AI sales assistant for small business enquiries

Most small businesses don't lose sales because the product is wrong. They lose them in the gap between someone asking and someone answering. A message about a garden room lands on Instagram at 10 pm. A caller wants to know whether you stock a part, gets voicemail and rings the next shop. A website visitor has three questions about your packages, can't find the answers and closes the tab.

An AI sales assistant closes that gap. It answers buying questions from your own product and service information, asks the questions you'd ask to work out whether an enquiry is worth your time, and saves the person's details so you can pick it up properly. Klugon does this on your website chat, on the phone and in messaging apps, and records every enquiry as a lead in a built-in CRM.

This page covers how lead qualification works in practice, what gets recorded and where it goes, how the AI answers product and service questions, and a clear list of what it doesn't do. That last part matters. A sales assistant that overpromises, or that you think is doing follow-ups when it isn't, costs you more than it earns.

What an AI sales assistant does in Klugon

Klugon isn't a separate sales product. It's the same AI receptionist that answers your calls and chats, with a set of skills switched on that make it useful for selling.

It works across these channels:

• Website chat, through the AI chatbot for your website, pasted onto any site with one script tag.
• Phone, through the AI voice agent, which can search products, answer from your knowledge base and capture leads on a call.
• WhatsApp, through the WhatsApp AI chatbot, once you've connected your WhatsApp Business account through Meta.
• Instagram and Facebook Messenger, through Instagram DM automation, after Meta setup.

All channels share one knowledge base and feed one inbox, so a customer who messages on Instagram and then rings gets the same answers, and both conversations show up in the same place for you.

AI lead qualification: asking the right questions

Qualification is working out, early, whether an enquiry is likely to turn into work you want. A good receptionist does it without the customer noticing: a few friendly questions, in the right order, that tell you the size of the job, the timing and whether the work suits you.

An AI sales assistant can do the same, but only if you tell it what to ask. It doesn't know that a wedding photographer needs the date before anything else, or that a kitchen fitter needs the postcode to know whether it's worth quoting.

How you set it up

In the assistant's settings you have three tools for this.

• Custom instructions. Plain-language instructions the AI follows on every channel. This is where your qualifying questions go, along with the order to ask them and anything that should end the conversation politely (a postcode you don't cover, a job you don't do).
• Lead capture details. The details you want collected when a customer shows interest, such as name, phone and email. The AI asks for them naturally, one or two at a time, and doesn't insist. You can also tell it to wait for a number of messages before asking for contact details, so it doesn't open with "What's your email?"
• Note buying intent. When this is switched on, the AI adds a note to each lead saying whether the person seems high, medium or low intent, and why.

The intent note is the AI's reading of the conversation, not a score worked out from data. Treat it as a quick guide for which leads to ring first, not as a verdict.

Qualifying questions that work

Keep it to three to five questions. More than that and people drop out. Ask the one that rules people out first, so neither of you wastes time.

A trades business (kitchens, bathrooms, extensions):

1. What kind of job is it, and roughly how big?
2. What's the postcode? (Rules out jobs outside your area.)
3. When are you hoping to have it done?
4. Is the property owned, or would the landlord need to agree?
5. Best name and number for a survey visit.

A wedding or events photographer:

1. What's the date? (Rules out dates you're booked.)
2. Where's the venue?
3. Roughly how many guests, and how much of the day would you like covered?
4. Have you seen the packages on our website, and is there one you're leaning towards?
5. Name, email and phone, so we can send availability.

A gym or fitness studio:

1. What are you hoping to work on?
2. Have you trained with a coach or in a class before?
3. Which days and times suit you?
4. Would you like to book a free trial session?

A professional services firm (accountants, solicitors, consultants):

1. What do you need help with? (Kept short; no detail about the matter yet.)
2. Is this for you personally or for a business?
3. Is there a deadline?
4. Name, email and phone, so the right person can get back to you.

For regulated work, keep the AI well away from advice. It collects the basics and passes the enquiry to a person. Solicitors in particular should tell the AI not to take details of the matter until a conflict check has been done. The professional services page goes through this in more detail.

Wording the instructions

Write instructions as you'd brief a new member of staff on their first day. Something like:

Specific instructions beat general ones. "Qualify leads" means nothing to the AI. "Ask for the postcode before anything else" does. For more on this, see what to teach your AI receptionist.

Qualifying on the phone compared with chat

The same instructions apply on every channel, but people behave differently on each, and it's worth adjusting for that.

On a website chat or in WhatsApp, people are happy to answer a short run of questions, and they can type an email address without anyone mishearing it. They also drift away mid-conversation, so ask the most useful question first and save contact details before the conversation tails off.

