Field Service & Back Office AI · 9 min read
Fencing Missed Call AI Recovery
The Omni model: one managed AI Employee owns one recurring workflow; specialist employees add capacity around the same business context; your team keeps the judgment calls.
A fencing company in suburban Dallas installed a new CRM two years ago, paid for a Google Local Service Ads budget, and started booking $40,000 a month in jobs. Then the owner looked at the actual numbers and realized something had gone sideways. The crew was on-site measuring driveways, the office phone was ringing during a backhoe delivery, and three calls hit voicemail in a single afternoon. None of those callers got a text back. None of them turned into appointments. The owner wasn't losing the leads because of bad marketing. He was losing them between the ring and the response.
This is the gap most home-services businesses ignore. Lead generation gets budgeted, staffed, and optimized. Lead recovery—catching the calls, web forms, and chats that didn't get handled in the moment—gets treated as an afterthought. For fencing specifically, the math is brutal because the call volume is bursty, the calls are long, and the field crews physically cannot answer the phone when they're 30 minutes into setting posts.
This article walks through a concrete missed-call recovery workflow we deploy for fencing clients: what the AI agent does, where the handoff to a human happens, what gets logged for review, and what the fallback looks like when the system fails.
Why fencing companies lose more leads than they think
The default assumption inside a small fencing operation is that the office staff handles calls. That's true for roughly 20% of the day. The rest of the time—the 60 minutes before a crew heads out, the entire window when the crew is at a job, and every minute after 5pm—the phone either rolls to voicemail or rings out.
A 2011 Harvard Business Review study by James Oldroyd (then at MIT's Sloan School) found that responding to a web inquiry within five minutes makes the contact 21 times more effective than responding 30 minutes later. Oldroyd's follow-up research showed that the odds of qualifying a lead drop by roughly 80% if you wait an hour. The study is old, but the response-time curve has held up in subsequent replication work—it is one of the most reliable findings in sales operations research.
For fencing, the comparable window is shorter, not longer. A homeowner calling about a $12,000 cedar fence replacement is usually calling two or three contractors in the same afternoon. The first one who texts back wins the appointment. If your office sends the lead to voicemail and gets to it 90 minutes later, you are calling someone who has already booked an estimate with your competitor.
The volume problem compounds it. A fencing company running $30K–$60K/month in revenue typically takes 150–300 inbound calls per month. Missed-call rates on busy weeks run 25–40%. That is 40–120 lost conversations per month that never enter your pipeline.
The missed call recovery workflow
A recovery workflow has four stages, and each one has a specific handoff. Here is the version we deploy most often.
Stage 1: Detection. The moment a call goes unanswered, the AI agent picks it up from the telephony layer (we typically use a SIP trunk or a hosted PBX integration). It logs the caller ID, the time of the call, and whether a voicemail was left. If there was a voicemail, the audio is transcribed and routed to stage 2. If there was no voicemail, the agent proceeds to stage 2 anyway, because the missed call itself is the signal.
Stage 2: First-touch text. Within 60–90 seconds, the AI sends an SMS from a business number (not a generic short code—carrier filtering and consumer trust are noticeably better on a long-code business line). The text reads roughly like this: "Hi—this is [Business Name]. Sorry we missed your call. Are you looking for a quote on a fence project? Reply YES and we'll get back to you with a time to talk."
The reply is the trigger. A "YES," a question about pricing, or any non-stop-word response moves the lead to stage 3. A "STOP" or no response within four hours drops the lead into a 24-hour nurture queue, then a 7-day re-engagement queue.
Stage 3: Qualification conversation. Once the lead is engaged, the AI agent runs a short conversation—usually 4–8 messages—to capture project type (privacy, chain link, ornamental, etc.), approximate linear footage if known, timeline, and property address. The conversation is templated but not rigid. If the lead asks something outside the script, the agent either answers from a knowledge base or escalates immediately. We configure the escalation thresholds during implementation: anything involving permits, HOA restrictions, or commercial bids goes straight to a human.
Stage 4: Booking handoff. If the lead is qualified and within scope, the agent offers two appointment windows for an on-site estimate, drawn from the field calendar. The lead picks one. The job is created in the CRM (we work with Jobber, ServiceTitan, Housecall Pro, and a few others) and a confirmation text is sent. The office manager gets a notification with the full transcript attached.
The human enters the loop at two specific points: the first time the lead asks something the AI is not authorized to handle, and any time a job is booked over a configurable dollar threshold.
Implementation scenario: a residential fencing company
Here is the version we built for a client last quarter. The company runs three crews in a metro area, books roughly $55K/month, and was missing about 90 calls per month based on their call logs.
The setup took about two weeks. Week one was workflow mapping: we sat with the owner and the office manager, pulled two months of call recordings, and tagged each call by outcome (booked estimate, lost to voicemail, lost to competitor, callback closed). Of the 90 missed calls per month, 38 were recoverable in the first 24 hours. Another 24 were recoverable in the 7-day window. The remaining 28 were either spam, wrong numbers, or duplicate calls from the same lead.
