The change order leak most owners don’t see

A technician is halfway through an HVAC replacement and finds a corroded condensate drain. An electrician opens a wall and finds aluminium wiring that needs remediation. A roofer discovers rotten decking underneath the shingles.

The customer says, “Yes, fix it.”

The crew does the work because that’s what good tradespeople do. They take a quick photo, maybe send a text to the office, then move to the next task. A week later, the invoice goes out for the original scope. The extra materials, labour, travel time, and margin are gone.

Nobody intended to give the work away. The technician was focused on getting the system working. The office was answering calls, routing another crew, following up an overdue quote, and trying to find a part. By the time someone notices the gap, the customer has received an invoice and may have no clear memory of approving the extra work.

For trades businesses doing $1 million to $25 million a year, this usually isn’t a single large mistake. It’s dozens of small misses and a few expensive ones. Across plumbing, HVAC, electrical, and roofing, we commonly see annual revenue leakage in the $50,000 to $200,000 range once unbilled extras, underpriced changes, and missing approvals are included.

The fix isn’t telling technicians to “remember change orders” one more time. You need a workflow that catches the moment scope changes, turns field evidence into a priced approval, and stops invoicing until the approval record is complete.

That’s where an AI operations agent can do useful, practical work.

Why change orders disappear in a working trades business

Most businesses already have a process on paper. The problem is that real jobs don’t follow the paper process.

A typical manual sequence looks like this:

1. A technician finds work outside the original scope.
2. They call or text the owner, project manager, or dispatcher.
3. Someone confirms a rough price, often while driving or between calls.
4. The technician gets a verbal approval from the customer.
5. Photos stay in a technician’s camera roll, text thread, or job notes.
6. The office tries to reconstruct the change when invoicing.
7. The invoice is sent without the extra work, or the team bills a number that doesn’t cover the full cost.

On a residential service call, an extra $350 drain line repair may slip through because it feels too small to chase. On commercial electrical work, a missing signed change order can mean thousands of dollars are disputed at the end of a project. On a roofing job, extra decking and flashing can become a margin problem quickly if the original bid is the only number that makes it to the final invoice.

The issue gets worse when the owner is still serving as dispatcher. Many firms at this stage lose 20 or more hours a week to phones, crew questions, customer updates, suppliers, and schedule changes. The owner may know every job needs a change order process. They simply don’t have a reliable way to enforce it while work is moving.

That same pressure creates related leaks. Calls go unanswered while the office chases job details. Estimates are sent but not followed up. In many trades firms, a disciplined follow-up sequence can bring back 15% to 25% of stale estimates. These are not separate problems. They point to one operating issue: critical revenue tasks depend on a person remembering them at the exact right time.

What AI should do in a change order workflow

AI isn’t there to decide if a customer should pay or to invent pricing. Your business still controls the price book, labour rates, markups, approval thresholds, and exceptions.

Its job is to make the process hard to miss.

A well-built workflow watches for signals that extra work occurred, prompts the field team for the missing information, creates a draft from approved rules, and routes the customer approval before invoicing. It also gives the office a clear exception list instead of making them hunt through job notes.

For a plumbing, HVAC, electrical, or roofing business, the workflow usually has five parts.

1. Detect a scope change at the job

The agent can monitor job notes, technician voice notes, photos, task codes, materials added, and messages sent from the field. It looks for phrases and events that indicate work outside the quoted scope.

Examples include:

• “Found damaged wiring behind panel”
• “Customer requested an additional return vent”
• “Three sheets of decking replaced”
• “Drain line needs replacement, not repair”
• A material entry that isn’t tied to the original work order
• A technician marking more labour hours than the estimated allowance

The agent then sends the technician a short prompt on the device they already use. Not a long form. Something like:

This appears outside the original scope. Please add two photos, select the reason for the change, and confirm whether the customer has approved the work.

For urgent service work, that prompt needs to take less than a minute. A technician can capture two photos, dictate a note, and choose from a few reason codes. If the work is safety-critical, the workflow can mark it accordingly and escalate a notification to the office.