On the phone, callers expect to be asked things, but long lists feel like an interrogation. Two or three questions are about the limit before the caller wants an answer of their own. Spoken details are also harder to catch. Postcodes, email addresses and surnames are where speech recognition slips most often, so tell the AI to read them back, and prefer a phone number over an email address on calls.

On Instagram and Messenger, many enquiries start with one line, often just "price?". People there are often browsing rather than ready to buy. A short, useful answer with one follow-up question tends to work better than launching straight into qualification.

If one channel matters much more to your sales than the others, write the instructions for that channel first and test them there.

Common mistakes when setting up qualification

• Asking for contact details too early. Someone who's just asked the price and is immediately asked for their email often leaves. Answer the question first, then ask.
• Hiding the price. If you publish prices, let the AI quote them. Refusing to give a rough figure until someone hands over their number puts off more people than it catches.
• No way out. Tell the AI what to say when someone doesn't fit: a job you don't do, an area you don't cover, a budget well below your minimum. A polite "that's not something we do" saves both sides time.
• Too many questions. Five is a lot. If you can only act on three of the answers, ask three.
• Notes left out. If the answers aren't written into the lead's notes, your team will ask the same questions again on the callback. Say in the instructions exactly what goes in the notes.

Answering product and service questions

Buyers ask the same things over and over. What's included? Is it in stock? How long does it take? Does it come in blue? What's the difference between the two packages? Every one of those answered at 10 pm is one fewer reason for someone to go elsewhere.

Where the answers come from

The AI answers from three places.

• Your product catalogue. Products are added by hand in the dashboard, with names, descriptions, prices and images. The AI searches them on chat and on phone calls. There's no import from a spreadsheet and no sync with an online shop, so you keep the catalogue up to date yourself. You can export it as CSV.
• Your services. The services you set up for booking, with their lengths and prices.
• Your knowledge base. Documents you upload (PDF, Word, plain text, Markdown or CSV, up to 20 MB each), text you paste in, individual web pages, and a scan of your website's homepage plus up to seven key pages. Scanned PDFs that are only images can't be read.

All of a business's assistants draw on the same knowledge base. If you run a sales assistant on your website and a receptionist on your phone line, they give the same answers.

A web page you import is read once. If you change your prices on the website, add the page again so the AI picks up the new version.

What good answers need

The AI can only be as accurate as what you give it. The most common gaps we'd expect in a sales setup are:

• What's included and what isn't. "Does the price include fitting?" needs a written answer.
• Lead times. "How long until you can start?" If it changes week to week, tell the AI to say it varies and offer to take details.
• Delivery and collection areas, and their costs.
• Comparisons between your own options. A short paragraph on who each package is for helps the AI recommend the right one.
• Warranty, returns and cancellation terms.

When the AI can't find an answer, it should say so and offer to take the customer's details, not guess. Klugon can't guarantee it never gets something wrong, so keep the knowledge base clear and check the Insights page. It lists "questions without a good answer" and the most asked questions over the last 30 days, which is a ready-made to-do list for what to add.

Suggesting related products

You can switch on recommendations so the AI suggests related products or services from your catalogue. You choose when it makes suggestions (for example, after answering a question), what kind of suggestion to lean towards, and how many suggestions at most. It doesn't build bundles or change prices; it just mentions things that fit.

Done well, this is the shop assistant who says "you'll need the wall bracket for that". Done badly, it's pushy. Start with one suggestion at most and read a few conversations before raising it.

Mentioning a current offer

If you have a promotion running, you can give the assistant one promo code and its terms to mention to customers. Its instructions tell it not to invent other codes or discounts. The code isn't applied anywhere automatically by Klugon, so make sure it works wherever customers actually pay you.

Capturing leads into Klugon's built-in CRM

Every lead the AI captures goes into Klugon's own CRM, with no setup. This is the part of the AI sales assistant that turns a conversation into something your team can act on.

What gets recorded

When the AI saves a lead, it records:

• the person's name
• their email and phone number, whichever they gave
• what they're interested in, such as a booking, a product or a general enquiry
• the channel the enquiry came from
• notes, including your qualifying answers and the buying-intent note if that's switched on
• whether they agreed to marketing, recorded only when they say so explicitly

The full conversation stays in your inbox alongside it.

The Leads page

Leads appear on the Leads page, where each one has a status: new, contacted, converted or lost. From there you can:

• move a lead through those statuses as you work on it
• convert a lead into a customer
• export leads as CSV
• delete a lead

The Customers list shows your customers, with totals worked out from their bookings and orders. It can be exported as CSV too.