Week two was configuration. We connected their existing phone system via SIP, integrated their Jobber account, and built the qualification template around their actual job types and average project sizes. We loaded their FAQ (HOA approval process, permitting timeline, payment terms) into the agent's knowledge base. The biggest decision during week two was the dollar threshold for human escalation—we set it at $15,000, which routed roughly one in seven conversations to the office manager for live follow-up.
The first month of operation: 91 missed calls detected, 74 first-touch texts sent, 41 replies received, 22 qualified leads, 14 booked estimates, 9 closed jobs at an average ticket of $11,400. Total closed revenue from the recovered leads: $102,600 against a $3,400 monthly operating cost for the AI agent. The owner was skeptical of the math until his controller reconciled it against the Jobber ledger.
This is not a guaranteed outcome for every fencing company—the volume and conversion math depend on local competition, ad spend, and seasonality. But the structural shape of the result holds up across our implementations: most recovered revenue comes from a quick text response, not from a phone-call recovery attempt.
Where humans stay in the loop
The most common question we get from owners is whether the AI is "calling customers." It is not. The AI is texting, qualifying, and booking. The humans—office manager, estimator, or owner—handle the conversations that require judgment.
There are three review points in the standard workflow:
- Daily transcript review. The office manager reads every conversation that ended in a booking or an escalation. This takes 15–20 minutes per day. The purpose is to catch any AI mistakes—an incorrect price range, a wrong service area answer, a misread of the property type.
- Weekly exception review. The owner or operations lead reviews the escalated leads, the ones the AI flagged as out-of-scope, and any leads that didn't book after full qualification. This is where the strategic conversations happen: which leads are worth a callback, which ones to drop, which patterns suggest the qualification template needs adjusting.
- Monthly performance review. We pull a report from the CRM comparing recovered leads, booked jobs, and closed revenue against the baseline. If the close rate on recovered leads drops below 30%, we investigate whether the AI is over-qualifying (pushing leads to estimates that should have been disqualified) or under-qualifying (letting leads through that should have been escalated).
Gartner's research on conversational AI in customer service has been consistent on this point: the deployments that fail are the ones that try to remove humans entirely. The deployments that hold up over 12+ months are the ones with clear escalation paths and regular transcript review.
Common failure modes
Three patterns show up repeatedly when we audit existing AI deployments for service businesses.
No escalation path. The AI is configured to handle every conversation, including ones it shouldn't. The first time a customer asks about an HOA dispute or a commercial bid, the AI improvises. The result is a quoted price that doesn't match the company's actual pricing model, and a customer expectation that has to be unwound by a human later.
Stale knowledge base. The AI was trained on the company's FAQ at launch, and the FAQ has not been updated in eight months. The agent is confidently telling customers the company doesn't do wrought iron work, which stopped being true four months ago.
No baseline measurement. The owner cannot tell you, before the AI goes live, how many calls they were missing or what those calls were worth. Without a baseline, there is no way to measure whether the recovery workflow is actually recovering anything. We always start with a 2-week measurement period before turning on the AI agent. That baseline becomes the comparison point for every monthly review.
A McKinsey analysis of AI in service operations found that the companies getting durable value from these deployments were the ones who treated the AI as a workflow change, not a software purchase. The technology is the easy part. The workflow design and the review cadence are what make the difference.
FAQ
How fast does the AI text back after a missed call?
Within 60–90 seconds. The first-touch text is the highest-conversion event in the entire workflow, and response time is the main lever. Anything slower than two minutes noticeably reduces reply rates in our testing.
What happens if the customer asks to speak with a person?
The AI acknowledges the request, transfers the conversation to the office number, and notifies the on-call human via SMS with a short summary of what the customer has said so far. The handoff takes 30–60 seconds.
Does the AI place outbound calls?
No. Outbound calling is a different workflow with different compliance requirements (TCPA, prior express written consent). The recovery system is text-first by design. If a customer explicitly asks for a call, the agent schedules a callback from a human rather than placing one itself.
What if the AI gives a wrong answer?
That is what the transcript review process is for. Every booking and every escalation is reviewed by a human within 24 hours. If the AI makes a substantive error, the lead gets a follow-up from the office manager and the knowledge base is updated.
Will this integrate with our existing CRM?
In most cases, yes. We integrate with Jobber, ServiceTitan, Housecall Pro, Service Fusion, and a handful of smaller systems. If your CRM is on a custom stack, we can usually build a Zapier-based bridge during implementation.
Get a free audit on your missed call workflow
If your fencing company is missing calls and you don't know how many, that's the first thing to find out. We run a free two-week missed call audit that includes:
- Baseline measurement of missed-call volume and timing
- Recovery potential estimate based on your actual call logs
- A workflow recommendation tailored to your CRM and phone system
No software purchase required, no commitment. Book a free AI automation audit and we'll send you the numbers within two weeks.