The point is to capture the evidence when the technician is standing in front of the issue, not later that night when details have faded.

2. Turn the field record into a priced draft

Once the technician records the scope change, the AI agent can turn the notes and photos into a structured draft change order.

It can pull the relevant original job scope, identify the likely labour category, list the added material, and map the work to your price book. If your HVAC business has set rates for condensate drain replacement, the draft should use those rates. If roofing decking is priced per sheet with a defined labour allowance, the workflow should apply that rule.

The office sees a clear summary:

• Original scope
• New condition found
• Evidence photos and technician notes
• Proposed added labour
• Proposed materials
• Suggested customer-facing description
• Price range or approved fixed price
• Approval status

This isn’t an automatic invoice. It is a draft created from your operating rules. For lower-value changes, you may allow the system to send the customer approval automatically. For larger changes, perhaps anything over $1,000 or any commercial contract work, a manager reviews it first.

That structure matters. A vague note saying “extra work done” gives an accounts receivable team nothing solid to bill. A documented change with photos, a clear reason, and customer approval is much easier to invoice and defend.

The approval has to happen before the invoice

The biggest control point is simple: no invoice should close a job with unapproved extra work sitting in the record.

An AI operations agent can check every completed job against the change order queue before the invoice is released. If a technician recorded extra work, used extra materials, or logged a scope-change code, the agent checks for a signed approval or a documented emergency exception.

If approval is missing, it creates an exception:

• Job number and customer
• Technician and date
• Value of the proposed change
• Photos and notes available
• Customer approval status
• Next action owner
• Time since the work was completed

The office doesn’t need to review every job. They review exceptions.

For customer approval, the workflow can send a branded SMS or email with a plain-language explanation, photos, amount, and an approval button. For larger commercial jobs, it can route a formal PDF or e-signature request to the authorised contact.

A good message doesn’t sound like a legal department wrote it. It should be direct.

During the scheduled electrical work, our technician found damaged wiring that wasn’t visible before access was opened. We recommend replacing this section to complete the job safely. The added work is $780. Photos are included below. Please approve so we can proceed.

If the customer doesn’t respond, the agent follows a set cadence. It can remind them after a few hours for an active job, then alert the project manager if the job risks delay. For completed work that was verbally approved, it flags the record for a human call rather than sending a surprise invoice days later.

This protects revenue, but it also protects customer trust. Customers are far more likely to accept a documented change at the moment the condition is discovered than an unexplained amount on a final invoice.

A practical end-to-end example

Consider a $14,000 residential roofing job. The estimator allowed for minor deck repairs. Once the existing roof is removed, the crew finds 12 damaged decking sheets. The original allowance covers four.

Without a workflow, the foreman calls the owner. The owner is in the middle of dispatching an emergency plumbing call and says, “Take photos, we’ll sort it out.” The crew replaces the decking because the roof can’t be completed otherwise. Days later, the office invoices the original contract value. The extra materials and labour might represent $1,500 to $3,000 in recoverable revenue, depending on the pricing model.

With an AI-supported process, the flow is different:

1. The foreman selects “additional decking” in the job app and takes required photos.
2. The agent compares 12 sheets with the original four-sheet allowance.
3. It creates a change order draft using the business’s per-sheet labour and material rules.
4. The project manager gets a notification to approve the price.
5. The customer receives a photo-backed approval request before the work continues.
6. The signed approval attaches to the job record.
7. When the final invoice is prepared, the approved amount is already included.
8. If the customer has not signed, the invoice queue marks the job as an exception.

No one has to remember the entire chain. The workflow carries the work from discovery to invoice.

That is the difference between an AI feature and an operating system. It has a trigger, a controlled decision, a customer action, an escalation route, and a final check.

Build it around the systems you already use

Most owners don’t want another app that crews ignore. Fair enough.

The strongest version of this workflow connects to the systems already running your jobs. That may include your field service platform, CRM, accounting system, photo documentation tool, messaging platform, and price book. The agent should write structured information back into the job record rather than creating a parallel spreadsheet nobody trusts.

At Enterprise DNA, this is the kind of workflow we map through Omni Ops. We start with the actual handoffs in your business, not an abstract list of AI capabilities.

