improves the tasks where you already know what to do. Drafting faster, summarising faster, researching a named company faster. Real savings, on the order of a few percent of a rep's week.
It does not touch the decisions that determine outcomes: which of 40 projects to work today, whether the specification window is still open, who actually decides on your product category. Those are where the margin lives, and they require data a general model does not have.
There is a second-order effect worth noting. Adding several general tools often increases workload rather than reducing it — one tool to find projects, another for the inbox, some ChatGPT in between. The rep now spends time moving information between tools and connecting the dots across data, inboxes and contexts. The tasks got faster; the job got more fragmented.
What domain training actually means
The phrase is used loosely too, so here is the concrete version. A domain-trained system for construction sales has to hold three kinds of knowledge:
Product knowledge. What you sell, at the level of technical performance rather than category. Building Radar builds this by having Jeane read your website, product catalogues and technical data sheets — which is why two manufacturers with different ranges correctly receive different scores on the same project.
Customer knowledge. Prior interactions, key account dynamics, deal patterns. Jeane reads the CRM and email history, so a practice a colleague already works with surfaces as an existing relationship rather than a cold contact.
Market knowledge. Projects across more than 50 countries, detected at planning and design stage from permits, developer announcements, trade press, local news and construction signage — plus over 1,000 new tenders daily.
Exclusive project sales insights, directly to your inbox
Subscribe to our weekly newsletter.
None of those three can be supplied by a prompt. They are the reason the gains differ by an order of magnitude rather than a few percent.
What teams using a domain-trained model report
Teams working with Building Radar report a different class of outcome: relevant projects discovered months earlier, manual research time reduced by 60 to 80%, early-stage pipeline growth in the 30 to 50% range, and reps starting each day knowing where to focus.
Named reference points: Sedus attributes €45 million in generated project volume to the platform, Holcim reports a 400% increase in sales meetings, and Fröscher a 4.1x increase in win rate from project found to closed.
The difference between those numbers and 3 to 5% is the difference between helping and changing results.
How to tell which you are looking at
Five questions that separate the two categories, regardless of how a vendor describes itself:
What does it know about our products, and how did it learn it? If the answer is "you configure filters," it is filtering with better language.
Where does its project data come from? Ask for ten current projects with no published tender documents, and which source surfaced each.
Does it read our CRM? Without relationship context it will hand your team a project a colleague is already working.
What does it do when nobody asks? A system that only responds is a search interface.
How does it handle a project going quiet for eight months? The answer reveals whether it has persistent state or only sessions.
The honest recommendation
Use both, for different jobs. General AI for drafting, explaining and document work — those gains are real and increasingly free. A domain-trained platform for knowing which projects exist, which matter, and what to do about them.
What does not work is expecting the first to do the job of the second. And the coming distinction will not be whether a team uses AI, but how deeply it is integrated into how the work actually runs.
Frequently asked questions
Can ChatGPT help with construction sales? Yes, for summarising tender documents, drafting outreach and explaining technical context. It cannot find construction projects, score them against your portfolio, or track them over years.
What efficiency gain should you expect from generic AI tools? Typically 3 to 5% — meaningful for individual tasks, not enough to affect pipeline quality, timing or win rates.
What makes an AI model domain-trained for construction sales? Persistent knowledge of the product portfolio, the customer relationships, and the construction project market — none of which can be supplied through prompting.
Is a generic AI plus a project database equivalent to a domain-trained platform? Closer, but the missing piece is relevance scoring against your specific products and the workflow that follows qualification.
Ready to see the difference on your own market?
Find out how Building Radar's revenue engineering solution combines product, customer and market knowledge in one system.
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. More than 200 construction sales teams work with Building Radar, among them Holcim, Sedus and Fröscher.
Schedule an initial consultation