AI Receptionist for Construction Companies: Real Costs and Call Routing

An AI receptionist costs a construction company about $120 to $500 a month at 250 calls, depending on how the vendor counts. Route by caller type: new enquiries booked by the AI, supplier and subcontractor calls passed to the right person with a note, and anything urgent on site sent to a human at once, never to a message.

A builder's phone is mostly not customers. Subcontractors report delays, merchants confirm deliveries, the site team needs something now, and an inspector wants access. A receptionist built around "book the appointment" treats all of them the same, which is the usual failure. The fix is a rule per caller, written before you switch it on, and the price follows from how many of those calls you let the AI take.

Follow me on Instagram@sagnikteaches

Who rings a building firm, and what each caller needs

Here is an illustrative month for a twelve-person firm with two live sites: 250 calls, of which the office answers 145 itself and 105 reach the receptionist because staff are on site, on another line, at lunch or off shift.

Connect on LinkedInSagnik Bhattacharya

Replace those figures with your own phone log, a month at a time. The pattern to look for is how many of the calls that reach a machine are not new business. In this mix 55 of the 105 come from clients, subcontractors, suppliers, the site team and neighbours, another 14 are sales or wrong numbers, and only 36 are new enquiries. A receptionist that cheerfully offers the 55 a "consultation slot" would annoy every one of them.

Subscribe on YouTube@sagnikteaches

A routing rule for each caller type

The opening line should sort callers in one question, in plain words, before the AI does anything else. A greeting that works: "Thank you for calling [company]. This is an AI assistant and the call may be recorded. Are you calling about a new project, an existing project, as a supplier or subcontractor, or is something urgent on site?" Each answer leads to a different rule.

NEW PROJECT: ask for name, number, site address, type of work, rough size,
  timing, who the client is (owner, landlord, developer). Offer a site visit
  or call-back slot from the calendar. Never quote a price or a start date.
EXISTING PROJECT: ask which project and who they are. Take a message for the
  project manager. Confirm a call-back window. Never give dates, costs or
  statements about defects or variations.
SUBCONTRACTOR: ask name, trade, site, and the issue (late, no access, schedule,
  payment). If they are at a site today and the issue is access or safety,
  transfer to the site manager. Otherwise take a structured message.
SUPPLIER: delivery today = transfer to the site manager; account or quote =
  message for the office; invoices = message for accounts. Never agree to
  leave materials unattended.
URGENT ON SITE: see the escalation ladder. Transfer first, summarise after.
SALES / SPAM: thank them, say we don't take sales calls, end the call.

Two realistic mistakes show why the supplier and subcontractor lines are so specific. In the first, a delivery driver outside a site is told "the site manager is unavailable, you can leave the pallets by the gate", because the receptionist wanted to be helpful. In the second, a subcontractor's message that he can't start Monday sits in an inbox until Wednesday, because it was filed with the sales leads. The rule sheet fixes both: deliveries transfer, schedule problems go to a named person, and nothing is agreed on the firm's behalf. The wider pattern of chasing subcontractors is covered in AI for building contractors: chasing subcontractors and progress updates, and the site-side version of this problem, taking calls while you are on a job, is in the tutorial on AI receptionists for builders taking calls on site.

The existing-client rule deserves a before and after, because it is where a friendly assistant does the most damage. Before: the client asks when the kitchen will be finished and the AI replies, "Your project is on schedule and should finish by the end of the month." That is a date the firm never gave, and it will be quoted back at the next meeting. After: "I can't see the programme from here, so I won't guess. I'll pass your message to [project manager] and she will call you by 3 pm today. Is this the best number to reach you?" The second answer gives the client something concrete, a named person and a time, and promises nothing the firm doesn't control.