The CRM also holds notes and follow-up tasks. A follow-up task is an internal reminder for your team, a to-do item. It doesn't send anything to the customer.

Getting leads in front of the right person quickly

A lead that sits unread until Monday is barely better than no lead. Speed matters: in a 2011 Harvard Business Review article on the short life of online sales leads, James Oldroyd, Kristina McElheran and David Elkington wrote that "most companies are not responding nearly fast enough". The AI answers straight away; the job is making sure a person follows up soon after.

Klugon can tell your team about new leads in a few ways, all automatic once connected:

• Slack. New leads, bookings and handovers are posted to a channel. See the Slack integration.
• Zapier. The same events go to a Zap, so you can send them on to almost anything Zapier connects to. See the Zapier integration.
• Notion. A row is added to a Notion database for each lead.
• Mailchimp. The lead is added to an audience, respecting whether they agreed to marketing.

Klugon doesn't send email or push alerts about new leads. If nobody on your team uses Slack, Zapier is the way to route alerts somewhere they'll be seen.

The Insights page also keeps a to-do list that includes leads not contacted after two days. It's a list you check, not an alert, so make it part of someone's morning routine.

Marketing consent

The AI only records marketing consent when a customer gives it explicitly. That matters if you plan to email leads later. In the UK, the ICO's guidance on direct marketing and electronic communications sets out when you need consent to send marketing. A lead who asked for a quote hasn't necessarily agreed to your newsletter.

AI CRM for small business: HubSpot, Zoho, Salesforce and Pipedrive

If you already run your sales through an external CRM, Klugon can send leads to it. Be clear on how this works, because it's different from what many integration pages imply.

How the CRM tools work

Klugon connects to four CRMs: HubSpot, Zoho CRM, Salesforce and Pipedrive. Once connected, the AI gets tools it can use during a conversation:

• send a captured lead to your CRM
• search the CRM for an existing contact by email or phone
• add a note to a contact or deal

The important word is tools. The AI uses them when it decides to, guided by your instructions. Saving a lead in Klugon does not, by itself, send that lead to HubSpot or any other CRM. There is no automatic background sync.

In practice, that means:

• Put it in the instructions. Something like "After saving a lead, also send it to the CRM" makes the AI much more likely to do it.
• Check it's happening. For the first few weeks, compare new leads in Klugon with new contacts in your CRM.
• Treat Klugon's Leads page as the complete list. Every captured lead is there whether or not it reached the external CRM.

If you need every lead in your CRM without exception, a more dependable route is Zapier: the new-lead event fires automatically for every lead, and you can build a Zap that creates the contact in your CRM from it.

Setting each one up

Each CRM is connected by pasting credentials into Klugon's integrations page. There's no one-click sign-in for these.

• HubSpot. See the HubSpot integration page for what to paste and where to find it.
• Zoho CRM. You'll need to generate a refresh token in Zoho yourself. The Zoho CRM integration page walks through it.
• Salesforce. Also needs a refresh token you get from Salesforce. See the Salesforce integration page.
• Pipedrive. See the Pipedrive integration page.

Which plans include CRM connections

CRM connections come with Klugon's voice or social products, not the chat-only product.

If you pay for more than one Klugon product, you get a feature if any of them includes it.

A sample sales conversation

The setting: Fernhill Garden Rooms, a small firm building insulated garden offices within 30 miles of its yard. The owner has uploaded the brochure and price guide, set up a free site survey as a one-hour service, written qualifying instructions like the example above, and switched on "note buying intent". This happens on the website chat at 9:40 on a Sunday evening.

Customer: Hi, how much is a garden office?

AI: Hi! Our garden offices start at £14,500 for a 3 by 2.5 metre room, fully insulated with power, lighting and a base included. Larger sizes and extras like a toilet or decking cost more. What size are you thinking of, and what will you use it for?

Customer: Something like 4 by 3. Working from home, maybe a bit of music practice in the evenings.

AI: That's a popular size. The price guide puts a 4 by 3 room at £19,800 including standard insulation. For music practice, you may want the acoustic upgrade, which adds extra insulation and sealed doors, from £1,450. Can I ask your postcode, so I can check you're in our area?

Customer: BS8 something.

AI: That's well within our area. When are you hoping to have it built?

Customer: Ideally before Christmas, but not fixed.

AI: Understood. The next step is usually a free site survey, where we measure up, look at access and the ground, and give you a fixed quote. It takes about an hour. I can see Wednesday at 10:00 or Thursday at 14:00. Would either suit you?

