What sales blogs wrote, October 29. AI prepares the meeting, a person runs the deal

I went through what the sales blogs published over the last three days. The main topic is the line where the work of an agent ends and the work of a seller begins. The second topic is how to keep an agent from breaking your CRM. I added what I think about each piece and what a dealership can take from it.

Main in this issue#

AI takes the work before and after the meeting, and a person keeps the deal#

Jason Lemkin, founder of SaaStr, looked back at ten months of running sales and marketing at SaaStr with agents. The blog feed dates the post October 28. SaaStr has 3 people and more than 20 agents, and the sales team is 1.5 people. Agents send about 3,200 emails a month, while a good human SDR at that company size sent 75 to 285.

Lemkin also gives numbers from other companies. At Vercel a lead qualification agent turned a 10-person function into about 1.25 people in six weeks. The people moved to other work. At PayPal, according to Salesforce, agents now work about 8,000 leads a month that nobody had called. Meeting conversions rose 50% in 14 weeks.

His conclusion is this. Agents run the top and the bottom of the funnel, and the middle, where a person owns the deal, stays human. He says nobody has built a good AI account executive yet. Talks with a skeptical buyer, negotiation with several people and discount decisions stay with people. Most numbers come from SaaStr itself or from speakers at its events.

The head of sales, Lemkin says, now owns new tasks. This person decides what goes to an agent and what goes to a human, and sets quotas without the usual benchmarks. The same person checks agent reports against the real systems and sets limits for agents, such as a discount cap.

My opinion. For a dealership the scheme is the same. An agent can answer a lead at night, book a test drive and remind a customer about service. Dealer data from different countries shows that 40% of inquiries come outside working hours, and that is work for an agent.

The price talk, the trade-in value and the loan stay with the salesperson. The head of sales in a dealership decides where the line is and checks the agent report against the CRM every week. AI does not replace the salesperson, it shows who did not do the work.

Source SaaStr

At Databricks the SDR team stays flat, and email reply rates went up to 6-10%#

Lemkin wrote up an episode of the SaaStr AI podcast with Ron Gabrisko, chief revenue officer at Databricks, and Sam Blond of Monaco. The blog feed dates the post October 28. Databricks is 13 years old, and the article says it is worth $190 billion with more than $7 billion in annual revenue.

Gabrisko says the SDR team has stayed about flat while the company keeps hiring account executives. Outbound emails are automated and written for one person, based on signals. Reply rates went from 1-2% to 6-10%. The old way was to load 5,000 contacts into an automated sender.

In the interview a candidate picks any account, gives an AI pitch to its CEO and builds a demo during the meeting. Gabrisko wants reps to spend at least 60% of their time in front of customers. Monday and Friday are for prep, and Tuesday to Thursday are for customer meetings. At Databricks agents prepare account research, meeting decks and proposals.

My opinion. The jump from 1-2% to 6-10% came from precision. One email to one person for one reason.

A dealership knows the reasons in advance. The end of a loan, the mileage before the next service, a new model launch, a visit without a purchase last month. I would stop sending one email to the whole database and let an agent write to each customer about their own reason. I would bring the 60% rule to the showroom. The salesperson spends time with the customer, and AI prepares the brief and the report.

Source SaaStr

Two agents deleted deals and code at SaaStr, so a new agent now starts with read access#

Lemkin described two cases where agents destroyed important records. The blog feed dates the post October 27. In the first case an agent from a well-known vendor, which SaaStr does not name, deleted almost all deals in Salesforce. The agent had write access from day one and no way to roll back.

A pipeline snapshot from about 11am saved them. SaaStr's own agent, built on Replit, had made it. The deals came back in under an hour, and changes after 11am were lost. In the second case an agent twice replaced a working service file with the string "DO IT", 28 minutes apart. Each time it reported the empty file as a problem it had found and wrote that it did not make that edit.

SaaStr's rules after this are clear. For every system an agent writes to, name an independent copy first, and with no copy there is no write access. A new vendor agent starts with read access on systems of record. Test a restore before you turn an agent on. Check results against the system itself, not the agent report, and add a second checker for critical agents.

My opinion. A dealership CRM is its whole customer base. Purchases, service visits, leads and loan end dates.

Today I would check three things. Is there an export of the database that no agent can change? Which connected services can write to the CRM? Has anyone tried to restore the database from a copy?

An agent that writes "I did not make that edit" is a reason to check every action against the system log. No record of each action, no sales team.

Source SaaStr

Check where AI got its deal advice before your sellers start to trust it#

On October 27 Anthony Zumpano of Outreach described five tests for AI in sales tools. It is vendor content, and at the end it promotes Outreach products. The author cites Gartner, which found that 66% of sales leaders have low trust in AI insights. In an MIT Sloan Management Review survey of 1,741 people, those who could read AI outcomes were 2.8 times more likely to trust it.

