24/7 Sales Without a Team: How AI Agents Handle Customer Enquiries

It is 23:40 on a Thursday in Riyadh. A contractor messages asking for pricing on 240 square metres of curtain wall, deadline eleven days. Nobody is at the desk. The message is read at 07:15, answered at 09:30, and by then he has priced the job with two competitors who replied inside ten minutes.

Every business with a sales desk has a version of that story. The usual response is to hire, or to ask the team to check their phones at night — both treat a coverage problem as a staffing problem. The useful response is to measure the gap precisely, which turns a complaint into an engineering brief.

Why "We Reply Fast" Is Not a Number

Ask a sales manager how fast the team responds and the answer is always an average, usually rounded down. Averages hide the only thing that matters: the tail. A desk answering ninety enquiries in twenty minutes and ten in nine hours has an average that looks fine and, for a tenth of the pipeline, a customer experience indistinguishable from not answering at all.

The GCC week makes this worse than it looks on paper. Sunday to Thursday leaves Friday and Saturday unstaffed, and those two days carry a disproportionate share of enquiries from retail buyers and fit-out contractors. Add the after-hours window on working days and a typical desk is dark for roughly 60 hours a weekwhile its buyers are awake.

The Five Numbers That Move When Enquiry Handling Goes Autonomous

These five figures are the first we instrument on an agentic sales deployment. Each can be pulled from systems a business already owns, and none can be improved by writing a nicer press release.

Number One: First-Response Time at the Worst Hour

Not the average — the ninetieth percentile, and specifically the enquiries that arrive between 22:00 and 06:00 or on a Friday. That single figure tells you whether coverage is real or theatrical. On the deployments we run it typically starts above four hours and settles under fifteen seconds, because the agent replies on the channel the buyer used rather than waiting for someone to open an inbox.

Number Two: After-Hours Capture Rate

The share of out-of-hours enquiries that receive a substantive reply — one that acknowledges the request, asks the qualifying question or quotes a range — within fifteen minutes. Most desks measure it for the first time and find it under a quarter. An agent that covers WhatsApp, email and web forms in one queue pushes it toward the high nineties, and every point of that move is demand marketing already paid for.

Number Three: Enquiry-to-Qualified-Meeting Rate

Lead counts are vanity; a qualified meeting is a conversation where budget, scope and timeline are established and a slot is on a calendar. The agent qualifies in the first exchange instead of the fifth, so the ratio moves both ways at once: more of the pipeline qualifies, and enquiries that were never going to buy stop consuming rep time. Several meetings a month then come from hours nobody was working.

Number Four: Human Minutes per Enquiry

This is the labour number, and the one finance cares about. Track how many minutes a person spends on an enquiry from first touch to handover — reading, re-typing details into the ERP, chasing a price, drafting a reply. Twenty-plus minutes is normal when the source is a chat thread. Five to seven is normal when the agent has already written the structured record and the rep is only reviewing it.

Number Five: Cost per Handled Enquiry

Total sales-operations cost — salaries attributable to enquiry handling, tooling, model usage and infrastructure — divided by enquiries actually handled. Do the accounting honestly, including the retainer for whoever maintains the guardrails; understating this figure is how automation projects get abandoned in year two. High single digits of dirhams per enquiry is achievable for volume-driven desks, against a manual baseline frequently four to five times that.

A Baseline You Can Build in One Week

• Export ninety days of enquiry timestamps per channel — WhatsApp, email, web form, phone notes. At least one channel usually has no timestamp discipline, which is itself the finding.
• Compute the distribution, not the mean — median and ninetieth percentile of first-response latency, split by working hours, evenings and Friday–Saturday. Two hours in a spreadsheet.
• Count meetings, not leads — meetings sourced from enquiries in each time band. If after-hours meetings are zero, the pipeline is clipped by schedule, not by demand.
• Time five replies by hand — stopwatch one rep on a real enquiry. The gap between perceived and actual handling time is the number that justifies the project internally.

What the Numbers Will Not Tell You

• Whether the relationship deepened. Speed wins the first reply; trust is built on the site visit, which no dashboard captures.
• Tone quality at the margin. Read twenty agent transcripts a week yourself. Metrics stay green while phrasing drifts bureaucratic.
• The enquiries that never arrived because someone looked at a slow reply and left. Absence has no row in the export.
• Pricing exceptions. Volume discounts and project-specific terms need a named approver, not an inferred one.
• When the buyer wanted a person. Some enquiries are a relationship request in disguise; escalation needs to be instant and graceful.

Where the Gains Usually Hide

Four pockets consistently hold more value than the complaint that triggered the project.

Reading Your First 90 Days

1. Weeks 1–2 — instrument only. Measure the five numbers on the current desk. Change nothing. This baseline is the only honest comparison you will ever get.
2. Weeks 3–5 — connect the record, not the chatbot. Pricing, stock and lead times must be readable by the agent, or it will answer confidently and wrongly.
3. Weeks 6–9 — go live on one channel. WhatsApp first for GCC buyers, with a human escalation path wired and tested before the first real customer touches it.
4. Weeks 10–13 — review transcripts weekly and adjust thresholds. The agents that stay useful are the ones someone is still reading.