The 5-Minute Sales Rep: An AI Agent That Answers Every Enquiry in Seconds

The 9:47 PM water heater: why the first reply wins

At 9:47 on an ordinary weeknight, a homeowner's water heater gives up. She finds a plumber with good reviews on her phone, fills in the "request a callback" form, then repeats the same form on two competitor sites for good measure. Five minutes later, at 9:52, one of them texts back — not a flat "we'll be in touch", but a genuine reply: it acknowledged the problem, asked two short clarifying questions, and offered a technician between 8 and 10 the next morning. She books on the spot. The other two respond late — one late the next morning, one not at all. (The story is told this way, with the 9:47 timestamp, in a practitioner piece on Medium this month: "The 5-Minute Sales Rep".)

The winner wasn't the better plumber. It was the one that answered first.

That story distils one of the oldest lessons in selling, and it is the reason this post exists. It is also, as of January this year, a job description that Meta's WhatsApp rules explicitly still allow an AI agent to do. The research says the first minutes decide who wins the customer. The policy says the business-scoped agent is the permitted shape. The engineering doctrine says the winning build is the simplest system that fits. And this month's market news says small businesses are already buying exactly this worker.

This post fuses all four into one thing: a job spec for the 5-minute sales rep that you can write this afternoon.

Speed-to-lead: the research behind the story

The water-heater story feels like folk wisdom. The numbers behind it are sharper — and they come with caveats worth stating plainly.

Finding 1: five minutes vs thirty

The most quoted speed-to-lead finding is from a 2007 InsideSales study of six companies' phone logs: the odds of contacting a lead drop about 100 times when you wait 30 minutes instead of 5; the odds of qualifying it drop about 21 times (expertise.ai, which links every stat to its original source).

Caveats, because they matter: it was a vendor study of six companies, presented at a 2007 conference, and it never measured close ratios — it measured contact and qualification. Treat the multipliers as direction, not promises. (UseCarly carries the same caveats.) Note also that the famous "78% of buyers go with the first responder" has no traceable primary source — it is debunked, and this post won't use it.

Finding 2: an hour vs a day

A 2011 Harvard Business Review audit of 2,241 US companies found roughly a sevenfold qualification advantage for firms that reached out within the hour, compared with firms that took longer. The same audit found the average response took 42 hours — and 23% of companies never responded at all (Brady Mills).

Finding 3: almost nobody is fast enough

Under 1% of first call attempts happen within five minutes (XANT, 2021), and only about 7% of tested B2B SaaS companies responded to an inbound enquiry that fast (Drift, 2017) — which means a five-minute reply puts you ahead of roughly nine in ten competitors (expertise.ai).

So the research agrees with the water heater: speed decides, and almost the entire competitive set is slow.

The question is who can afford to be fast — because a ten-person business can't post someone by the phone at 2 AM for the handful of leads that might drift in.

Two businesses, one enquiry: the lead-response comparison

Put the research into one concrete comparison. The same enquiry lands at 10 PM on a WhatsApp Business number. Two shops handle it differently.

The tomorrow shop isn't underperforming. It runs the way almost every small business runs — coverage during business hours, silence outside them — and the research says that pattern loses the water-heater customer every time. The 5-minute shop isn't a bigger team. It's the same team with one worker that never sleeps: the lead-response agent.

Why this lead-response agent is allowed on WhatsApp

Here is where most AI-on-WhatsApp conversations get nervous, and where this one gets simple. Meta updated the WhatsApp Business Solution terms in October 2025, effective for all accounts on January 15, 2026. The change bans one specific thing: AI providers offering general-purpose assistants — a bot that answers questions on any topic and is itself the product — as a distribution channel over WhatsApp. ChatGPT, Perplexity, Copilot as open-ended assistants: out. But AI as a supporting part of a defined business function — a retailer's order-status bot, a clinic's appointment system, support triage — is explicitly unaffected.

The independent explainers converge on one line: "the bot as product is out, the bot as business tool is in" (Wamafy, Singoo Cloud).

Read the "still allowed" list closely, because the 5-minute sales rep sits directly on it. Wamafy's explainer lists what businesses may still do, and the list includes:

• Answering questions about your own products and services
• Booking appointments
• Qualifying leads and routing them to a person — word for word

That is the job spec of this post: qualify the lead, route the hot one to the owner.

Meta's rule doesn't merely tolerate this agent. It describes it.

Two honest notes on the rules. First, this is a policy reading, not legal advice — Meta's terms are the final word for your account.

Second, there is a cost shape to respect, and it changed this month: since October 1, free-text replies inside the 24-hour service window are billed as service messages, and business-initiated template messages have been billed per message since Meta's July 2025 per-message pricing change (Wamafy, Singoo Cloud). We publish no figures here, per our standing rule — the qualitative point is what matters for the build: keep the agent tight, not chatty. Every pointless reply is a billed reply. The guardrails in the build section below are partly about this.