The workflow needs clear rules before it goes live:

• What events count as a scope change
• Which technicians can submit one
• Which photo and note requirements apply
• Who can approve pricing internally
• Dollar thresholds for automatic customer requests
• Different handling for emergency, residential, and commercial work
• What a valid approval record looks like
• Who owns a missing-approval exception
• When an invoice must be held

You may also choose to set different rules by department. Service HVAC calls need speed. Roofing and electrical projects often need more formal documentation. Plumbing emergency work needs a safety exception path, with documentation completed as soon as the immediate risk is controlled.

The right process has enough discipline to protect revenue without slowing crews down.

If you want to see where this fits beside call handling, follow-up, and job administration, see Omni for trades businesses. The audit is built around the operating constraints that matter in a working trade business.

Don’t ignore the call and estimate leaks around it

Unbilled change orders are often a symptom of an overloaded front office. Fixing them can free attention, but you should also look at the other revenue moments your team misses.

The 24/7 Dispatch Voice Agent answers calls when the owner, dispatcher, or office manager is tied up. It can qualify whether the job is an emergency or scheduled service, book directly into the dispatch tool, and send a confirmation text to the customer. That matters because a missed call can be a $500 service job or a $3,000 replacement opportunity that goes to the next company on the search results page.

The Estimate Follow-Up Agent tracks every estimate and follows up on day 2, day 5, and day 14 with messages appropriate to the job size and trade. It prevents good estimates from quietly dying because nobody had time to make the second call.

The Review and Reactivation Agent asks happy customers for a review the day after the job and contacts previous customers at the appropriate service interval. That can be a seasonal HVAC check, plumbing maintenance reminder, or roof inspection prompt.

These agents shouldn’t be bolted on as disconnected tools. They should share the same view of customer, job, estimate, invoice, and outcome. You can read more about the call side of the model in Omni Voice, or review the broader operating approach in our AI operations resources.

If after-hours calls are part of your current problem, use this After-Hours Call Recovery Plan for Trades as a working checklist for who answers, what gets booked, and where follow-up breaks down. You can also get the printable version here: download the call recovery plan.

What to measure in the first 90 days

Don’t judge this work by how many AI messages were sent. Judge it by revenue recovery and operational compliance.

Start with a baseline from the previous 60 to 90 days. Review completed jobs for notes, materials, and labour that indicate extra work. Compare those jobs to invoices and signed change orders. You won’t get a perfect historical number, but you’ll identify patterns.

Then track:

• Scope changes identified per 100 jobs
• Percentage with required photos and notes
• Time from discovery to customer approval
• Percentage approved before work proceeds
• Percentage billed before job closeout
• Average change order value by trade and job type
• Value of changes held for missing approval
• Disputed change orders
• Revenue recovered compared with the prior baseline

The goal isn’t to maximise the count of change orders. It is to ensure real additional work is documented, approved, and billed fairly.

A healthy process may initially show more change orders because you’re finally seeing work that was always happening. Over time, it should show fewer missing photos, faster approvals, and very few completed jobs with undocumented extras.

Find the leak before you buy more software

You don’t need a generic AI strategy session. You need to know where a change order starts disappearing in your specific operation.

In a 60-minute Omni Audit, we map the workflow from technician discovery through customer approval and final invoice. You leave with three outputs: the highest-value revenue leaks, a practical agent workflow for the next 90 days, and a prioritised implementation plan. No deck, no theatre.

Book a 60-min Omni Audit if you want to identify what your crews are completing but your invoices aren’t capturing.

You can also review the AI audit for trades businesses to see the operational areas we assess. Start with change orders if that is the immediate pain, then connect dispatch, estimate follow-up, and customer reactivation once the core job workflow is reliable.

The money lost on unbilled work is rarely mysterious. It is sitting in technician photos, text messages, material entries, and conversations that never made it to an approved invoice. Put an agent in charge of the handoffs, and your team can spend more time doing the work customers hired them to do.

When you’re ready to map that workflow against your systems and pricing rules, Book my Omni Audit.