The AI customer concierge: what you are actually buying

That last clause is the one to take into your next vendor call. Deflection is a cost metric. If the AI you are being sold is framed entirely as a way to stop customers reaching you, you are being asked to optimise the wrong end of your business.

What makes it a concierge and not a bot

The word concierge is doing real work in the definition above. A hotel concierge is not an information booth. They do not just tell you where the restaurant is; they book the table, remember that you prefer the terrace, and when your flight is cancelled at midnight they wake up somebody who can fix it. Anticipation, action and escalation. Most businesses already have the FAQ bot. Very few have the concierge.

1. It knows your business, not the internet

A concierge is grounded in your catalogue, your pricing, your delivery windows, your refund policy and your past conversations. It does not improvise a shipping rate. When it does not know, it says so and fetches someone who does — which, counter-intuitively, is what makes customers trust the answers it does give. If you want the mechanics of how that grounding works, connecting an agent to your knowledge is the shortest version.

2. It works where the customer already is

WhatsApp at 11pm. An Instagram DM under a reel. A comment on a Facebook ad. Live chat on your pricing page. A reply to a shipping email. Same brain, same memory, five doors. The customer who DMs you and then emails a week later is one person, and an agent that treats them as two strangers will be noticed — usually out loud.

3. It does things, not just says things

Qualifying a lead, checking stock, holding an item, booking an appointment, raising a ticket, taking a deposit. If the outcome of every conversation is only ever a nice reply, you bought a writing tool with a chat interface. This is the whole distinction between agentic and generative AI, and it is the single most expensive thing buyers get wrong.

4. It hands over cleanly

The most important feature in the entire product is the one that turns the AI off. When someone is angry, when the order is worth four figures, when the question is genuinely novel — a real person steps in with the full thread already in front of them, and nobody has to start again. Get this wrong and one bad handover undoes every second the AI saved. There is a whole design discipline to it, covered in AI and human support, paired.

Three things get sold under one name

Vendors use “chatbot”, “AI agent” and “AI concierge” interchangeably, which is convenient for vendors and expensive for you. Here is the honest split, and it is worth having open during a demo.

If you want this comparison at full length, with the failure modes written out, AI agent vs chatbot takes it line by line, and AI agent vs live chat vs helpdesk covers the adjacent question of whether you need a helpdesk at all.

The gap has a price, and you can calculate it

Three findings have been repeated so often in sales research that they have stopped being interesting and started being the whole game. Leads contacted within five minutes are roughly twenty-one times more likely to qualify than leads contacted at thirty minutes. Around 78% of customers buy from whoever responds first. And a widely cited audit of 2,241 companies found an average response time of 42 hours, with 23% never responding at all.

Read those together and the conclusion is not flattering to any of us. Most businesses are not losing to a better competitor. They are losing to a faster one. Speed is not a feature of the product — on messaging channels, speed is the product.

Slot your own numbers in and the shape of the answer rarely changes: the value sitting in unanswered conversations is almost always a large multiple of what answering them would cost. That is not a software argument, it is an arithmetic one, and it is why the honest recommendation is to test on a free tier and count rather than to take anyone's word for it — ours included.

The three pricing models, and which one bites

This is where buyers get hurt, months after signing, and it has almost nothing to do with the headline number on the pricing page. What matters is how the bill behaves when things go well.

There is a psychological trap in metered pricing that is worth naming plainly. Once the meter is running, every instinct pushes you to narrow the agent's scope, shorten its answers and route more conversations away from it — which is the exact opposite of what you bought it to do. Jugl publishes four flat tiers with the AI included and a free tier that does not expire, specifically so the tool never has to argue with your growth. If you want the metered maths done against real published rates, the per-resolution pricing breakdown and the usage-based pricing comparison both model it at volume.

Twelve questions to score any vendor with

Print this, or just tick it as you go. It works on any product in the category, and it is deliberately written so that a vendor cannot pass it with a good demo — every line is something you can ask them to prove in ten minutes.

The free agent question, answered fairly

If you run a business on WhatsApp or Instagram you have already seen the prompt: the platform's own free AI agent, built into the app, live in minutes, no dashboard and no developer. Let us be fair to it, because fairness is more useful to you than a hit piece.

It is genuinely good at what it does. Setup friction is close to zero, it costs nothing to start, it lives natively in the apps your customers already use, and it does hand off to a person when it gets stuck. For a solo operator handling a modest number of conversations a day entirely inside those apps, turn it on. You will learn something about your own inbox in a week.

The ceiling only becomes visible when you grow, and it is a sharp one. Free platform tiers typically have no shared team inbox — one person sees everything, with no assignment, no routing and no service levels. No CRM connection, so nothing links a conversation to a customer record or an order. No API or webhooks. And most consequentially: they work only inside the platform's own apps. If a customer emails you, or lands on your pricing page, the agent you set up does not exist there.

That last point is the one that catches most buyers, because for a great many businesses the website is where the highest-intent traffic lands. The person reading your pricing page has already got past the ad and onto your property. A platform-locked agent leaves that half of your funnel uncovered while showing you a dashboard that says everything is handled. The full side-by-side is in Jugl vs Meta Business Agent.

