You paid for the click. The visitor landed on your site at 10:40 p.m., asked a question about price, and nobody answered until the next morning. By then they had booked with someone else. That's the problem most businesses actually have, and it's worth keeping in mind when you search for an AI sales agent, because almost every page that ranks for the term is selling you something else: a bot that writes cold emails to strangers. Salesforce's 2026 survey found 54% of sellers have already used AI agents and nearly 9 in 10 plan to by 2027 (Salesforce State of Sales, 2026). The word is everywhere. What it means depends on which way the conversation is flowing.
This guide separates the three products that all call themselves "AI sales agents", shows what each one does and where each one has gone wrong, and gives you a simple test for which one you need.
What is an AI sales agent?
An AI sales agent is software that carries out part of a sales conversation on its own: it reads or listens to a prospect, decides what to say, and takes a next step such as qualifying, booking or handing off, without a person typing each reply. That's the shared definition. Underneath it sit three very different products.
The first is the outbound AI SDR (sales development rep). It finds prospects, researches them, writes cold emails or LinkedIn messages, and tries to book meetings. 11x, Artisan and AiSDR are the best-known names. This is what most "AI sales agent" listicles mean.
The second is the AI calling agent: a voice model that places or takes phone calls, usually priced per minute. Bland AI, Retell AI and Synthflow sit here.
The third is the inbound answer-qualify-book agent. It doesn't go looking for anyone. It waits for the person who already found you, through an ad, a search result or a referral, and then answers their question, asks the qualifying questions you would ask, and books the appointment or captures the lead. It works on the website, in messaging apps and sometimes by text.
The difference matters because the economics are opposite. Outbound agents try to create demand from strangers who didn't ask to hear from you. Inbound agents try to stop losing demand you've already paid for. One is a volume game against spam filters and reply rates. The other is a speed game against your own office hours.
How is an outbound AI SDR different from an inbound AI sales agent?
An outbound AI SDR starts the conversation; an inbound AI sales agent responds to one the customer started. That single difference changes the risks, the rules and what "success" looks like.
Here is the comparison side by side:
| Outbound AI SDR | AI calling agent | Inbound answer-qualify-book agent | |
|---|---|---|---|
| Who starts the conversation | The agent | Either side, by phone | The customer |
| Typical channel | Cold email, LinkedIn | Phone | Website, WhatsApp, Telegram, text |
| Main job | Research, sequence, book meetings | Talk, qualify, set appointments | Answer, qualify, book or capture |
| Usual pricing | Annual contracts or high monthly fees | Per minute | Flat monthly or per seat |
| Biggest external risk | Spam filters, platform bans | Robocall consent rules | Wrong answers, slow handoff |
| Success metric | Meetings booked from cold lists | Calls handled, appointments set | Share of enquiries answered and booked |
Read the risk row closely. An outbound agent is fighting gatekeepers that exist precisely to stop unsolicited messages. An inbound agent has a much easier job in that respect: the person asked. Its risks are quality risks, such as getting the price wrong or failing to hand a serious case to a human, and those are under your control.
The success metric differs too. An outbound tool is judged on how many meetings it pulls out of a cold list. An inbound agent is judged on something simpler: of the people who reached out, how many got a correct answer fast and ended up booked. If you've ever looked at your ad dashboard and wondered why leads don't turn into appointments, that second number is the one you care about. We go deeper into why the "chatbot vs AI agent" naming causes confusion in AI agent vs live chat vs chatbot.
What can an AI sales agent actually do in 2026?
In 2026 an AI sales agent can reliably handle the repetitive middle of a sales conversation: answering known questions, asking a fixed set of qualifying questions, offering times and recording the result. It is much weaker at judgment calls, negotiation and anything emotionally loaded.
On the outbound side, sellers mostly expect agents to save research and writing time. Salesforce's survey of 4,050 sales professionals across 22 countries found sellers expect agents to cut prospect research time by 34% and email drafting time by 36% (Salesforce State of Sales, 2026). The same survey found 87% of sales organisations already use some form of AI. Those are expectations from a vendor survey, not measured results, but they tell you where the work is: preparation, not closing.
On the inbound side, the useful work is narrower and more concrete. A good inbound agent should be able to:
- Answer the questions people ask before they buy: price ranges, availability, location, what's included, what happens next. It should answer from your own website and documents, not from guesswork.
- Qualify by asking the two or three questions your best salesperson would ask: budget, timing, eligibility, service type.
- Book or capture: offer a slot, take contact details, or pass the conversation to a person with the context attached.
- Hand off cleanly when the question is outside what it knows, instead of improvising.
