It is 6:40 on a Tuesday morning in Ōtaki and the foreman is standing in a paddock of formwork with his phone in one hand and a coffee going cold on the ute bonnet. Sixty square metres of driveway, boxed and steel-tied since Friday. Two loads of 25 MPa (megapascals) booked for a nine o'clock start. The radar has a green smudge sitting off the coast that was not there when he went to bed.
So begins the ring-around. The batch plant, to ask whether the load can shift to eleven. The pump operator, who is already halfway to Levin. The homeowner, who has taken the day off work and would quite like to know either way. The two labourers on their way in from Palmerston North. Five calls before seven, each one short, each one carrying a decision that changes the other four.
Concrete crews are very good at this. The ring-around is a skill, built over years of reading cloud and knowing which plant dispatcher will hold a slot for you. What it costs is the foreman's morning, every morning, and a phone that cannot be put down while the crew is placing.
We build AI agents that carry that coordination load. Not a booking widget bolted to a website, but an agent that watches the forecast against your pour list, phones the customer when a window closes, holds the conversation with whoever picks up, and writes the outcome back into the job. Below we will walk through what makes a pour slip in the first place, how an agent handles the truck slot chain, what it does with a same-day cancellation, how to design the triggers so it phones people only when there is something worth saying, and what you need in place before any of it earns its keep.
What actually makes a concrete pour slip?
Pours slip for three reasons: weather that arrives inside the finishing window, a truck slot that moves at the plant end, and a site that is not ready when the agitator arrives. Weather is the biggest of the three and the only one with a forecast attached, which makes it the easiest to get ahead of.
The awkward part is that none of the three announce themselves at a convenient hour. A front that was due Thursday afternoon comes through Wednesday night. A plant reshuffles its run to cover a commercial job. A plumber has not finished the underslab work. Each one shows up as a phone call at an hour when someone is either driving or already placing.
That is what makes pour scheduling a coordination problem rather than a calendar problem. The calendar is easy. Keeping six people and one truck agreed on the same version of the calendar, at 6:40 in the morning, is the work.
How much does a rained-out pour cost?
A washed-out residential pour tends to sit between $2,000 and $4,000 once you add it up. Concrete at roughly $290 to $350 per cubic metre delivered, depending on mix and region. Pump hire from around $700 for a half day with an operator. A crew of four or five standing on site for a morning that produces nothing.
Those figures are indicative and worth checking against your own plant and region, but the shape holds. The concrete itself is often the smaller number. If the load has been batched and dispatched it is paid for regardless, and once it is on the truck the clock is running on it.
Then there is the finish. Rain landing on fresh concrete inside the first few hours washes the surface paste and leaves a weak, pitted top that no amount of trowelling will save. That is a grind-and-reseal at best, a break-out and repour at worst, and either way it is the conversation nobody wants to have with a homeowner who has just watched their new driveway go grey and spotty.
When does a pour become uncancellable?
The practical cut-off is the moment the load is batched. Many plants want notice by early afternoon the day before to move a slot cleanly, and once the mix is in the drum you are usually paying for it. Short-load and cancellation fees vary by supplier, so the number that matters is the one on your own account terms.
Which puts a hard edge on the decision. A call made at 3pm Monday moves a Tuesday pour for nothing. The same call at 6:40am Tuesday costs you a load of concrete and a pump booking. Fourteen hours is the difference between a scheduling change and an invoice.
Most of the time nobody is sitting there at 3pm on Monday reading a marine forecast for the next morning. They are on site, gloves on, finishing yesterday's slab. This is precisely the kind of watch a machine keeps well and a person keeps at the cost of their attention.
How does an AI agent manage the batch plant chain?
The agent holds the sequence rather than the relationship. It knows which pours are booked, which plant each load is coming from, what the notice window is on each account, and who needs to be told when something moves. When a date changes it works down that chain in order and phones each party with the version of the change that matters to them.
Note what it is not doing. It is not ringing the dispatcher to negotiate a favour, and it is not deciding whether Thursday's window is good enough to place in. Those are judgement calls that belong to the foreman, who has stood in that particular gully at that particular time of year and knows how long the sun takes to get over the ridge.
What the agent removes is the mechanical part. The four confirmation calls that follow the decision. The text to the labourers. The note in the job file that says why the date moved, so that when the homeowner asks in three weeks there is an answer.
How does an agent hold and confirm a truck slot?
For most crews the agent handles everything around the slot and leaves the plant call to a human. It prepares the request with the volume, mix, slump, pump requirement and site access notes, phones the customer to confirm the new time first, then either books through the plant's ordering system or hands the foreman a single call to make with every detail already settled.