The simplest AI agent system that fits

Now the engineering doctrine, because it tells you what not to build. Anthropic's "Building Effective Agents" — still the standing reference, verified verbatim on the live page this month — is blunt about where successful agents come from: the most successful implementations "weren't using complex frameworks or specialized libraries. Instead, they were building with simple, composable patterns." The standing rule: "finding the simplest solution possible, and only increasing complexity when needed." And in its Appendix 1, the teams found agents add the most value for tasks that do four things: "require both conversation and action, have clear success criteria, enable feedback loops, and integrate meaningful human oversight" (Anthropic).

Score the 5-minute sales rep against those four criteria, and it fits almost suspiciously well:

• Conversation and action. It talks (answers, asks the qualifying questions) and acts (books the appointment, logs the answers, routes the lead). Neither a chatbot that only chats nor a script that only books would do the job.
• Clear success criteria. Two numbers: median seconds-to-first-reply, and qualified leads handed to the owner per week. You can measure both without an AI degree.
• Feedback loops. The owner reads the week's transcripts and fixes the answers — the agent gets better the way a new hire does.
• Meaningful human oversight. The agent qualifies; the owner closes. The human stays at the end of the line, which is also what keeps the agent on the right side of Meta's rule.

The practitioner world agrees on the shape. "Reliability comes from architecture, not just a better model" (SuperMia), and the most common mistake is teams jumping "straight to multi-agent systems because they sound more powerful... Start with a single agent" (IABAC). Anthropic even names the exact workflow pattern: routing — "classifies an input and directs it to a specialized followup task." Qualified lead → owner. Cold lead → follow-up path. Spam → parked. That is the whole architecture: one agent, one routing step, one human at the end.

And this month, the market is buying exactly this worker.

On October 13, Dapta announced a $6M seed round (led by Chingona Ventures) for "Dapti," an AI sales agent that "instantly answers, qualifies and schedules meetings without human intervention the moment a prospect reaches out" — the company's own announcement numbers: 200K+ users, about 4K of them new each week, $4M in annual recurring revenue, and contact rates doubling for companies using it (GlobeNewswire). All of those figures are company-reported via its funding announcement — treat them as the company's story, not an audited fact. Closer to home, Kraya AI reports 1,000+ Indian SMBs running WhatsApp-first enquiry response, AI qualification, and automated follow-up sequences across real estate, education, healthcare, fitness, and logistics — again company-reported, via its September announcement (Latestly).

The pattern is the same one the research demands and the doctrine blesses: answer, qualify, route.

One more design line worth stealing, from an Indian deployment write-up of an outbound voice agent for lead qualification: the agent "doesn't try to close the deal. It simply qualifies and routes" — lifting daily qualification capacity from ~250 to 1,000+ calls, the qualification rate from ~5% to 12%, and quintupling qualified leads escalated per day. That write-up is vendor-published with no confirmed date, so read it as a design anecdote, not evidence — but the design is right: qualify, don't close. The owner closes. (MyOperator)

The lead-response agent job spec: one worker, four jobs

Everything above collapses into a one-page job spec. This is the worker: one agent, one job, four steps in sequence, every time an enquiry arrives.

Step 1 — Answer

Every enquiry gets a first reply in seconds, any hour of the day or night. Not a canned "we'll be in touch" — a response that picks up the exact problem they described and engages with it. A fast answer also does relational work: it tells the lead they're being taken seriously, which buys the minutes the qualifying questions need.

Step 2 — Qualify

The agent leads with the 2–3 triage questions your best salesperson would ask first — budget band, timeline, scope. No discovery-call depth; just enough to tell who's buying from who's browsing. This is the "qualifying leads" half of Meta's permitted shape.

Step 3 — Hand off

The routing step. A qualified lead goes to the owner with the answers attached — no "call this number," the full picture. An ambiguous or cold lead goes to the nurture path, not to your phone. Spam gets parked. This is the "routing them to a person" half — the human oversight Anthropic's doctrine requires.

Step 4 — Follow up

Leads that go quiet get a structured sequence: one nudge, one value-add, one final check — then parked. A plain, repetitive run of reminders that humans always intend to send and never do — and that wins back a large slice of otherwise-lost revenue. The worker holds every thread, never gets swallowed by a busy week, and sends the third nudge without the awkwardness a person would feel.

That is the spec. Here it is as the one-page card you will actually fill in — copy this, print it, pin it above the desk:

THE 5-MINUTE SALES REP — ONE-PAGE JOB CARD
==========================================
Business:  <your business name>
Channel:   WhatsApp (business number)
Date:      <today's date>

JOB 1 — ANSWER (in seconds, any hour)
First reply goes out immediately and references what they asked.
Never a flat "we'll be in touch."