That fourth point is where weak tools fail. An agent that always answers will sometimes answer wrongly, and a wrong price quoted at midnight is worse than no answer. If you want the longer version of what "agentic" actually means in customer conversations, see agentic AI that takes actions.
Why has the AI SDR category run into trouble?
The AI SDR category has run into trouble because the volume model that makes it attractive is exactly what email providers, platforms and buyers are built to resist. Several public warning signs have stacked up since 2025.
The most reported case is 11x. In March 2025 TechCrunch reported that the a16z- and Benchmark-backed startup had shown logos of companies that weren't customers. ZoomInfo told the publication: "We did not give them permission to use our logo in any manner, and we are not a customer." One employee was quoted saying, "We were losing 70-80% of customers that came through the door," while 11x said its "retention rate is currently 79%" and put its highest churn down to early cohorts in late 2023 (TechCrunch, 2025). You don't have to decide who was right to take the lesson: the category's headline numbers were contested in public by its own staff.
The second problem is structural. Since February 2024, Yahoo has required all senders to keep their spam complaint rate below 0.3%, and bulk senders must support one-click unsubscribe and honour unsubscribes within two days (Yahoo Sender Hub, 2024). Google introduced similar bulk-sender rules at the same time. An outbound AI agent's whole advantage is sending more personalised cold email than a human could. Push volume up and complaint rates follow, and a domain that crosses the line can see its mail filtered for everyone, including your existing customers.
The third problem is that buyers don't yet feel the benefit. Gartner predicts that by 2028 AI agents will outnumber sellers ten to one, yet fewer than 40% of sellers will say agents improved their productivity (Gartner, 2026). That's a prediction, based on a survey of 210 sales leaders, not a measurement. But it captures the risk well: lots of agents, lots of activity, not much more revenue.
None of this means outbound AI is useless. For B2B companies with a clean list, a real offer and someone managing deliverability, research and drafting help is valuable. It means "AI sales agent" is not a synonym for "more revenue", and the outbound version carries risks a small team may not be set up to manage.
Are AI sales agents legal to use?
Yes, AI sales agents are legal in the US, but the rules depend on direction and channel. Outbound AI voice calls face the strictest rules; inbound agents answering people who contacted you face the lightest.
For phone, the key document is the FCC's Declaratory Ruling FCC 24-17, released on 8 February 2024. It confirmed that AI-generated voices, including voice clones, count as an "artificial or prerecorded voice" under the Telephone Consumer Protection Act (FCC, 2024). In practice that means outbound AI voice calls to mobile phones need the recipient's prior express consent, and telemarketing calls need prior express written consent. The ruling didn't ban AI calls. It applied the existing robocall rules to them, and it governs calls you place, not calls a customer places to you.
For text conversations, the main questions are disclosure and data. California's bot law (SB 1001) makes it unlawful to use a bot to knowingly deceive someone about its artificial identity to drive a sale. Utah's AI Policy Act requires businesses to disclose AI when a consumer clearly asks. The safe habit for any inbound agent is to say it's an AI in its first message and offer a human. We cover the full state-by-state picture in US chatbot disclosure laws, and the messaging-specific rules in TCPA rules for SMS and WhatsApp.
The simple version: if your AI is the one starting conversations, check consent before anything else. If it only answers people who reached out first, focus on honesty and data handling.
Which kind of AI sales agent does a business that pays for ads need?
If you pay for ads that send people to your website or WhatsApp, the inbound kind is almost always the one you need first. Your problem isn't a shortage of people to talk to. It's that the people you paid to attract don't get a fast, correct answer.
The evidence on phone leads shows how much leaks even in the channel businesses take most seriously. CallRail's platform data shows 28% of business calls go unanswered on average (CallRail, 2026). Invoca's 2026 benchmark, covering more than 70 million calls across 10 industries, found only 56% of callers to businesses speak with a person. Of the calls a person did answer, 38% were leads, and 42% of those leads converted on the call (Invoca, 2026). Both are vendor datasets, but they point the same way: a large share of buyers who reach out never reach anyone.
And many buyers don't call at all. Ruler Analytics' 2026 benchmark of more than 5 million conversions found phone calls make up 56.3% of conversions in legal but only 7.8% in beauty and cosmetics, where web forms carry nearly everything else (Ruler Analytics, 2026). Ruler's data leans UK, so treat it as a pattern rather than a US figure. The pattern is clear enough: in a lot of verticals, the lead arrives in writing, not by phone.
Here's a quick test for which agent you need:
- You have more inbound enquiries than you can answer quickly, especially at night and on weekends. You need an inbound agent.