Some plants take orders through an online portal, in which case the agent can write the booking straight in. Others run on a phone relationship with a dispatcher who has known the foreman for fifteen years, and there is no good reason to put a machine in the middle of that.
The design question is where the human boundary sits, and it is worth being honest about it early. We usually draw it at anything involving negotiation or a favour, and let the agent own everything that is pure information transfer. That split holds up well in practice and it keeps the plant relationship where it belongs.
What happens when the customer moves the date?
The agent takes the call, confirms the new date against the pour list and the forecast, checks it against the crew's existing bookings, and then works the chain: plant, pump, crew, any subcontractor with underslab work. Each party gets the detail relevant to them. The job record is updated once, and everyone is working from it.
The customer-side call is where a voice agent earns its place, because those calls arrive at hours when nobody is free. Saturday afternoon. Half seven at night. Sunday, from a homeowner who has been thinking about it all weekend and has decided the plumber needs another week.
Answering those in the moment, with a real conversation rather than a message left on voicemail, keeps the pour list current going into Monday morning. It is the same pattern we built CallCover around for trades more broadly, tuned here to the specific shape of a pour booking.
What happens when a job cancels on the morning?
A same-day cancellation leaves two things on your hands: a crew with nowhere to be and, if the load is already batched, concrete with a shelf life measured in hours. The agent's job is to fill both gaps fast, working a pre-built list of jobs that are formed up, signed off and close enough to reach.
This is where the gap-filling logic has to be honest about geography. A shortfall load that could go to a Waikanae job is worth calling about. The same load with a Feilding job as the only candidate is not, once you have added the travel and the risk of the mix going off in the drum.
The realistic outcome is not that every cancelled slot gets refilled. It is that the ones which can be refilled get refilled, inside the window where it is still possible, without the foreman running a phone tree from the cab of a ute.
Who does the agent call first when a slot opens?
Ranked by readiness, then by distance, then by how long the job has been waiting. Readiness means formwork up, steel in, base compacted and the customer already agreed to a short-notice call. Distance is measured from the batch plant, not the yard, because the truck is the thing with the clock on it.
That standby list is the piece worth building properly. It comes from your existing job system, and it wants a simple readiness flag that the crew can set on site with one tap. Fergus, Tradify, NextMinute and simPRO all handle a custom field or a job status for this without any drama.
Once the list exists, the calling is mechanical. Three or four calls in ten minutes, each one asking the same question, each answer written straight back to the job. A person can absolutely do this. A person cannot do it while screeding.
How do you keep short-notice calls from becoming a nuisance?
By making the standby list opt-in and treating it as a standing agreement. When a customer books, they are asked whether they want to be on the short-notice list, and told plainly what that means: a call that might come at 7am with an offer to pour today. Everyone who says no is never called about it again.
The customers who say yes tend to be the ones who benefit most, because a short-notice slot often comes with a better date than the one they were quoted. Get that framing right at booking time and the calls are welcome. Get it wrong and you have built an automated nuisance, which is worse than no system at all.
Frequency caps matter too. One short-notice call per customer per week is plenty. The agent should also know not to phone before seven or after eight, and to stop entirely once the slot is filled, so nobody gets a call about a load that went out twenty minutes ago.
How do you stop the agent crying wolf about weather?
By tying the triggers to conditions that change the pour decision, not to rain probability on its own. A pour cares about rainfall inside the finishing window, wind and heat driving evaporation off a fresh surface, and overnight temperature falling low enough to stall the set. Everything else is weather the crew can work in.
This is the part that gets skipped, and it is the part that decides whether anyone trusts the thing in month three. An agent that phones about every 40 percent chance of showers gets ignored inside a fortnight. An agent that phones twice a month, and is right both times, gets listened to.
Start deliberately conservative on the alert side. It is better to miss a marginal call in the first month and tighten the thresholds afterwards than to open with a fortnight of false alarms that teach the crew to swipe the notification away.
Which forecast signals are worth acting on?
Three of them, in practice. Measurable rain forecast inside the pour and finishing window, which for most residential slabs is the pour time plus four to six hours. High evaporation conditions, which means hot, dry and windy at the same time. And an overnight minimum heading toward five degrees, which slows the set enough to matter.
Set thresholds against your own mixes and your own region. A Kāpiti crew placing in July is watching the temperature line. The same crew in February is watching wind and humidity, because plastic shrinkage cracking on a big exposed slab is a nastier problem than a bit of drizzle.
The forecast side is straightforward to wire up. MetService point forecasts and rain radar, read hourly against each pour on the list, with the site's own coordinates rather than the nearest town. Coastal Kāpiti and a site four kilometres inland behind the ridge do not get the same afternoon.
How far ahead should the agent be looking?
