AI support agent or AI sales agent: which do you need?

Most owners end up comparing an ai support agent with an AI sales agent before they know which questions their site actually receives. One test settles it: would the answer change whether this visitor buys? If it would, the question is pre-sale, whatever channel it arrives on. Price, fit, availability, insurance cover, specs and lead times are pre-sale. Order status, password resets and invoice copies are post-sale.

Start with your last 20 enquiries, not a product category

Search either term and you get "types of AI agents" explainers and "top 8" rankings, written by people who have never seen your inbox. Your inbox is the only evidence that matters here, and 20 minutes gets you enough of it.

1. Pull the last 20 enquiries

   Take them from wherever they actually live: contact-form emails, WhatsApp, notes from missed calls, DMs, chat transcripts.

2. Write the question in one line

   Strip the channel. "Do you take my insurer?" not "web form, mobile, 9.12pm".

3. Mark each one P, S or N

   P for pre-sale, S for post-sale, N for neither.

4. Count, then note the pattern

   How many in each pile, what time of day they arrived, and which page the person was on.

P is the pile where the answer changes a buying decision:

• Clinic: "Do you take my insurer?" or "How soon can I be seen?"
• Agency: "Can you do a Shopify build by 14 March, and what does that cost?"
• Estate team: "Is the two-bed on Elm Street still available, and what are the fees?"
• Coach: "What's included in the six-session pack?"
• SaaS: "Does Pro include API access?"

S is work you owe someone who has already paid: "Where is my order", "How do I reset my password", "Can I get a copy of last month's invoice", "I need to move Tuesday's appointment".

N is noise: supplier pitches, job applications, someone selling SEO.

The detail that decides everything: your inbox files these by channel, not by intent. That is how a pricing question ends up logged as a support ticket. Plenty of what arrives wearing support clothing is somebody deciding whether to buy at all.

Take a five-person clinic whose after-hours enquiries are almost all P. People are asking whether the clinic treats their problem and whether their insurer covers it, because they will not book until they know. The reply they get is "our team will respond within one working day". That is not a resolution. It is a delayed sale, or no sale.

What the piles tell you: most support questions are pre-sale

Read your own count. If most of the 20 are P, a support-first setup is scoped to a job your visitors are not doing yet. If most are S, read the rest of this in reverse, because the support case is genuinely yours.

Why P questions look like S questions: on most small sites the only door is a footer "contact us" link or one generic form, so buyers use it for everything. The visitor is not confused about what they are asking. Your site only has one inbox, and everything lands in it.

The tell is what has to happen after the answer.

One honest caveat: some enquiries are both. "I'm on the starter plan, can I upgrade and does the new price include the add-on?" needs an account lookup and a buying decision. The normal shape for a mixed question is one shared first answer, then a handover to whichever side owns the rest.

Why an ai support agent politely answers the wrong things

A support setup is built around a help centre as its source of truth, plus order and account lookups, ticketing, assignment, escalation and resolution reporting. Its vocabulary is deflection, resolution rate, CSAT and triage. None of that is wrong. It is aimed at a different visitor.

Put that same setup on a pricing page and here is what it sounds like:

• "I don't have access to our pricing. Please contact the sales team."
• "Here is an article about delivery times."
• "Thanks, your request has been received. We reply within one working day."

All three are technically correct. All three are commercially useless to someone deciding today.

The category has moved past scripted rules: current agents interpret intent and can carry out multi-step actions, as this guide to how AI support agents work sets out. That shift is aimed at repeat support volume. It says nothing about a first-time buyer asking what a service costs, and that buyer is most of your traffic.

The sales side deserves the same fairness. An AI sales agent is built to qualify, score fit and hand over. The definition worth borrowing comes from Creatio's explanation of what an AI sales agent does: agents that interact with prospects in real time, qualify leads, recommend products and pass them to a rep. That is closer to the job your enquiries are asking for.

