The work gets done before the approval arrives

Unapproved change orders are one of the quietest ways a trades business loses money.

A plumber opens a wall and finds corroded pipe beyond the original scope. An HVAC technician discovers the quoted replacement won’t work without electrical remediation. An electrician gets asked to add two circuits while already on site. A roofer finds decking damage after the old materials come off.

The customer says, “Yes, go ahead.”

The technician does the work because the crew is there, the customer wants the problem fixed, and stopping feels like poor service. The office hears about it later through a field note, a photo folder, a text message, or a job costing review that happens after the invoice has already gone out.

By then, the business has a problem. The extra work may be hard to prove. The customer may dispute the amount. The technician may not remember every material used. And the invoice may go out at the original quoted value because nobody wants to reopen the conversation.

For trades businesses doing $1 million to $25 million in annual revenue, that pattern can easily account for part of the typical $50,000 to $200,000 annual leakage band. It isn’t usually one large disaster. It’s dozens of $350, $900, $1,800, and $4,000 decisions made on active jobs without a reliable approval trail.

The answer isn’t telling your technicians to fill in more forms. They’re already balancing safety, workmanship, customer communication, and the next job on the board.

The answer is building a workflow that catches scope changes where they actually show up, then gets the right approval before the extra work turns into donated labour.

Why change orders break down in real trades operations

Most owners already have a change-order process on paper. The issue is that the process depends on people noticing, remembering, and chasing things at the right moment.

Here’s how it typically plays out.

A technician calls the office from site. The owner is in a supplier meeting, driving between jobs, or handling an unhappy customer. The call goes to voicemail. The technician makes a judgment call and keeps moving.

Or the technician sends a text with two photos and writes, “Found more damage. Customer wants fixed.” That message lands in a group chat alongside dispatch questions, supplier updates, and staff messages. No formal change order gets created.

In a larger business, the technician may update the job in the field service platform. The note could say, “Needed extra line set, approved by homeowner.” An admin sees it later but doesn’t know if that is enough to support a charge, what price should apply, or who should send the approval document.

Roofing and restoration jobs have another variation. A project manager learns about changed conditions from photos uploaded at 4:45 p.m. The insurer, property manager, or homeowner needs to approve the variance. The crew returns the next morning because the schedule is tight. The job moves forward, but the commercial record stays incomplete.

This isn’t just a paperwork issue. It affects four areas at once:

Gross margin: Labour and materials are absorbed without recovery.
Cash flow: Approved extras are billed sooner. Unclear extras sit in dispute or never reach the invoice.
Technician confidence: Good technicians feel caught between doing right by the customer and protecting the company.
Job costing: Original estimates look less profitable than they really were because the extra scope isn’t separated and priced.

The more jobs you run, the harder it becomes for an owner to spot every exception. That is exactly where an AI-enabled operations workflow earns its place.

If you’re looking at broader operational gaps across calls, dispatch, estimates, and job handoff, See Omni for trades businesses. Change orders rarely exist as an isolated problem.

What an AI change-order workflow actually does

An AI agent doesn’t replace the technician’s judgement or approve a variation on behalf of the customer. It does the administrative work that causes good change-order processes to fail.

The workflow listens for signals that scope has changed. It assembles the evidence. It creates the approval request. It sends the request to the right person. It follows up. It updates the job record once approval is received. It escalates the exceptions that still need a human decision.

That sounds simple. The detail matters.

1. Identify scope-change signals from the field

The first job is finding possible changes before they vanish into unstructured communication.

For a plumbing business, the signals might include terms like:

“Additional leak found”
“Access panel required”
“Pipe condition worse than expected”
“Customer requested replacement”
“Need to replace valve”

For HVAC, it may be “electrical upgrade,” “ductwork repair,” “line set replacement,” “code issue,” or “permit required.”

For electrical, look for “panel capacity,” “additional circuit,” “trenching,” “damaged wiring,” or “customer added fixtures.”

For roofing, the agent can flag terms such as “rotten decking,” “flashing damage,” “fascia repair,” “underlayment issue,” or “supplement required.”

The AI can review field notes, technician dictation, job photos with captions, internal text messages, email threads, and updates entered into the field service system. It doesn’t need to treat every mention as billable extra work. It needs to identify a likely scope-change event and place it in a review queue.

A useful trigger could be:

Technician note includes a scope-change phrase, a request for additional materials, or a statement that the customer has asked for work not listed in the approved estimate.