Two windows, doing two different jobs. A 48-hour look-ahead that flags pours worth watching, and a firm decision check the afternoon before, timed to land before your plant's notice cut-off. The second one is the call that saves money, because it is the last point where a change is free.
The 48-hour flag goes to the foreman as a heads-up and nothing more. No customer contact, no plant contact. It exists so that when the decision check comes around the next day, nobody is surprised.
The decision check is the one with a phone call attached. The agent presents what it is seeing, the foreman makes the call, and the agent then works the chain. Human judgement in the middle, mechanical work either side of it. That shape shows up in most of the agent work we do, whatever the industry.
What do you need in place before this works?
Three things: a job list a machine can read, site coordinates rather than street addresses, and clarity about which decisions stay with the foreman. Get those right and the build is a few weeks. Get them wrong and you have an agent making confident calls about the wrong paddock.
The job list is usually the easy one, because most crews already run Fergus, Tradify, NextMinute or something similar, and those systems have an application programming interface (API) or at minimum an export. The data that matters is thin: date, time, volume, mix, plant, pump requirement, site location, customer contact and a readiness status.
Coordinates matter more than people expect. Rural jobs on the Kāpiti Coast and through the Horowhenua routinely have addresses that geocode to a letterbox two paddocks from where the truck needs to be, and the weather difference across that distance can be real on a windy day.
What does the agent need to know about each job?
Enough to hold a sensible conversation and no more. Volume and mix so it can talk to the plant. Site coordinates for the forecast. Pump requirement, access notes, the customer's name and number, the crew assigned, and a readiness flag. That is the working set for most residential and light commercial pours.
Resist the urge to feed it everything. An agent with access to the full job history will occasionally volunteer something from an old note that the customer did not need to hear, and every extra field is another thing to keep current. Thin and accurate beats comprehensive and stale.
Placing and curing requirements sit outside the agent's remit entirely. NZS 3109 covers that ground and it is the foreman's call, informed by the mix design and the conditions on the day. The agent schedules; it does not specify.
How does this sit with the Privacy Act 2020?
Cleanly, provided you handle three things: tell people calls are recorded, collect only what the job needs, and set a retention period rather than keeping recordings forever. The Privacy Act 2020 does not stand in the way of a voice agent. It asks you to be straight with people about what you are collecting and why.
In practice that means a short disclosure at the start of the call, a written record of consent for the short-notice list, and a retention rule that deletes call audio after a set period while keeping the structured outcome. Ninety days is a common setting and it covers the "what did we agree" question comfortably.
Worth noting that this is a step up from the status quo for most crews. A conversation held on a ute phone with no record at all is not more private, it is less accountable. A logged, consented, time-limited record is easier to stand behind if a job ever goes sideways.
Practical Takeaway
Start with the decision check the afternoon before the pour, because it is the single point where information turns into saved money. Everything else in this post is an extension of that one habit.
Here is the sequence we would work through. First, write down your plant's actual notice cut-off, from your account terms rather than memory. Second, pick your three weather thresholds and put numbers on them, in your region, for your mixes. Third, add a readiness flag to your job system that the crew can set on site with one tap. Fourth, ask every customer at booking time whether they want to be on the short-notice list, and record the answer.
Those four steps are worth doing whether or not you ever build an agent. They are what makes a pour list legible, and a legible pour list is the thing an agent needs. Automating a schedule that only exists in the foreman's head is how these projects fail.
Once that is in place, the agent build is small. A forecast watch against the pour list, an outbound call for the decision check, a customer-side line that answers at half seven on a Sunday, and a gap-filling routine for the mornings that go wrong. Weeks, not months. If you want to talk through what that looks like for your crew, get in touch.
Frequently asked questions
- Can an AI agent book concrete trucks directly with the batch plant?
It depends on the plant. Where there is an online ordering portal, an agent can write the booking directly. Where orders run through a dispatcher by phone, most crews keep that call human and have the agent prepare every detail first, so the foreman makes one short call instead of four.
- How much does a rained-out concrete pour cost in New Zealand?
Typically $2,000 to $4,000 for a residential job once you include a batched load at roughly $290 to $350 per cubic metre, pump hire from around $700 for a half day, and a crew standing idle for a morning. Remedial grinding or a repour pushes it considerably higher.
- Will an AI agent decide whether it is safe to pour?
No. The agent watches the forecast against your pour list and raises the decision with the foreman before the plant's notice cut-off. The call about whether conditions suit the mix, the site and the finish stays with the person who knows the site. The agent handles the calls that follow.
- Do I need to record calls to use a voice agent for pour scheduling?
Recording is not required, though most crews want it for the "what did we agree" conversations. If you do record, the Privacy Act 2020 expects you to tell callers at the start, collect only what the job needs, and delete audio on a set schedule. Ninety days is a common retention period.