- Your inbound is steady and answered, and you need more top-of-funnel pipeline in B2B. An outbound AI SDR may help, if someone owns deliverability.
- Most of your leads phone you and many calls go to voicemail. You need phone coverage: a person, an answering service or a voice AI. We compare those in AI receptionist vs AI chat agent and after-hours answering service vs AI.
Most businesses that pay for ads land in the first group. The leak is between the click and the reply, and outbound tools don't touch it.
How should you evaluate an inbound AI sales agent?
Evaluate an inbound AI sales agent by testing it on your own business, with your own customers' questions, before you pay. Demos built on a vendor's sample site tell you nothing about how it will handle your prices, your exclusions and your edge cases.
A practical checklist:
- Ask it your hardest real question. Pick the one your team answers most often and most carefully, such as "How much for X, and is there anything extra?" See if it answers from your content or invents a number.
- Check qualifying. Does it ask the two or three questions that separate a serious lead from a browser, without interrogating them?
- Check booking and handoff. Can it offer a time, take details, and pass the full conversation to a person when it should?
- Check pricing structure. Per-resolution or per-conversation pricing can punish you in a busy month. Flat pricing makes the cost predictable. Our breakdown of Intercom Fin's per-resolution pricing shows how usage-based bills add up.
- Check the channels. It should sit where your leads actually arrive: the website and the messaging app your customers use.
- Check how it asks. Invoca's 2025 benchmark found only 35% of human agents asked phone leads to buy or book (Invoca, 2025). An inbound agent should always offer the next step. That one habit is where much of the value is.
Speed is the other half of the evaluation. Harvard Business Review's 2011 audit of 2,241 US companies found an average 42-hour response to web leads, and 23% never responded at all (HBR, 2011). The data is old, but the point stands: the gap between an instant answer and a next-day reply is where leads are lost. Our speed-to-lead playbook turns that into a working routine, and the full set of numbers is in lead response time statistics for 2026.
Where does Omago fit among AI sales agents?
Omago is an inbound AI sales agent: it answers, qualifies and books on your website, WhatsApp and Telegram, from your own site content. It doesn't send cold email and it doesn't answer phone calls. If phone is where your leads go missing, you need a different tool for that leak.
It's priced at US$49–99 a month flat, with no per-resolution fees, as of October 2026 (see pricing). If you're weighing it against general automation tools, the ManyChat alternative comparison sets out what each is built for. And as with any tool in this category, don't take the description on trust. Test it on your own site.
Paste your website URL into Omago and ask it the question your customers ask at 11 p.m.
Frequently asked questions
What does an AI sales agent do?
An AI sales agent handles part of a sales conversation without a person typing each reply. Outbound versions research prospects and send cold emails or LinkedIn messages. Inbound versions answer questions from people who have already contacted you, ask qualifying questions and book appointments or capture contact details. Which one you need depends on whether your problem is finding leads or answering them.
Is an AI sales agent the same as an AI SDR?
No. An AI SDR is one type of AI sales agent: the outbound type that prospects and books meetings from cold lists. Inbound AI sales agents don't prospect. They respond to enquiries from your website and messaging channels. The two share the label but have opposite economics and very different risks.
Are AI SDRs worth it?
They can be for B2B teams with a strong list, a clear offer and someone managing email deliverability. The category has had public problems, including TechCrunch's 2025 report on 11x's customer claims, and email providers now require spam complaint rates below 0.3% (Yahoo Sender Hub, 2024). For a business whose leads come in through ads, an inbound agent usually pays back faster.
Can an AI sales agent make phone calls?
Some can, but outbound AI voice calls in the US fall under the TCPA. Since the FCC's 2024 ruling, AI-generated voices count as artificial voices, so outbound calls need prior express consent, and written consent for telemarketing (FCC, 2024). Many AI sales agents, including chat-based inbound ones, don't handle phone at all.
How much does an AI sales agent cost?
Pricing varies by type. Outbound AI SDRs are often sold on annual contracts, voice agents are usually charged per minute, and inbound chat agents tend to be flat monthly or per seat. Always check whether you're billed per conversation or per resolution, because usage pricing rises in your busiest months.
Sources: Salesforce State of Sales (2026), Gartner (2026), TechCrunch (2025), Yahoo Sender Hub (2024), FCC Declaratory Ruling FCC 24-17 (2024), CallRail (2026), Invoca Lead Conversion Benchmarks (2026), Invoca Call Conversion Industry Benchmarks (2025), Ruler Analytics (2026), Harvard Business Review (2011). General information, not legal advice.