AI agents for sales: where they help, where they hurt, and how to start

AI agents for sales are most useful around the conversation, not in it. They reply to new leads within minutes, ask the qualifying questions, book the meeting, research the account before the call, write up the notes afterwards and keep the CRM current. The parts of selling that depend on trust, such as discovery, negotiation and pricing, stay with your reps.

This guide is for founders and sales leaders. It covers the jobs an agent does well, the ones it should not touch, the rules on outreach, how to choose between a CRM add-on and a custom agent, and what to measure.

#The jobs a sales agent does well

• Replying to inbound leads at once. A form fill at 9pm gets a useful reply at 9:01pm, not the next morning. Well-known research on inbound leads, including a Harvard Business Review study, found that the chance of reaching and qualifying a lead falls sharply within the first hour. This is usually the best first job.
• Qualifying. The agent asks your questions about need, timing, budget and fit, in a natural exchange, and scores the lead against your rules.
• Booking the meeting. It offers times from the right rep's calendar, sends the invite and handles rescheduling.
• Account research. Before a call the rep gets a one-page brief: what the company does, recent news, who is on the call, and any earlier contact recorded in your CRM.
• Call notes into the CRM. After a recorded call, the agent writes the summary, fills in the fields you care about, and creates the next task.
• Follow-up drafts. A recap email with the agreed next steps, ready for the rep to edit and send.
• First drafts of quotes and proposals, built from your templates, your price list and the call notes.
• Pipeline upkeep. Flags deals with no activity, missing next steps or close dates in the past, and asks the owner to update them.

#What should stay with your reps

• Discovery and demos. Understanding what a buyer actually needs is the job.
• Negotiation and pricing exceptions. An agent should never offer a discount you have not approved.
• Relationships. Renewals and expansions are won by people who know the account.
• The final say on outbound messages, at least until you trust what the agent writes.

#Inbound first, outbound with care

Inbound work is low-risk. The person has asked to hear from you, a fast, helpful reply is what they want, and mistakes are easy to catch.

Outbound is where AI has a poor reputation, mostly earned. Mass-produced "personalized" messages are easy to spot, they damage your sender reputation, and in many countries they are restricted by law. If you use an agent for outbound:

• Use it for research and first drafts, with a person choosing the accounts and approving the message.
• Keep volumes modest. Deliverability suffers quickly when sending spikes.
• Make sure any personal detail in a message is true and relevant. An invented compliment is worse than none.

#The rules still apply

An AI writing the message does not change the law.

• Email. In the US, CAN-SPAM requires honest headers, a working opt-out and a postal address. The EU and UK generally require consent or a narrow legitimate-interest case for marketing email, and have strict rules on personal data.
• Calls and texts. In the US, the TCPA requires consent for automated calls and texts, and regulators have made clear that AI-generated voices are covered.
• Platforms. LinkedIn and similar networks prohibit automated messaging in their terms of use.

This is general information, not legal advice. Check the rules for every country you sell into.

#Guardrails for a sales agent

• Facts come from your systems. Prices, plans, terms and availability are read from your price list and CRM, never improvised.
• No promises. Discounts, delivery dates and custom terms go to a person.
• Limited CRM permissions. The agent can write to specific fields and create tasks. It cannot delete records or change owners.
• Draft mode first. For anything sent to a customer, a person approves until the results are consistently right.
• Honest about what it is. In live chat, say the customer is talking to an AI, and offer a person.
• Everything logged, so you can see what was said to whom.

#CRM add-on, sales tool or custom agent?

Two notes. First, some of what is sold as a "sales agent" is plain automation: a form fill that creates a CRM record, alerts the rep in Slack and sends a templated email needs no agent, and costs far less. Second, a custom agent makes most sense when the knowledge that qualifies a lead is particular to you, for example usage data from your own product.

#What to measure

• Time to first response for inbound leads. This should fall from hours to minutes.
• Lead-to-meeting rate, and whether the meetings are with good-fit accounts.
• Show rate. Meetings that are booked and then held.
• CRM completeness. The share of open deals with a next step, an amount and a close date.
• Hours returned to reps each week. Ask them.
• Reply quality on outbound, not reply volume. Count conversations started with accounts you wanted.

Be careful about crediting the agent with revenue. Measure the steps it controls.

#What it costs

• CRM add-ons are usually priced per seat, on top of your existing plan.
• Specialist sales tools range from a few hundred to several thousand dollars a month depending on seats and volume.
• A custom sales agent typically costs $8,000 to $40,000 to build, with running costs that are usually small next to a rep's time. Details are in how much does an AI agent cost.
• Plain automation, such as lead routing, alerts and templated follow-up, often costs a few thousand dollars and is sometimes all you need.

Our own pricing, for reference: a single automation starts from $4,800 and takes about two weeks. A custom AI agent starts from $9,800 and takes three to five weeks. Both are quoted as one fixed price before work begins, and include a 30-day fix window.

#Where to start

Look at how long new inbound leads wait for a first reply, and what share of them ever get a meeting. If the wait is measured in hours, instant response and booking is your first job, and it may need only your CRM's own tools. If your qualifying rules or data are more particular, book a free 20-minute call and we will tell you what a build would involve.

This guide is part of our series on AI agents for business.

Common questions.

• Can AI agents do sales?

  They can do much of the work around selling: respond to new leads right away, ask qualifying questions, book meetings, research accounts, write call summaries, update the CRM and draft follow-ups. They are poor at the parts that win complex deals, such as discovery, handling objections and negotiating. Treat an agent as a very fast assistant to your reps, not as a rep.

• Will an AI sales agent replace SDRs?

  It replaces tasks more than roles. Instant replies to inbound leads, research, data entry and routine follow-up can move to an agent. Judgment about which accounts matter, real conversations and relationships stay with people. Many teams find the same number of reps can cover more leads, and that rep time moves from admin to calls.

• Is AI-generated cold outreach legal?

  The same rules apply whether a person or an AI writes the message. In the US, commercial email falls under CAN-SPAM, and automated calls and texts fall under the TCPA, which requires consent. The EU and UK have stricter consent rules for email and much tighter rules on personal data. Check the rules for each country you sell into, and take advice if unsure.

• How do I create an AI sales agent?

  Start with one job, usually replying to inbound leads. Write down your qualifying questions and what counts as a good fit. Connect the agent to your website form, your calendar and your CRM with limited permissions. Run it in draft mode, where a person approves each message, until the replies are consistently right. Then let it send on its own for that one job.

• How much does an AI sales agent cost?

  Agents built into CRMs are usually a per-seat add-on. Standalone AI sales tools range from a few hundred to several thousand dollars a month depending on volume. A custom sales agent typically costs $8,000 to $40,000 to build. A simple lead-routing automation with no agent involved can cost much less, often a few thousand dollars.