JOB 2 — ASK (2-3 questions, one at a time)
1. <your question 1>
2. <your question 2>
3. <your question 3>

JOB 3 — HAND OFF (route, don't chat on)
QUALIFIED → owner, WITH the answers attached: <your handoff rule>
COLD / AMBIGUOUS → follow-up path, not the owner's phone
SPAM / WRONG NUMBER → parked, no reply beyond one polite line

JOB 4 — FOLLOW UP (the gone-quiet sequence)
Day 3:  one nudge
Day 7:  one value-add (useful, not "just checking in")
Day 14: final check, then park

SUCCESS CRITERIA (measured weekly)
□ Median seconds from enquiry to first reply
□ Qualified leads handed to owner per week

DOES NOT DO (the fence)
□ General chit-chat — only our products, prices, booking
□ Promises on price, delivery date, or availability
□ Anything outside <your products/services>
□ Cold outreach — it only answers inbound enquiries

Notice what the card is not: it is not a technology choice. There is no model name on it, no vendor, no "AI platform." That is deliberate, and it is Anthropic's doctrine applied — the simplest system that fits. The card is the product; whatever software runs it is an implementation detail. Write the card first. Choose the tooling after.

Build your lead-response agent this afternoon: fill in the job card

Here is the follow-along build: one afternoon, no code, seven steps. Each step states the action and what you should have at the end of it.

Use the worked example where it helps — it is fictional, a composite, not a real business: a 10-person modular-kitchen studio in Bengaluru whose WhatsApp enquiries arrive at 10 PM, long after the showroom closes.

Step 1 — Measure your current reply time honestly

Before you write anything, find the truth. Tonight at 9 PM, put yourself in the lead's shoes: submit your own website's enquiry form or WhatsApp the business number from a personal phone, and clock how long the reply takes. Then look back through your chat history — for each of your last ten enquiries, measure the gap between the lead's message and your first reply. Almost every owner guesses wrong here, and far too optimistically — the Medium piece's implementation playbook starts with this measurement for a reason: you can't put a value on the fix without the real number in front of you.

Expected result: one honest number ("we reply in 4 hours on average; nights and Sundays go unanswered until morning").

Step 2 — Write your 2–3 qualifying questions

Bring in your best salesperson — or channel one — and write the two or three questions that reveal whether you're talking to a buyer or a browser. Not ten questions. Two or three. The test: if a stranger answered only these, would you know whether to call them back tonight?

QUALIFYING QUESTIONS WORKSHEET
==============================
My business: <what you sell, in one line>

Question 1 (the money question — can they afford us?):
  <e.g. "What budget have you set aside for this?">

Question 2 (the timing question — when does this happen?):
  <e.g. "When do you need this done by?">

Question 3 (the scope question — what exactly do they need?):
  <e.g. "How many rooms / what size / which service?">

Rule: if a lead answers all three, they are qualified.
Rule: if a lead dodges the money question twice, they go to
the follow-up path, not to the owner.

Worked example (fictional — the Bengaluru modular-kitchen studio): Question 1, budget band — "Kitchens like ours usually run in three bands; which is closest to yours?" Question 2, timeline — "When are you hoping to have the kitchen ready?" Question 3, scope — "Is this a full kitchen or a renovation of specific units?" If all three land, the agent books the site visit.

Expected result: three written questions, in your own words, that your best salesperson nods at.

Step 3 — Write the handoff rule

Decide exactly what "qualified" means and exactly what the owner receives. Vague handoffs ("send me the good ones") are where deployments die. Write the rule on the job card, and write the exact message the owner gets:

HANDOFF MESSAGE (sent to the owner, every time)
===============================================
🔥 Qualified lead — <name>
Wants: <scope from Q3> | Budget: <band from Q1> | By: <date from Q2>
Contact: <phone number> | First reply sent in <N> seconds
They expect your call. Full transcript attached.

RULE: the owner replies to qualified leads within 2 hours in
business hours. The agent never sends this message for cold,
ambiguous, or spam leads — those go to the follow-up path.

Expected result: one rule and one message template, exact wording, ready to paste.

Step 4 — Write the follow-up sequence

The gone-quiet leads get three touches, then rest. Write the actual messages — short, specific, human. The October billing change is the reason to keep them short: every pointless reply is a billed reply.

FOLLOW-UP SEQUENCE (gone-quiet leads only)
==========================================
Day 3 — the nudge (one line, references their question):
"Hi <name>, you had asked about <their question>. Still
looking? Happy to help whenever."

Day 7 — the value-add (useful, not "just checking in"):
"Sharing this in case it helps — <one useful thing:
a price guide, a completed-project photo, a checklist>."

