Getting started
Video · 6 min read
The short answer
Watch: Where should a construction business start with AI?
Watch the explanation, then explore the practical implications below.
Start with a business constraint
Individual AI experiments can be useful while the overall process remains slow. A quotation drafted quickly still cannot leave if the survey is incomplete, the rates are disputed or nobody can approve an exception. The first task is to identify what prevents a piece of work from reaching a useful conclusion.
Choose a recurring workflow with a clear beginning and end. Quoting, tender handovers or scope-change reviews can provide a bounded starting point. Follow an actual recent example with the people responsible, including the steps that happen in email, calls and spreadsheets.
Look for search, retyping, waiting and checking
Search means people hunting for the latest record or asking a colleague where it lives. Retyping means the same information moving manually between forms, documents and systems. Waiting exposes missing inputs, unclear ownership or approval queues. Checking shows where judgement is essential and where avoidable inconsistencies create work.
Record who does each activity, what they need and what happens when the input is missing. Separate time spent doing the work from elapsed time waiting for someone else. This makes the constraint visible before choosing a tool.
Prioritise value and readiness together
Assess commercial impact, capacity, dependence on key people and the cost of leaving the problem unresolved. Include the effort needed to prepare information, train users and review outputs. A process with high potential may need basic ownership or data decisions before implementation.
For a quotation workflow, agree how you will measure improvement: time from complete requirements to an approved quotation, review effort, missing-information requests and corrections. Compare similar jobs and include the new checking work in any estimate of time saved.
Make the information dependable for one workflow
Data first does not require cleaning every record in the company. It means agreeing which information the selected workflow can rely on, who owns it and how changes are approved. For quoting, this could mean current specifications, project references, survey evidence and an approved rate card.
If two teams use different rates, first resolve which rates apply and who can authorise exceptions. Connecting both versions to AI would make the disagreement easier to distribute. Keep the original evidence available to the reviewer.
Choose the simplest approach that fits
Begin by checking what existing software can already do. Use a straightforward automation for predictable transfers and routing. Use AI where interpreting varied information or preparing a draft adds value. Consider a custom system when important work crosses tools and needs company-specific decisions, permissions or approvals.
A connection is only part of an implementation. Access rights, field mappings, project identifiers, update frequency and failed transfers need to be understood. The choice should follow the workflow and operating responsibilities.
Define a first implementation the team can assess
Give the process a business owner, agree the scope and document what a better result looks like. Test with representative records, including incomplete surveys and changed requirements. Decide when the system should stop and ask for help.
Keep high-consequence commitments with authorised people. The system can gather evidence, prepare a scope and organise a review queue; the responsible person approves the commercial decision. Review results with users before expanding to another workflow.
Questions for your leadership team
Which recurring piece of work gets stuck?
Bring one recent example and trace it from the first request to the finished result. Where did the team search, retype, wait or check?
What outcome matters commercially?
Identify the baseline, the cost of the delay and the measure that would show the process is improving.
Which information can we trust?
Name the owner of the relevant rates, scope, specifications and project references. Resolve competing versions before connecting them.
What can our existing tools already do?
Check native features and simple automations before assuming a custom AI system is needed.
Who owns the first implementation?
Agree who can change the workflow, approve exceptions, review results and decide whether to expand.
Full video transcript
The full video transcript is included below for reference.
Video transcript
Picture Monday morning in a growing construction business. One person is asking ChatGPT to draft a quotation. Another is explaining the company to copilot. A third has built a small app to organize project updates.
There is plenty of activity, but the estimator is still waiting for the survey. The commercial manager is still checking the rates, and everyone still needs the same experienced colleague to explain what happens next. So where should you start with AI? Start with the work that keeps getting stuck.
At CSM, we look for four things, where people search, retype, wait, and check. Then we follow one real piece of work from the first request to the finished result. Take quoting. Where do the requirements arrive?
Who decides what is included? Where are the approved rates? What happens when the survey is incomplete?
Those questions tell you far more than a list of AI tools.
Next, prioritize.
We look at commercial impact, time and capacity, dependence on key people, the risk of doing nothing, and whether the business is ready to make the change safely. A process can be valuable and still need groundwork. If nobody agrees which rate card is correct, connecting an AI model will not resolve that disagreement.
That is what we mean by data first.
For AI to become dependable across the business, the information behind it needs an owner, a clear meaning, and an agreed use.
You can begin with the information needed for one workflow. You do not have to clean the entire company first, then choose the right approach. An existing software feature might solve it. A simple automation might move the information.
AI might help interpret a survey or draft a scope.
A custom system makes sense when the important work falls between those tools and needs your company’s rules and approvals.
Through building CoreOS we have established API connections with over 3,000 software tools. That gives us a head start in connecting what a business already uses. Each implementation still needs the right access, mappings and checks.
Our first recommendation is simple: pick one recurring process, give it an owner, agree what a better result looks like, then work backwards through the information and decisions required.
Keep high consequence decisions with the people authorized to make them. Let the system prepare the evidence and remove the chasing around those decisions. The opportunity is to give your business more capacity without adding the same amount of coordination every time you grow.
If you want help finding that first process, book a call with CSM. Bring one piece of work that repeatedly slows your team down. That is enough to start.
Start with one process that matters
Bring a workflow where delays, repeated work or unclear decisions are limiting the business. We can explore the outcome you want, what needs to change and where AI could help.