Once triggered, the workflow links the note to the job number, original estimate, assigned technician, customer contact, and current job stage.

That context is what turns a vague message into something the office can act on.

2. Compare the new work against the original scope

The next step is where most manual processes slow down.

Someone has to find the estimate, read it, compare it against the field note, check pricebook items, and decide if the requested work is genuinely outside scope. On a busy day, that can mean 10 to 20 minutes per exception. Multiply that across several crews and it becomes another task the office does after hours.

An AI agent can prepare the comparison for a human reviewer.

For example, it can produce a short summary:

Original scope includes replacement of one 50-gallon water heater and standard connection work. Field note reports corroded shutoff valve and non-compliant venting. These items do not appear in the original estimate. Suggested action: issue change order for valve replacement and venting remediation.

The office manager, estimator, or project manager reviews that summary rather than starting from scratch. They can accept the recommendation, adjust it, or mark it as included work.

This is an important control. You don’t want an automated system inventing a price or treating normal workmanship as a variation. The agent should work from your approved pricebook, margin rules, and approval thresholds.

A common rule might be that any proposed addition above a set dollar amount requires a manager review before it goes to the customer. Another rule may require a project manager to approve any change that affects schedule, permit requirements, or subcontractor cost.

You can find more examples of how operations agents fit around human controls in Omni Ops.

3. Generate a customer-ready approval request

Once the proposed change is reviewed, the agent creates a customer-ready request.

It should be clear, brief, and tied to the job. It should explain what was found, why the work is needed, the price, and what happens if the customer chooses not to proceed.

A good approval message isn’t legal language pasted from a template. It is specific enough that the customer understands the decision.

For example:

During the installation, our technician found that the existing shutoff valve is seized and cannot be safely reused. Replacing it will add $485 plus tax. This prevents a leak risk when the new unit is installed. Please approve below so we can complete the work today.

For commercial jobs, the approval may need to go to a facilities manager, general contractor, or property manager. The workflow should use the contact and approval route attached to that account, not automatically text the person who first booked the call.

The customer receives the request by text, email, or a branded approval link. Their response is captured against the job record. If the platform supports e-signature, the approval can be stored there. If approval is given in writing by email or text, the system records the message and attaches it to the change-order file.

A phone conversation can still work, but it needs a follow-up confirmation. The agent can send a message such as, “Thanks for approving the additional venting work by phone. Please reply YES to confirm the added $620 charge.” That removes the ambiguity that causes problems at invoicing.

The workflow needs a stop-work decision

There is a point where automation has to hand control back to the business.

Some extra work is urgent. A plumber may need to isolate a dangerous leak. An electrician may find a safety issue that makes continuing impossible. In those cases, your policy may allow technicians to take immediate protective action up to a defined limit, then submit the event for review.

Other work can wait for approval. Cosmetic upgrades, optional replacements, additions requested by the customer, and non-critical remediation should not be started until approval is captured.

Your agent should recognise the difference and route it accordingly.

A practical decision framework looks like this:

Safety or asset-protection work: Alert the manager immediately. Record the condition and any urgent work completed.
Required work to complete the original job: Send an expedited approval request with photos and a clear explanation.
Customer-requested extras: Create a standard change order and hold the additional work until approved.
Unclear scope: Route the job to an estimator or project manager before pricing is sent.

The goal is not to slow crews down. It’s to give them a clear path instead of asking them to guess what the office wants.

Follow up before the job closes

Approval requests often fail because nobody follows up. The customer is busy. The property manager needs to ask someone else. The email is missed. Meanwhile, the crew is waiting, the schedule gets disrupted, or the work happens without approval anyway.

The agent should run timed follow-up based on urgency.

For a same-day service job, it might send a reminder after 30 minutes, then alert the dispatcher or manager after 60 minutes. For a roofing project or commercial variation, it might follow up the next business day and then every two days until the request is accepted, declined, or escalated.

The workflow also needs a job-close check. Before an invoice is issued, it should compare field notes, materials used, technician time, and open change-order records. If the technician reported extra work but there is no approved change order, the invoice cannot simply pass through unnoticed.

It gets flagged for review.

That one control prevents a lot of leakage. The office can resolve the issue while the job details are fresh, rather than trying to reconstruct events 30 days later.

The same principle applies to estimates. The Estimate Follow-Up Agent can chase proposals on day 2, day 5, and day 14. A trades business that lets approved change orders and original estimates sit without follow-up is leaving two types of revenue uncollected.