What it actually does all day

Abstract definitions do not sell anything and do not help you evaluate. Here is the same technology in five different businesses — the standard, repeatable jobs a concierge takes over.

E-commerce — the 11pm stock question

“Is this in the blue, medium?” is the most common message an online store receives, and it is almost always sent from a sofa after nine. A concierge reads the product the customer is looking at, checks live inventory, answers with the variant and the delivery date, offers to hold it and sends the payment link. What it replaces: the morning-after reply that arrives after the customer has bought elsewhere. The mechanics, including returns and order status, are in automating e-commerce with AI agents.

Home services — qualifying before you drive

A plumber gets thirty enquiries a week; perhaps twelve are inside his area with jobs worth quoting. A concierge asks the qualifying questions — postcode, property type, urgency, photos of the issue — attaches them to the lead and books the site visit into the calendar. What it replaces: two hours a night of thumb-typing, plus every wasted van trip.

Clinics and salons — the booking and the no-show

Appointment businesses lose money twice: once when a slot never gets filled, and again when it gets filled and nobody turns up. A concierge handles booking, rescheduling and confirmation in the same thread the customer used to enquire, in their own language, without a phone call. What it replaces: the front desk answering the phone while three people wait at the counter.

B2B and software — routing by intent, not by form field

Not every “can I get a demo” is worth the same. A concierge asks about team size, current tooling and timeline, answers the technical questions it can, then routes a 400-seat prospect to a human immediately and a 3-seat prospect to self-serve signup. What it replaces: a contact form, a two-day queue and a lot of guessing. Related: sales automation and turning website traffic into revenue.

Hotels, restaurants and travel — the original concierge job

Opening hours, dietary questions, group bookings, parking, whether the terrace is open in this weather. Endless, repetitive, and every unanswered one is a table that goes to the place down the road. What it replaces: a manager reading DMs during service. The hospitality version, including upgrades and multilingual guest service, is in AI agents for hotels.

Notice what none of these are: replacing your team. In every case the concierge takes the first layer — repetitive, out-of-hours, low-judgement — so the people you already pay spend their day on the conversations that need a person. What an AI agent for customer support can and cannot do draws that boundary properly.

Why Jugl, in plain terms

We built Jugl around the gaps above, so the argument for it is specific rather than adjectival. Five things.

Every channel, not just one platform's. Jugl runs across WhatsApp, Facebook, Instagram, your website and email — one agent, one memory, one inbox. The customer who DMs you on Instagram and emails a week later stays one conversation with one history.

Trained on your business, in your voice. Your catalogue, policies, past conversations and tone, not a generic assistant with your logo on it. It answers like your best employee on their best day.

Built to sell, not to deflect. Jugl's agents detect buying intent, recommend from your catalogue, qualify leads, schedule appointments and prevent churn inside the conversation. The metric is revenue, not ticket volume — which is why the calculator above measures recovered orders rather than deflected tickets.

The handoff is a feature, not an apology. Conversations pass to your team with the context intact, on triggers you define. You are not building a wall between you and your customers; you are building a filter.

A Meta Business Partner, so you keep the reach without the ceiling. Jugl runs natively on the WhatsApp and Instagram rails you already use — and then keeps going, across your website and email, with the shared team inbox, CRM structure and published flat pricing that free tiers do not offer.

The honest recommendation, and it has not changed: if you are one person answering a handful of messages a day inside one app, use the free platform agent. The moment a second person starts answering messages, your website starts producing enquiries, or a single customer is worth more than one order — you have outgrown it, and that is the day a concierge starts paying for itself rather than costing you. If you are weighing specific vendors, the full comparison set covers Zendesk, Freshdesk and Gorgias with the pricing worked through.

How to buy one and be live this week

This is not an IT project. It is a few hours of thinking and about twenty minutes of clicking — and the thinking is the part that decides whether it works.

1. 01Pull your last two hundred messages and list the ten questions that repeatThis is both your training set and your acceptance test. Almost every business finds that ten questions cover the large majority of inbound volume. If a vendor cannot handle those ten from your own material, nothing else on their feature list matters.
2. 02Connect the two channels that produce the most enquiriesUsually WhatsApp and Instagram, or your website chat and email. Resist switching on all five at once — you want a volume of transcripts you can actually read in the first week.
3. 03Point it at the truth, not at a promptYour website, catalogue, pricing, delivery and refund policy, and the systems that hold live answers: orders, calendar, CRM. The most common cause of an AI saying something wrong is being trained on something out of date. Fix the source, never patch the prompt.
4. 04Write the no-go list before launchName what the agent must never handle on its own: refunds over a threshold, complaints, anything legal or medical, any customer who asks for a person, any order above a value you choose. Rules written before go-live are policy. Rules written after an incident are damage control.
5. 05Launch narrow and read every transcript for a weekAn hour a day, reading real conversations and correcting misses, will produce a better agent than a quarter of configuration by a team that never reads one. This is the step buyers skip and then blame the software for.
6. 06Measure revenue, not deflectionFirst-response time, out-of-hours conversations answered, leads qualified, bookings taken, orders recovered. If those numbers are not moving inside a month, change the configuration — or change the vendor. Deflection rate is a cost metric that tells you nothing about whether you sold anything.

Questions buyers ask before they sign