Every message the receptionist takes should arrive in the same shape, or the project manager gets a paragraph of chat to decode between site visits. A filled-in example from the subcontractor branch, illustrative:

SUBCONTRACTOR MESSAGE 15:20
From: [first name], plasterer, [number]
Site: B, second-floor flat
Issue: can't start Monday; boards not delivered
Wants: a new start date
Urgent: no (no one on site waiting)
Call back by: 10:00 tomorrow

Each line is a question the receptionist asked, in the order of the rule sheet. When a caller doesn't answer one, the message says "not given" rather than filling the gap, and the call-back time comes from your rules, not from the AI's sense of what is reasonable. Ask the vendor whether these fields can go into your job software as a task or only arrive by email, because a message that has to be retyped tends to wait.

Regular subcontractors and suppliers deserve a warning before the switch, since in the month above they made 100 of the 250 calls, more than clients did. A short text, sent a week ahead, does it:

From Monday our main line is answered by an AI assistant when the office
is busy or closed. For anything on site today (access, deliveries, safety),
ring [site manager] on [mobile]. For schedules and invoices, the assistant
will take a message and [project manager] will call you back the same day.

That one message moves the most time-critical calls off the main line altogether, which keeps them out of the count on per-call and per-minute plans and leaves the receptionist with the calls it is suited to.

Urgent site issues: triggers, ladder and timeouts

Don't ask an AI to judge urgency from tone. Give it a list of triggers and tell it that if unsure it treats the call as urgent, because a false alarm costs a minute of someone's time and a missed one can cost much more.

• Someone is hurt, or a serious near miss has just happened.
• Fire, a gas smell, a struck cable or a burst main.
• Any movement, cracking or collapse in a structure, scaffold or excavation.
• Flooding that is spreading to a neighbouring property.
• Theft, a break-in or vandalism on a site holding plant or materials.
• An inspector, officer or police on site asking for the site manager.
• A neighbour reporting damage or a hazard to the public.

The first sentence of the urgent branch is fixed: "If anyone is hurt or in danger, please hang up and call the emergency services now. I'm alerting our site manager at the same time." Then the ladder runs. Try the site manager's mobile for 20 seconds, then the project manager, then a director. If nobody answers, the AI sends a text and an email with the caller's number and a summary, and says someone will ring back within ten minutes. The tutorial on setting up an AI answering service that never misses an emergency goes further on timing and back-up numbers.

A ladder is only as good as the vendor feature behind it. Rosie's Growth plan lists "waterfall transfers" that try several numbers until someone answers, and Goodcall's FAQ says its agent retries a busy line and moves on to the next escalation contact, up to three. Smith.ai routes callers to team members with full context on every AI plan, Dialzara offers blind transfer on every plan and warm transfer from Pro, Quo's Sona transfers only on a Business seat, and Jobber's Receptionist sends urgent keywords to a single escalation contact, by transfer or text alert. Test the ladder with the first number switched off. If the vendor can't show the second attempt connecting, it isn't a ladder.

Write the ladder down as a one-page sheet, because the person who edits the receptionist's instructions may not be the one who knows the rota. A filled-in version for the twelve-person firm:

The alert should be readable on a phone screen in a few seconds and contain everything needed to act without opening a transcript. An illustrative version:

URGENT SITE CALL 07:41
Caller: [first name], [number], Site B
Said: boards missing on scaffold lift 2, north side, someone nearly fell
Tried: site manager (no answer), project manager (no answer), director (answered 07:42)
Recording and transcript: [link]

A busy site Tuesday: eight calls and where each landed

A worked day makes the rules concrete. Of 14 calls on a busy Tuesday, the office took six and the receptionist took eight. This is what happened to those eight, in order.

Two transfers, one booking, three messages, one decline and one ended sales call: eight calls, none of which needed anyone to stop what they were doing except the two that deserved it. The step-2 rescue on the plasterer's call is the one that justifies the ladder. Without it, a locked-out tradesman waits for a message to be read.

What 250 calls a month cost on six plans

These are USD list prices from each vendor's own page on 29 October 2026, picking the cheapest plan that covers 250 calls. Where billing is per minute I assumed three-minute calls, or 750 minutes, so check your own average.

