If work is being lost between the first phone call and the quote, the missing thing is a shared system of record. A CRM is one way to get that. AI on top of what you already run is another. The deciding question is how many people need the same memory.
Key takeaways
Almost every contractor who asks this question is describing the same symptom: an enquiry came in three weeks ago, somebody meant to follow up, and nobody is quite sure whether it was quoted, lost or forgotten. That is not a software problem yet. It is a memory problem, and the honest first question is whose memory is supposed to hold it.
If the answer is "mine, and there is only me", a disciplined inbox with a follow-up rule genuinely can be enough, and AI can make it better without buying anything new. If the answer involves two or more people — an estimator, an owner, whoever answers the phone — then you need a shared record, and the argument for a CRM starts.
A CRM is a database with a discipline attached. It holds contacts, opportunities, activity history and a next action for each, and it makes that history visible to more than one person. Nothing about a language model provides that. A model can draft an excellent follow-up email; it does not remember that you sent one in March unless something recorded it.
That distinction is the whole decision. AI is a capability layered onto a system of record. If there is no system of record, adding AI produces faster individual actions and the same lost opportunities.
It is also worth separating a CRM from project software. Tools like Procore manage work you have already won — a company directory of people and companies you work with, and prime contracts once there is a contract to manage. That is a different question from tracking the eleven opportunities you have not won yet.
When a shared system of record has become necessary
Two people answer enquiries. The moment a second person can take a call, the history has to live somewhere both can see, or you will call a client twice and another one never.
You cannot answer "what is outstanding". If listing every open quote requires searching an inbox, you are already losing work you never learn about.
Repeat and referral work is the majority of your revenue. Long gaps between jobs make personal memory unreliable at exactly the moment it matters most.
Absent all three, the honest advice is to defer the purchase and fix the process. A single shared spreadsheet with owner, next action and next date, reviewed every Monday for ten minutes, has rescued more contracting pipelines than any subscription.
Drafting the follow-up. The reason follow-ups do not happen is rarely disagreement about whether they should. It is that writing the fourth one this week is tedious. A model drafting from the actual thread, in your voice, removes that friction — and you still read it before it goes.
Summarising a long thread. Twenty-two emails about a kitchen renovation compress into a paragraph of what was agreed, what is outstanding and what the client last asked. Useful before a call, and more useful when a colleague picks it up.
Extracting structure from unstructured enquiries. An enquiry arrives as three paragraphs of prose. Pulling out address, scope, timeline and budget signals into fields is exactly the extraction work document models are built for — the same class of capability documented for structured fields in Microsoft's Document Intelligence and across Google's Document AI processors.
Answering after hours. An assistant that acknowledges an enquiry, asks two qualifying questions and books a site visit is genuinely valuable in a trade where the first responder often wins. Keep it to acknowledging and scheduling — not quoting, and not committing you to a date you have not checked.
This is the part contractors most often skip, and it applies to a two-van operation as much as to a large firm.
Commercial electronic messages. Canada's anti-spam legislation governs commercial electronic messages, and the CRTC's spam and malware pages are the regulator's own starting point. Automation makes volume easy, which is exactly why consent and unsubscribe handling need to be right before you scale sending. Our sister firm covers the practical detail in CASL email marketing compliance and the distinction that catches people out in implied versus express consent.
The contact data itself. Names, addresses, phone numbers and notes about identifiable people are personal information. The Office of the Privacy Commissioner's summary of PIPEDA requirements sets out the obligations, and our sister firm addresses the small-business angle in whether PIPEDA applies to your business and customer data collection for small businesses.
A specific caution about AI: if you are pasting client conversations into a general-purpose assistant, you are moving personal information into a third-party system. That is a decision to make deliberately, with a look at where the data goes and whether it is retained, rather than by habit.
Choosing between the three options
One person, low volume, mostly referral: shared spreadsheet plus a weekly review, with AI drafting follow-ups. Cost: nothing. Revisit when a second person starts answering the phone.
Two to ten people, mixed residential and small commercial: a general CRM you will actually maintain, configured simply — contact, source, next action, next date. AI for drafting and summarising on top.
Bidding institutional or larger commercial work: something that tracks bid invitations, submission deadlines and outcomes, because the pipeline shape is different and the deadlines are external. Project software handles what happens after the award, not before.
Any of the above: whatever you choose must let you export contacts and history in an open format. The lock-in risk is real and it compounds.
A four-person renovation contractor, illustrative
The complaint. "We need a CRM." What was actually happening: enquiries arrived in a shared inbox, the owner quoted the ones he remembered, and nobody knew what happened to the rest.
What they did first. One shared list. Every enquiry logged with source, next action and next date within a day of arriving. Ten minutes every Monday reviewing it.
What AI added. Extraction of address, scope and timeline from each enquiry into the list, and drafted follow-ups for the owner to review. Both saved real minutes, and neither was the fix.
The result. Within two months the pattern was visible: enquiries from one referral source converted at several times the rate of the rest, and roughly a third of enquiries had historically received no second contact at all.
Then they bought a CRM. By that point they knew exactly which four fields they needed, which made the choice fast and the setup honest. Buying first would have produced an expensive, empty database.
It depends on whether you need the construction-specific parts — bid invitations, plan rooms, takeoff links. For residential and service work, a general CRM you will keep current beats a specialised one you abandon. The failure mode is always abandonment, not features.
It can classify and remind, which is genuinely useful. It cannot be the record, because a record has to be something you can query, correct and export — and it should not decide who gets chased and who gets dropped. That prioritisation is a commercial judgment about your business.
Only for work you have won. A project system manages contracts, directories and delivery. If your problem is opportunities going cold before they become projects, that problem lives upstream of it.
Approach with care. Commercial electronic messages are regulated, and a bulk reactivation campaign to contacts whose consent basis you cannot demonstrate is exactly the scenario CASL addresses — the CRTC's guidance and our sister firm's note on implied versus express consent are the place to start before you press send.
A 30-minute call is enough to tell you whether AI pays for itself here.