If you want a practical checklist for another common leakage point, use the After-Hours Call Recovery Plan for Trades. It helps you map missed-call handling, call-back ownership, and the handoff from first enquiry to booked work. You can also download the worksheet directly for your office manager or dispatcher.

Connect change orders to dispatch and customer communication

Change-order leakage and missed-call leakage often come from the same root issue. Important information is arriving through too many channels, and nobody owns the next action.

A crew is at a property waiting on approval. At the same time, new callers are reaching voicemail because the owner or dispatcher is trying to resolve field issues. That creates a costly trade-off between protecting work already sold and winning the next job.

The 24/7 Dispatch Voice Agent helps remove that pressure. It can answer every inbound call, identify emergencies versus scheduled work, book suitable slots in the dispatch tool, and text a confirmation to the customer. That doesn’t directly approve a change order, but it keeps the front door covered while your team handles high-value exceptions properly.

The Review and Reactivation Agent closes the loop after work is complete. It can request a review from satisfied customers the day after the job, then reactivate them at an appropriate service interval. When your team has a cleaner record of what work was approved and completed, those follow-up messages are more accurate too.

This is how an operations system should work. Each agent owns a clear handoff. No one has to search a group chat to find out whether a customer approved $1,200 in additional work.

Measure the money before you automate everything

Don’t start by trying to automate every kind of variation across every department. Start with the job type where you have frequent scope changes and clear unit pricing.

That might be:

Water heater replacements and repipes
HVAC replacements and duct modifications
Electrical service upgrades and panel work
Roof replacements with decking repairs
Commercial maintenance calls that regularly turn into quoted repairs

For 30 days, track five numbers:

Number of jobs where field notes mention work outside original scope.
Number of change orders created.
Percentage approved before work starts.
Dollar value of approved extras invoiced.
Dollar value of extra work identified after the invoice was sent.

The fifth number is the uncomfortable one. It shows what the existing process is failing to capture.

Then look at the response times. If it takes four hours to send a change order on a two-hour service call, the process isn’t designed for field reality. If a project manager spends six hours every Friday sorting through notes and photos, that work is a strong candidate for an AI-assisted workflow.

You don’t need perfect data to identify the first opportunity. You need enough evidence to see where work is being completed without a commercial decision attached to it.

For a broader view of the operating model behind these workflows, browse the Omni advisory approach. The right design starts with how your crews, office, pricebook, and systems actually operate.

Build controls your crews will use

The best change-order workflow is simple from the technician’s perspective.

They should be able to dictate a note, attach photos, select a reason code, and continue with their job. They shouldn’t need to write a formal proposal on a phone while standing in a hot roof cavity or a cramped mechanical room.

Give them a small set of field prompts:

What did you find?
Is it urgent or can it wait?
Is the customer asking for additional work?
Did you explain the issue to the customer?
Upload two to four clear photos.

Everything else can happen in the workflow.

Office staff need clear service levels too. If the workflow flags a same-day scope change, somebody must own the review within a stated time. If there is no response within that window, the escalation should go to the owner, operations manager, or project lead.

This isn’t about watching people. It’s about preventing the business from accidentally giving away margin because no one was available at the right moment.

Find the leaks before they become accepted losses

Unapproved change orders are rarely a technician problem. They are an operating-system problem.

Your crews see the issue first. Your office needs the information in a usable form. Your customer needs a clear decision to make. Your accounting system needs proof before the invoice goes out. When any link is missing, the extra work becomes vulnerable.

An AI workflow can watch the field signals, prepare the comparison, produce the approval request, chase the response, and stop unclear items from disappearing at job close. Your people still make the commercial and technical calls. They just stop doing repetitive chasing across notes, texts, photos, and inboxes.

If you want to map where unapproved work, missed calls, stale estimates, and manual dispatch are costing you money, Book a 60-min Omni Audit. In 60 minutes, you’ll leave with three practical outputs: your highest-value workflow opportunities, the handoffs that need fixing, and a clear first build plan. No deck, no vague transformation discussion.

You can also review the AI audit for trades businesses before you book. It outlines where we typically see revenue leakage in plumbing, HVAC, electrical, and roofing operations.

The main point is straightforward. If your team is doing extra work, your business should know about it, get approval for it, and bill for it. Book my Omni Audit when you’re ready to put that process in place.

Stop Losing Money on Change Orders