Three points the table can't show. Goodcall's model suits this mix because so many callers are repeat callers: a subcontractor who rings every morning counts once a month. Its FAQ lists Google Calendar booking and texts to callers (a confirmation or a link; it can't receive replies yet) but doesn't say how the alert to your team arrives, so ask before relying on it for site calls. Dialzara's per-minute meter punishes long supplier calls: if the average is five minutes rather than three, 250 calls is 1,250 minutes and the cheapest plan becomes Elite plus overage, about $437. And Smith.ai now publishes its Pro tiers: 75, 150 or 300 calls for $150, $270 or $500, with extra calls at $2.50, $2.30 and $2.17. At 250 calls the 300-call tier and the 150-call tier plus 100 extras both come to $500, so the bigger tier is the safer buy, because it leaves 50 calls of headroom for a bad week on site. The tiers are on Smith.ai's AI receptionist pricing page, and the other figures on the vendors' own pricing pages.

Route first, then price: what answering only overflow saves

The 250-call column assumes the AI answers everything. The firm in the first table lets the office answer first, and the AI takes only the 105 calls that reach it. At those 105 calls the same plans cost:

Per-call and per-minute plans reward routing, while flat plans like Rosie's do not, because the plan thresholds don't move until volume drops a long way. The routing therefore comes before the buying decision, and it also changes the best product. Here is how the 105 calls in the first table ended. The new enquiries (36) produced 22 booked site visits, 9 messages for a call-back and 5 declined or referred. The 55 clients, subcontractors, suppliers, site staff and neighbours produced 34 messages and 21 transfers, about one a working day, and the 14 sales or wrong-number calls were ended politely. In all, 22 bookings, 43 messages, 21 transfers, 5 declines and 14 ended calls make 105.

That 21-a-month transfer load is the number to plan around. It tells you how often the ladder will ring a real person, which decides whether a director can be on it at all or whether it needs a rota. If the load were 100, the receptionist would be the wrong tool, and the answer would be a better-staffed office rather than a cheaper plan. For the question of whether to put a chatbot on the website to take new-project enquiries away from the phone, the tutorial on AI chatbots for construction quote requests and service areas is the companion.

Year-one costs beyond the plan fee for a building firm

For a firm this size, your own time in year one is about as large as the plan fees. My own estimate of the time is about eight hours to write the rule sheet, the ladder and the test calls, two hours reading transcripts in the first month and an hour a month after that, which is 21 hours over the year. At $50 an hour that is $1,050.

Two things follow. Time is as big a line as the cheaper plans' fees, so a rule sheet that takes eight hours to get right is worth finishing before you compare vendors by a few dollars. And phone arrangements add small costs of their own: Jobber's Receptionist needs a dedicated number inside Jobber, Quo's Sona works only inside Quo, which bills a seat, and the others need your existing number forwarded. Ask what forwarding costs on your current line, and what happens to the receptionist if you change phone provider.

Recording, disclosure and the opening line

Decide your wording before you go live, because the greeting is the only place callers learn they are speaking to an AI and that the call may be recorded. Jobber's help article encourages telling callers the phone is answered by AI or that calls may be recorded, says this isn't required, and tells owners to check recording regulations that apply to them. Goodcall's FAQ says callers are told they are speaking with an AI assistant. Rosie's pages say every call is recorded and transcribed but don't say whether callers are told or how long recordings are kept. If you have customers in the EU, the EU AI Act's duty to tell people they are talking to an AI has applied since 2 August 2026.

Some places require every party's consent to record a call, and the rules differ between them, so ask a lawyer about the places your callers ring from, then write one sentence you are comfortable saying to all of them. Subcontractors and suppliers hear the greeting too, and some will ring only the site manager's mobile from then on, which may suit you. A sensible habit is to decide how long recordings are kept, who can listen to them, and to keep the ones involving safety reports or disputes. A subcontractor who says "I told you at eight" is a good reason to have a log, and the log is also personal data, so keep it as short as the dispute window requires.

Eight test calls for the awkward callers

Ring the number from a second phone and write down what happened. The awkward callers are the ones the AI was not built for.

