AI Sales Agents: Salesforce's $88M Unanswered-Lead Lesson

An AI sales agent earns its keep on the leads your team never gets to, not the ones it already handles. That is the plain lesson in what Salesforce said on 18 October at the Israel Information Technology Conference in Tel Aviv. Itai Margalit, a Salesforce sales vice president, said the company had about 100 million leads and that roughly 75% of them went unanswered before it put AI agents on first contact. The agents now reach every lead and hand off to a salesperson when a human is needed. He said the initiative has generated more than $88 million in revenue, and that matching the first-response volume with people would have taken around 400 extra employees.

Treat the figures with care. As CTech reported, Salesforce has not said over what period the $88 million was earned or how it was attributed, and the numbers came from one executive on stage. But you do not need the exact number to see the pattern, and it is one most growing businesses share: more people ask than anyone can answer, and the unanswered ones quietly buy elsewhere. This post is part of our guide to rolling out agents at wizeb.com/blog/how-to-implement-ai-agents-in-business.

The Problem: Your Most Expensive Leads Are the Ones Nobody Answers

Every company counts the leads it works. Very few count the ones it doesn't. A form fill lands at 9pm. A pricing question arrives by email while the only salesperson is on a call. A webinar sign-up list of 400 names gets ten follow-ups and 390 silences. A lead from a trade show sits in a spreadsheet for two weeks. None of these show up as a loss in any report, because a lead that was never contacted never becomes a lost deal. It just disappears.

The cost is real anyway. Research published in Harvard Business Review found that firms contacting leads within an hour were seven times more likely to qualify them than firms that waited even an hour longer. Most small and mid-sized companies do not respond within an hour to most leads, and they never respond to a large share at all. The usual fix is to hire, which runs straight into the problem Salesforce described: first contact is high-volume, repetitive work, and staffing it with people is slow and expensive.

What an AI Sales Agent Actually Does

An AI sales agent is not a chatbot waiting on your website, and it is not an email sequence that sends the same three messages to everyone. It is an agent with a defined job, access to your CRM and calendar, and rules for when to step aside. A typical first-contact agent does five things:

1. 1Responds to every new lead within minutes, by email, WhatsApp, SMS or a phone call, in a message written for that lead's actual question.
2. 2Answers the practical questions from information you approve: what you offer, rough pricing ranges, timelines, locations, who it is for and who it is not for.
3. 3Asks the three or four qualifying questions your sales team would ask, such as budget, timeline, team size or decision-maker, and records the answers in the CRM.
4. 4Books a meeting straight into a salesperson's calendar when the lead is a fit, with a short summary so nobody starts the call blind.
5. 5Follows up on leads that go quiet, then closes them out politely, so every record ends as booked, not a fit, or not now.

The handoff is the part that matters most. Salesforce's own description was agents making first contact and passing to a human when a personal touch is needed. That is the right design. The agent does the volume; your people do the conversations that win deals. For more ways agents split work with staff across sales, support, operations and finance, see wizeb.com/blog/ai-agent-examples-by-business-function.

Where AI Sales Agents Pay Off, and Where They Don't

Agents work best where leads are plentiful, questions repeat, and a booked call is a clear next step. They are a poor fit where every sale is a long, custom negotiation from the first message.

• Strong fit: inbound form fills and enquiries, out-of-hours leads, event and webinar lists, old leads that were never worked, and quote requests with standard questions.
• Good fit with tighter rules: renewals and upsell outreach to existing customers, where tone and account history matter more.
• Poor fit: enterprise deals where the first conversation is already a negotiation, and any market where unsolicited automated outreach breaks local rules.

If you are deciding between this and other candidates for your first agent, our scoring method at wizeb.com/blog/first-ai-agent-for-business weighs volume, risk and data readiness. Lead response usually scores well on all three.

How to Set Up an AI Sales Agent in Four Weeks

1. 1Week 1, count and scope. Measure the unanswered-lead baseline, list the ten questions leads ask most, write the qualifying questions and decide which leads must always go straight to a person.
2. 2Week 2, connect and write. Connect the agent to your CRM, calendar and one channel. Load approved answers on offers, pricing ranges and policies. Decide what it may never say, such as discounts or commitments.
3. 3Week 3, test on history. Run the agent against 100 to 200 past leads and have a salesperson score every reply. Fix wrong answers at the source, not just in the prompt.
4. 4Week 4, go live in stages. Start with leads nobody would have reached anyway, such as out-of-hours enquiries and the backlog, then widen once the booked-meeting rate and reply quality hold up.

Guardrails That Protect the Brand

• The agent says it is an AI assistant. Pretending to be a person is the fastest way to lose trust when a prospect finds out.
• No promises on price, discounts, delivery dates or contract terms. Those go to a person with a summary.
• Respect consent and opt-out rules for every channel and country, and stop contact the moment someone asks.
• Cap follow-ups. Three touches over ten days is plenty for most inbound leads; more reads as spam.
• Log every message in the CRM, so sales can see exactly what was said before they pick up the phone.

What It Costs Compared With Hiring

The Salesforce comparison was 400 employees. Your version is smaller but the shape is the same. A sales development rep costs a full salary plus tools and management, works one conversation at a time, and is offline evenings and weekends. An AI sales agent for a single channel and CRM is typically a one-off build plus a monthly running cost driven by message and call volume, usually a fraction of one hire. The honest comparison is not agent versus rep. It is agent versus the leads nobody is answering today, because those are the ones it recovers.

Case Study: A B2B Services Firm's Lost Leads

Consider a 40-person B2B IT services firm with two salespeople. A 90-day count found 1,150 inbound leads, of which 610 got no reply within 24 hours, mostly evening form fills and a webinar list nobody worked. The firm's normal close rate on contacted leads was 4%, at an average first-year deal value of USD 9,000. Even at half that close rate, the uncontacted leads represented around 12 deals, or roughly USD 110,000, in a quarter.

The firm put an AI sales agent on email and WhatsApp for out-of-hours leads and the webinar backlog only. It answered within five minutes, asked about team size, current provider and timeline, and booked fits into the salespeople's calendars. In the first eight weeks it contacted every new out-of-hours lead and the full backlog, booked 46 meetings, and passed 19 pricing questions to a person. The salespeople stopped prospecting their own inbox and spent the time on calls that were already qualified. This is an illustrative example, but every number in it is one you can measure in your own CRM in an afternoon.

How WizeB Builds AI Sales Agents

WizeB builds AI sales agents around the leads you are already losing. We start with the unanswered-lead count, connect the agent to your CRM, calendar and channels, test it against your own past leads, and put it live in stages with clear handoff rules for your team. You own the build, and every conversation is logged where your salespeople can see it. Many clients pair it with our voice agents so phone leads get the same instant response. See how we work at wizeb.com/services/ai-agents.