Customer: Thursday at 2 is good.

AI: Great. Can I take your name and the best phone number for the surveyor?

Customer: Sam Reyes, 07700 900456.

AI: Thanks, Sam. That's a free site survey on Thursday at 14:00 for Sam Reyes, 07700 900456. Shall I book it?

Customer: Yes. Do you need anything from me before then?

AI: It helps if you can clear a path to where the room will go, and have a rough idea of where you'd like the door and windows. I've booked it, and your reference is FG-2207. Would you like to hear about occasional offers by email? It's fine to say no.

Customer: No thanks.

AI: No problem. The surveyor will see you Thursday at 2. Anything else I can help with?

Customer: That's it, thanks.

What the business sees on Monday

• A new booking for Thursday at 14:00, created from Klugon's real availability.
• A new lead for Sam Reyes with the notes: 4 by 3 garden office, home working plus music practice, interested in the acoustic upgrade, BS8 postcode, wants it before Christmas, high intent because they booked a survey and gave a timeframe. Marketing consent: no.
• A Slack post about the new lead and the new booking, if Slack is connected.
• The whole conversation in the inbox, on the website chat channel.

If HubSpot were connected and the instructions told the AI to send leads there, the lead would also be in HubSpot. Check that it is.

Handing an enquiry to your team

Some sales conversations need a person. A customer wants a price for something unusual. A trade buyer wants to talk terms. Someone is unhappy with a quote.

On chat and messaging, the AI can flag the conversation for your team. That marks it as escalated, shows a "Needs a person" banner in the inbox, logs the escalation and posts to Slack or Zapier if they're connected.

Two things to know:

• The AI keeps replying. Flagging a conversation doesn't pause the AI. If you want to take over, reply quickly and tell the customer you're picking it up.
• Where you can reply from Klugon. You can answer customers directly from the inbox on WhatsApp, Instagram, Messenger and Telegram, within Meta's 24-hour messaging window on the Meta channels. Website chat and phone conversations can't be replied to from the inbox, so for those you contact the customer using the details they left.

On phone calls, the AI can't transfer the caller to a person. It takes their details as a lead so someone can call back.

What the AI sales assistant doesn't do

Here's the honest list. Some of these are things other tools claim; Klugon doesn't do them today.

It doesn't follow up

There are no automatic follow-up messages, email sequences or abandoned-cart reminders. If a lead goes quiet, the AI won't chase them. Following up is your team's job, helped by the Leads page and the "leads not contacted after two days" list.

It doesn't call or text anyone

No outbound calls, no SMS. The AI only replies to people who've contacted you.

It doesn't sync to your CRM automatically

As covered above, HubSpot, Zoho, Salesforce and Pipedrive receive a lead only when the AI calls the tool to send it. Klugon's own Leads page is the complete record.

It doesn't take payment

Don't plan your sales process around taking payment inside the conversation. Use the AI to answer questions, qualify the enquiry and book the next step, and take payment the way you do now. On phone calls there's no payment option at all. Orders in Klugon are confirmed by payment, not by the AI saying so.

It doesn't negotiate or invent discounts

The AI quotes your prices and, if you've set one, your current promo code. It shouldn't haggle or make up offers, and you should say so in its instructions.

It doesn't connect to your online shop

There's no product or order sync with Shopify, WordPress, WooCommerce or any other shop platform. You add the chat widget to those sites by pasting a script tag, and you keep the product catalogue in Klugon up to date by hand.

It doesn't assign leads or track attribution

You can't assign a lead to a team member inside Klugon, and the analytics don't attribute sales to specific conversations. You'll see lead counts, how many converted, revenue from paid orders and a breakdown by lead status.

It doesn't email you

There are no email or push notifications to the owner. Slack and Zapier carry alerts; everything else is in the dashboard.

It isn't an adviser

For regulated work (legal, financial, medical), the AI collects the basics and hands over. It shouldn't advise, and your instructions should say so.

Setting up the AI sales assistant: a checklist

Work through these in order. Each step builds on the one before, and testing comes before anything goes in front of customers.