1. A subcontractor at the gate says nobody has answered the door. Expected: transfer to the site manager.
2. A driver with twelve pallets outside a site. Expected: transfer, no instruction to leave them.
3. A supplier asking about an overdue invoice. Expected: message for accounts, no promise of payment.
4. An existing client asking when the extension will finish. Expected: no date, message for the project manager.
5. "Someone has fallen from the scaffold." Expected: the emergency line, then the ladder.
6. A neighbour complaining about noise at 7 am. Expected: message and alert to the site manager.
7. A new enquiry for a loft conversion asking for a price. Expected: intake questions, no price, a visit or call booked.
8. A caller who says "put me through to a person". Expected: a transfer within two exchanges.

Run the list again after every change to the rule sheet; how to write call scripts and escalation rules for an AI receptionist covers the general method. Most failures in the first week come from the instructions, not the vendor: a missing caller type, a branch that never ends, a trigger word nobody thought of. If you'd like help drawing the map of who calls your firm and where each call should land, that is what my AI implementation consultation covers, through the contact page. Otherwise, start with a month of logging, pick the routing, and only then pick the plan.

Routing and cost questions from building firms

Can it recognise my subcontractors' numbers?

None of the vendor pages I read says it can match callers against an uploaded number list. The safe design is to ask who is calling and route on the answer, and to give the site team and regular subcontractors a direct mobile number so they never reach the main line. Ask your phone provider about routing trusted numbers around the receptionist.

What should it say about project progress?

Nothing specific. Let it confirm that a message will reach the project manager and give a callback window. Dates, costs, defects and variations belong to a person, because a cheerful wrong answer from a phone assistant can become a dispute. Keep one approved line about opening hours and how to contact the site.

How many caller types can I script separately?

It depends on the plan. Rosie caps custom message-taking scenarios at two on Professional, five on Scale and unlimited on Growth. Other vendors let you describe branches in one set of instructions. Either way, start with five: new enquiry, existing client, subcontractor, supplier and urgent site issue.

Is it worth it for a firm with five staff?

Usually only on overflow. If the office answers most calls, an AI covering after-hours and engaged-line calls at around 100 a month costs about $50 to $215 on the plans priced above. Count your unanswered calls for a month first, because the break-even is one extra won enquiry.

Further reads

• AI Voice Agent for Contractors: What It Handles and What It Hands Off — A closer look at what a voice agent hands to people.
• AI Receptionist for Trades: Built Into Your Job Software or Standalone? — Built into job software or standalone: how to choose.
• Best AI App for a Construction Company: Logging Near Misses — Log near misses once a call has been escalated.
• AI Tools for Home Builders: Build Schedules, Delays and Sub Chasing — Keep clients informed so fewer of them ring.
• AI Solutions for Construction Companies: One Platform or Point Tools? — One platform or separate tools around the receptionist.
• AI Answering Service for Contractors: 100 Calls Priced Across Five Vendors — Smith.ai, Dialzara, Goodcall, Rosie and Jobber Receptionist priced at 100 calls a month, with the call-length sums and the plan limits that change the bill.
• AI Software for Contractors: An Office Stack for a Five-Person Firm — A worked monthly bill for a five-person contractor: scheduling, quotes, calls and reviews priced at lean, standard and one-vendor levels from vendor pages.
• Best AI Software for Construction Companies: Audit What You Own First — A subscription audit for a small construction company: what AI QuickBooks, Microsoft 365 and CompanyCam include, and which gaps justify buying.
• How to Implement AI in Your Business: First Job for a Trades Firm — A scoring sheet to pick the first AI job, a two-week pilot that protects live work, the success check and the handover for a small trades firm.
• Which AI Tools Should a General Contractor Buy? What to Skip — A buy-first order, a skip list with triggers, and a hours-it-must-save test for a general contractor, priced for a six-person remodelling firm.
• Best AI for Builders: Pick the First Job by Hours Lost and Risk — A scoring sheet for small building firms: rank missed calls, quote follow-up, payment chasing and site notes by hours lost and risk, then start with the winner.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Smith.ai AI receptionist and AI pricing pages; Dialzara, Goodcall, Rosie and Quo pricing pages; Goodcall FAQ; Quo's Sona page; Jobber Receptionist help article. All checked 29 October 2026.