So a support agent is not broken, it is scoped. If your post-sale volume is heavy, it is the right purchase, and a later section says when.

What each setup needs before it is worth paying for

Both setups need the same hard work first: written content that answers your top question. The cheap failure mode is buying the ticketing stack and only then discovering that the sentence answering "do you take my insurer" exists nowhere, in any system. That is the same job in both cases, and it is the one people skip.

Clinic and agency make the split concrete. An agency's pre-sale question is "do you work with Shopify and can you start in March", which needs a pricing page and a portfolio page, not a password-reset flow. A clinic's after-hours question is "do you treat this, and does my insurer cover it", which needs the treatment page and a booking request. If you have already decided the support case is yours, what post-sale support work actually involves sets out the rest of it.

Where one assistant covers both on a small site

For most lead-driven small businesses, the P pile is the one with money in it, and the tools sold as support agents are aimed past it.

BusinessLyft is the AI sales assistant for your website. It answers from your own website and documents, shows the sources it used, and says when it doesn't know. That single knowledge source is what lets the same assistant handle "what does the six-session pack include" and "how do I change my appointment" without buying two products.

The "I don't have our pricing" failure has a specific fix. PageSense reads the page the visitor is on, including prices, plans, specs and collapsed FAQs, so someone sitting on the pricing page is answered with the plan in front of them. It never reads what they type into forms, passwords or hidden fields.

The routing rule that keeps one assistant honest: pre-sale answers are captured, scored and handed over; post-sale answers are resolved from your content where the content exists, then routed to a human. In practice the route is business hours and holidays, an API, or a webhook to Zapier, Make or n8n. It scores every lead Hot, Warm or Cold with an intent tag and a one-line summary, so whoever follows up is not opening a blank email.

What it is not: it qualifies and hands over. It does not close deals, and it is not a ticketing system with assignment and SLAs. If those words are in your requirements, you are the two-tool reader and the next section is yours. A basic assistant can be live in about 30 seconds. The free plan covers one assistant, and Pro is $10 a month with a 7-day trial and no card to start. See the plans.

When two separate tools is the better call

Buy twice if these are true: hundreds of repeat order-status or account questions a month; an existing helpdesk where support already works; SLAs and assignment you have to report on; account actions that need identity verification; support across email, voice and messaging as well as the website; support content owned by a different team.

The support case cuts the other way here, honestly. Repeat service volume is exactly what agentic support is being built for, and that curve is heading up.

The trade-off is not mainly cost. It is two sources of truth falling out of step the week someone renames a plan on the site and forgets the help centre.

The middle path suits most readers: one assistant on the marketing site for pre-sale work, a helpdesk for account work, a written rule about which question goes where, and the assistant telling people where account help lives. The one-line test: if your sorting exercise produced mostly S, buy support first and keep a light pre-sale assistant on the pricing page. Mostly P, do the reverse.

Four mistakes that waste the money either way

1. Buying support to fix a sales problem. The unanswered question was about price or fit, not about an order. If your P pile is mostly out of hours, start with capturing after-hours enquiries without a receptionist.
2. Leaving prices, lead times and availability out of the knowledge source. Every answer then politely defers. If you won't publish a price, decide what the assistant should say instead, and say it the same way every time.
3. Running two versions of the truth. A plan name changed on the site but not in the help centre is worse than having no assistant at all.
4. Letting it improvise a number, or counting a captured lead as a resolved ticket. Quotes must come from your rate card or your own formulas, never generated. A lead marked "resolved" looks like success in a dashboard and is a lost sale in reality.

One more worth saying: don't interrupt people constantly. The polite version is 1 to 3 messages a visit, a cooldown, and silence after a dismiss. Anything more trains visitors to close the bubble on sight.

Frequently asked questions

If your sort came back mostly P, put one assistant on the site, point it at your pricing and service pages, and read what your after-hours enquiries were really asking. You can start on the free plan or see the plans first.