# Half Your CRM Data Is Wrong: AI Prospecting in a World of 49% Exec Turnover > Source: > Published: 2026-10-25 11:00:12+00:00 Earlier in the day at the latest SaaStr AI Day, Jason and Amelia talked about a core weakness in most AI prospecting tools: they run almost entirely on third-party data. Every team buys from the same data vendors, so every agent ends up writing the same emails to the same people. In the next session, Steve (CMO) and Richa (head of product) from Aurasell showed a prospecting agent built to also use first-party data: your own CRM history, past meeting transcripts, and internal conversations. Amelia had a preview the week before and plans to benchmark it against the agents SaaStr already runs. ## An AI SDR Steve Asked About the Commanders vs. the 49ers Steve opened with the outbound most of us now get daily. Emails with bold text and emojis that are obviously not personalized. AI SDR threads where the agent keeps following up with itself while the recipient never replies. The day before the session, Steve asked one AI SDR who would win the Commanders vs. 49ers game. It answered him. The question had nothing to do with what it was selling, and it engaged anyway, because it was built to keep the prospect talking. Aurasell decided not to ship another agent like that. The team spent a few months researching why AI outbound ends up this bad before releasing its first version. ## The First Meeting Came From a Prospect’s Home Fish Tank The first version of the agent booked a meeting. Then the SDR on the account asked where the fish tank came from. The email’s subject line asked Chris whether the fish tank was still thriving, and the body opened with a line hoping the fish tank crew was doing well. The rest was short and relevant to his business. It didn’t read like AI. No one on the SDR team had mentioned a fish tank. It turned into a long Slack thread across product, engineering, and the SDR team trying to figure out the source. The answer was in the CRM. Chris had met with Aurasell more than a year earlier, and the timing hadn’t worked out. The agent read that meeting’s transcript and found that the intro small talk was about Chris’s large home fish tank. It used that as the opening. No third-party data vendor has that detail. The only place it existed was Aurasell’s own CRM. ## A Personal Hook Needs a Problem Attached Steve was clear that a personal detail alone doesn’t get a meeting. “Hey Steve, I saw you like golf” gets a shrug. The hook has to lead into a problem the prospect actually has, something the email makes them aware of that they might need to solve. Recency matters too. Aurasell researches how old a signal can be before it stops working. A signal from six months ago is usually weaker than one from last week. He listed situations where first-party data does most of the work: - **Startups out of stealth.** Early on, founders take dozens of feedback meetings. A year later the product has changed, and those people are the best list to go back to. - **Enterprises launching a new product.** Years of past conversations in the CRM hold context on accounts that can open the door for the new product. His view is that most outbound needs both first-party and third-party signals, chosen based on the goal of the campaign. ## Each Contact Gets a Different Number of Steps, Replanned After Every Signal Richa showed the product live. The main difference from a standard sequence tool is that there’s no fixed sequence. The agent builds a plan per contact and changes it as new signals come in. The demo went like this: 1. **Prompt for contacts.** For example, VPs and CROs at large transportation and logistics companies in Washington state. The agent searches an 850 million contact database and builds a list of the most qualified people. 2. **A plan per contact.** One person gets nine steps, another gets seven. The plan depends on how that person has been contacted before and whether they’ve been active on LinkedIn recently. 3. **Human review.** Reps see each plan, can coach the agent, and confirm before anything goes out. 4. **Replan on signals.** In the demo, a prospect named Marcus opened the first follow-up 48 seconds after it arrived. The agent logged it as a medium signal of interest and rewrote his next steps. Email clicks, LinkedIn replies, and account-level changes like new leadership all trigger a replan. 5. **Ongoing direction.** Reps can ask for 20 more prospects, clone the campaign for directors of operations at freight forwarders in Texas, or add instructions like “only reach out during reasonable hours in their time zone.” When a contact opens three emails and clicks links but doesn’t reply, the agent changes the next step from another email or LinkedIn touch to a phone call. Richa described it as working toward one goal, a booked meeting, using whatever channel and data it has. ## Job Changes Become a Signal Instead of a Dead Contact High turnover in B2B GTM roles came up from the livestream. The agent tracks each contact’s current profile. If someone has moved, it reaches out to them at the new company with the new email address, and also finds the right person to contact at the old company. When a contact looks stale and LinkedIn hasn’t been updated, the agent searches for news or press about the person and uses that. Most execs have some public coverage when they change roles. ## Deliverability: Multiple Email Vendors, Warmed Secondary Mailboxes On deliverability, Richa said Aurasell sources emails from multiple vendors and verifies them before sending. Users connect a primary Gmail or Outlook mailbox and can add secondary mailboxes. A built-in SendGrid setup warms new mailboxes and spreads sending volume across them. ## Richa’s Recruiting Agent Booked 20 Meetings While She Was on Vacation Aurasell uses the agent for more than sales. The team has used it to recruit PMs and to fill seats at field marketing dinners. A couple weeks before the session, Richa started a recruiting agent to source PMs in India and then left on vacation to see fall colors. While she was away, the agent messaged candidates on LinkedIn, answered their questions in their own time zone, sometimes within two minutes, and booked them onto her calendar. She came back three days later to 20 booked meetings. Time zones have been a real weakness for agents. Amelia said that even a year ago, telling an agent to send something at 6 a.m. could confuse it, because agents default to UTC rather than Pacific time. The agent has to know where the contact is or it will email them at 3 a.m. ## 14 to 16 Meetings a Week: 60% Phone, 40% LinkedIn, Almost No Email Steve runs marketing and manages the SDRs. Jason, Aurasell’s CEO, runs sales. Their SDR team books 14 to 16 meetings a week. About 60% come from cold calls through a power dialer, with humans on the phone. Most of the rest come from LinkedIn. Email produces the occasional meeting. Their buyers are sales leaders who get flooded with email every day and almost never reply to it. Steve is careful with email because of what’s at stake on each send. Marketing overused email for 20 years because it was cheap to send. An unsubscribe removes that contact for good, so he only sends when the message is genuinely useful to the person. ## SDR Candidates Write a Prospect Email Without AI Steve tests SDR candidates on skills agents don’t have. Candidates go to Aurasell’s website without AI and write a prospecting email. The exercise shows how quickly they can learn the product and whether they can write outreach by hand. Steve thinks human creativity is often missing now, and that it’s what lets a team beat competitors using the same tools. Amelia said the CRO of Databricks told her the same thing in an interview for The Agents coming out next week. Databricks has plenty of AI, and he still screens SDRs on cold calling and walking into offices. Agents handle account research, automated email, and follow-up in the right time zone. When step four of the sequence is a phone call, an agent won’t pick up the phone and make it. Many new SDRs grew up with AI, so these skills are getting harder to find. Steve added that a sales leader had replied to one of their reps that morning praising the pitch and the call control. If a rep can’t pitch a sales leader well, they’re unlikely to do well in the role. ### **A few more learnings from the session:** - One AI SDR had a full back-and-forth with itself in Steve’s inbox while he never replied. Agents set to keep pushing until a goal is hit need a hard stop. - A first-party signal from a past conversation booked a meeting that third-party data couldn’t have. For companies with years of CRM history, past meetings are a sourcing channel. - When opens and clicks don’t produce a reply, switching the next step to a call is something reps should do anyway. The agent makes the switch automatically. - Amelia pointed out that agents work 24/7 and won’t stop on their own. Give a prospecting agent a goal and it will use every channel it has to reach it, which is why the guardrails and human review step matter. To try the prospecting agent, request a [demo at Aurasell’s website.](http://www.aurasell.ai)