Day 14 — the final check, then park:
"Hi <name>, I'll stop here so I'm not a bother. If the
<kitchen/order/project> comes back on the table, just reply
and we'll pick up where we left off."

Expected result: three messages, exact wording, saved where the person who runs the tool can paste them.

Step 5 — Write the starter prompt

This is the artifact the whole build hangs on: the copy-paste starter prompt for the agent. It encodes the job card in the agent's own instructions. Never describe a prompt — print it. Here it is, with the fictional studio filled in so you can see the shape; replace the bracketed parts with yours:

YOU ARE the first responder for <Studio Name>, a 10-person
modular-kitchen studio in Bengaluru. You answer inbound WhatsApp
enquiries. Your job is to respond fast, qualify, and hand off —
nothing else.

YOUR FOUR JOBS, IN ORDER:
1. REPLY within seconds to every enquiry, any hour. Reference
   what they asked about. Never send a flat "we'll be in touch."
2. ASK the qualifying questions below, ONE AT A TIME, in order.
   Wait for each answer before asking the next.
3. When all answers are in, STOP and hand off. Send the owner the
   handoff message with every answer attached. Do not keep chatting.
4. If a lead goes quiet after your first reply, send the follow-up
   sequence on schedule. Nothing more.

QUALIFYING QUESTIONS (one at a time):
1. "Kitchens like ours usually run in three bands — which is
   closest to yours?" [replace with your money question]
2. "When are you hoping to have the kitchen ready?"
   [replace with your timing question]
3. "Is this a full kitchen or a renovation of specific units?"
   [replace with your scope question]

HANDOFF RULE:
- All three answered → qualified → handoff message to owner.
- Money question dodged twice → follow-up path, not the owner.
- Spam, wrong number, or abuse → one polite line, then park.

THE FENCE (never cross):
- Answer ONLY about our kitchens, prices, and site-visit booking.
- NEVER answer general questions (weather, news, anything else).
  If asked, say: "I only handle <Studio Name> kitchen enquiries —
  want me to connect you to the team?" then hand off.
- NEVER promise a price, delivery date, or slot you were not given.
- Keep every reply to two lines or less. Tight, not chatty.

SUCCESS: every enquiry answered in seconds; qualified leads reach
the owner with answers attached.

Expected result: the prompt saved in a document, bracketed parts replaced with your business's words, reviewed once by your best salesperson.

Step 6 — Run the two-guardrail check

Before anything goes live, check the card against two guardrails.

Guardrail 1 — the Meta shape

The agent must stay business-scoped: your products, your prices, your booking — never general chit-chat. The test from the policy section: does it do a named job ("qualifies leads and hands them to the owner")? Would it refuse a question about something you don't sell? The starter prompt's fence handles both. This is what keeps you on the "business tool" side of Meta's January 2026 line. (One related note, kept to a line since it has its own subject: this worker is text-only — it answers messages, it doesn't place calls.)

Guardrail 2 — the cost shape

Since October 1, in-window free-text replies are billed as service messages, and template messages have been billed per message since July 2025 — so the agent must be tight, not chatty. The two-lines-or-less rule in the prompt isn't style advice; it's cost control. Count the replies per conversation when you review transcripts in week one, and cut any that don't move the lead forward.

Expected result: the fence section of your job card initialled — literally, put your initials on it — and the two-lines rule in the prompt.

Step 7 — Set the two numbers and the weekly review

The job card's success criteria become a weekly 15-minute habit. Every Monday: pull the median seconds from enquiry to first reply, and count the qualified leads handed to the owner that week. Then read a sample of actual transcripts — the Medium piece's playbook calls this the most-skipped review step and the highest-leverage one: fifteen minutes a week inside real conversations catches misdirected leads, clumsy wording, and qualifying questions that need retuning. Run it like a new hire: short weekly coaching, with the conversation log as the raw material.

Expected result: two numbers tracked weekly, and a recurring 15-minute calendar block titled "read the agent's transcripts."

That is the whole afternoon: one measured number, three questions, one handoff rule, three follow-up messages, one starter prompt, two guardrails, two numbers. Nothing here costs money. The card is the product; the software is an implementation detail.

This is what custom AI agents work is really about at small-business scale: not a big risky project, but one worker with one job, designed to the permitted shape and the simplest system that fits.

Two lead-response numbers to watch

If you take nothing else from this post, take the two numbers — because they are also the answer to the objection this whole post is built to defeat: "an AI agent sounds like a big, risky project." The 5-minute sales rep is the smallest agent that provably pays: the one that stops you losing enquiries you already earned.

Median seconds-to-first-reply tells you whether the worker is doing its job. Qualified leads per week tells you whether the job matters.

When both move in the right direction for a month, you have your proof — measured, not promised. The 2007 and 2011 studies gave you the direction; your own two numbers give you the verdict for your shop.