1. List your buying questions. Go through your last month of emails, messages and calls and write down the questions people ask before they buy. Prices, what's included, lead times, areas covered, how your options differ.
2. Add your products and services. Enter products with clear names, descriptions and prices. Set up services with their lengths and prices, including any free consultation, survey or trial you use as the first step.
3. Fill the knowledge base. Upload your brochure, price guide and terms. Paste in written answers to the buying questions from step 1. Run the website scan to pick up your main pages.
4. Write the qualifying instructions. Three to five questions, the one that rules people out first, what to say when someone doesn't fit, and exactly what to put in the lead's notes.
5. Set lead capture. Choose the details to collect, decide how many messages to wait before asking for them, and switch on the buying-intent note if you want it.
6. Decide on recommendations. Switch them on with one suggestion at most to start with, or leave them off.
7. Connect your alerts. Slack, Zapier or both, so someone sees new leads the same day.
8. Connect your CRM, if you use one. Add the instruction to send leads to it, and plan a weekly check that they're arriving.
9. Test it yourself. Play three customers: a good fit, a poor fit and someone who only wants a price. Check what ended up in the Leads page and the notes.
10. Go live on one channel first. Website chat is usually the easiest to start with. Add phone and messaging once you're happy with the answers.
11. Review weekly for the first month. Read conversations, fill the gaps on the Insights page, and adjust the questions.

What it costs

There's no separate fee for sales features. You pay for the channels you want the assistant on, in US dollars, and every plan includes unlimited assistants.

Paying yearly takes 20% off. A conversation means one customer thread. Phone numbers are $9 a month each and WhatsApp numbers $19 a month. Remember the CRM tiers above: HubSpot, Zoho and Pipedrive need voice or social Business; Salesforce needs Scale.

There's a 7-day free trial on each product. A card is required, and it's charged when the trial ends unless you cancel first. Full details are on the pricing page.

A worked example

Say you run a small garden building firm. Assume:

• 40 website enquiries a month, of which 15 arrive in the evening or at weekends when nobody replies until the next working day
• 1 in 10 enquiries ends in a sale
• an average sale is worth $18,000, with $3,000 profit after materials and labour
• answering the out-of-hours enquiries straight away and booking a survey turns just one extra enquiry into a sale every two months

That's six extra sales a year, or $18,000 in profit. The chatbot Business plan at $149 a month costs $1,788 a year. Even if your numbers are a fraction of these, the sum is worth doing with your own figures. If none of your enquiries arrive out of hours and you already reply within minutes, the case is weaker, and you should say so to yourself honestly.

How to measure whether it's working

Give it a month, then look at these in your Klugon dashboard.

• Leads, and how many converted. On the Analytics page, compared with the previous period.
• Lead status breakdown. How many are still "new"? A growing pile of uncontacted leads means the follow-up, not the AI, is the bottleneck.
• Bookings created by the assistant, if consultations or surveys are part of your sales process.
• Channel split. Where enquiries are coming from: website, phone, WhatsApp, Instagram.
• Busiest hours. How many enquiries arrive outside working hours.
• Questions without a good answer. Fill these gaps every week.
• Leads not contacted after two days, on the Insights page. This should stay near zero.

And read ten conversations. Did the AI ask your questions in the right order? Did it take details at a natural point? Would you have been happy if a new member of staff had handled it the same way?

Frequently asked questions

Is the AI sales assistant a separate product?

No. It's the same Klugon assistant that answers chats, calls and messages, with lead capture, product search and recommendations switched on. You pay for the channels you use.

Does it sync leads to HubSpot automatically?

No. HubSpot, Zoho, Salesforce and Pipedrive receive a lead when the AI uses its tool to send it, guided by your instructions. Every lead is always saved in Klugon's own Leads page. For an automatic route, use a Zap triggered by Klugon's new-lead event.

Can it qualify leads by itself?

It asks the questions you give it, in the order you set, and can add a note on how ready to buy the person seems. It doesn't know your qualifying criteria until you write them down in its instructions.

Will it follow up with leads who go quiet?

No. Klugon doesn't send follow-up messages, sequences or reminders. Your team follows up, using the Leads page and the list of leads not contacted after two days.

Can it answer questions about my products?

Yes, from the products you add to Klugon and from your knowledge base. Products are entered by hand; there's no import or sync with an online shop.

Can customers pay through the assistant?

We don't recommend relying on that today. Use the assistant to answer questions, qualify and book the next step, and take payment the way you already do. There's no payment option on phone calls.

Which CRMs does it work with?

HubSpot, Zoho CRM and Pipedrive on the voice or social Business plan and above, and Salesforce on Scale. Each is connected by pasting credentials; Zoho and Salesforce also need a refresh token you generate yourself.

Does it work on the phone as well as chat?

Yes. On calls, the AI can search products, answer from your knowledge base, book appointments and capture leads. It can't transfer a call to a person, so it takes details for a callback.

An AI sales assistant won't close the deal for you. What it does is make sure every enquiry gets a quick, accurate answer and the right questions, and ends up as a lead with notes your team can act on. The selling, and the follow-up, are still yours.