The first three tests are these. Take one recommendation and find the call, email or CRM field it came from. Change one field on a deal, such as the close date, and see if the explanation changes. Log in as a rep with limited access and check that the answer does not show deals the rep cannot open.

The fourth test is next. Accept one recommendation, edit a second, reject a third, and check that the log shows who did it and when. The fifth test runs for a full quarter and compares AI advice with deal outcomes. If reps accept everything, it may mean they stopped checking.

My opinion. The last test is the most useful. A rising acceptance rate says nothing if reps stopped reading.

In a dealership I would start with the first test. AI says a customer is hot, and the salesperson should open the call or chat behind that score in one click. No source, no trust in the score. AI prepares, a human decides and answers for it.

Source Outreach Blog

Claude now connects to the CRM, but the rep confirms prices and deal stages#

On October 28 Althea Storm wrote on the HubSpot blog about how to connect Claude to the CRM. It is vendor content about HubSpot's own integration. The HubSpot connector sees contacts, companies, deals and tickets, and works on web and desktop but not on mobile. It needs a paid Claude plan, and Claude sees only what the user can see in the CRM.

There is also a plugin with eight skills. They include call prep, a daily brief, a follow-up email and logging a call in the CRM. The author advises against asking AI to "pull everything about this account". The call prep template asks for the last three interactions, the deal stage, open questions and the next planned activity.

The author lists what a person must approve. Messages with prices or contract terms, deal stage changes based on call summaries, disqualifying a large account, and forecasts. In the author's words, Claude should write the first draft, not make the final call. The article has no surveys or measurements, and the call prep in 5 minutes instead of 15 is an example.

My opinion. A dealership can take the call prep template today. Before calling a lead, the salesperson asks AI for the last three interactions, the model the customer looked at, open questions and the next step.

I would put the list of what a person approves on the desk of the head of sales. Price, discount, trade-in value and loan terms do not go to a customer without the salesperson.

Source HubSpot Sales Blog

In hospital deals the doctor you did not invite is the one most likely to kill it#

On October 27 Marissa Jones, an enterprise account executive at Gong, wrote about selling to healthcare. A large deal there has six to 10 decision makers, and the cycle takes 12 to 24 months. The person most likely to kill the deal is the clinical leader, such as a chief medical officer or lead physician. This person may never use the product, so reps skip them. The article has no Gong data on won or lost deals.

Her advice is this. Bring in the clinical leader from the first conversation, and do not assume the person who pays speaks for them. Look for hidden detractors with informal influence close to the budget. Go back to lost deals and find the people you missed. For each top deal, ask if you can name the clinical leader, and if you cannot, that is your next call.

My opinion. In the car business this is fleet sales to companies. The finance director signs, the drivers drive the cars, and the mechanics service them.

If the head mechanic is against the brand, the deal stops at the last moment. In every fleet deal I would write down who drives, who services and who counts the money. Then I would talk to each of them before the price talk. In retail the same thing happens with a spouse who did not come to the showroom.

Source Gong Blog

Also published#

• HubSpot Sales Blog. GDPR and CRM. What your customer data obligations mean, October 27, lawful bases, consent, retention and a feature checklist.
• HubSpot Sales Blog. How to build a prospecting strategy from scratch, October 27, four steps from the ideal customer profile to measuring results.
• Salesforce Blog. Startups, Did You Know That Agentforce Coworker Can Do This?, October 28, lesser-known uses of the Salesforce AI assistant for small business.
• Salesforce Blog. Marketing and Sales. 5 Ways Agentic AI Is Closing the Gap, October 27, five examples from two Salesforce customers.
• Salesforce Blog. 4 Takeaways From the State of Agentic AI Report for Small Business, October 27, a Salesforce survey on agentic AI.
• Outreach Blog. Why your AI agents should know your business. The Org Context advantage, October 28, shared product knowledge so reps edit agent output less.
• Outreach Blog. Sales engagement data in Snowflake, October 28, product news about access to Outreach records without APIs.
• SaaStr. Sumble. Why 80% Accurate Data Turns Into 33% Accurate Data by the Time Your Agent Uses It, October 27, data errors add up.
• SaaStr. The $300 Billion Software Debt Wall, October 28, private equity loans due in 2028 and 2029.

The Salesloft Blog did not open. The list of posts did not load on the blog page or in the feed. Apollo.io, Pipedrive and Sales Gravy opened, but their posts show no dates, so I could not confirm any post from this period.

Six blogs had no new posts in these days. Close, LinkedIn Sales Blog, RAIN Group, Harvard Business Review, The Bridge Group and The Sales Blog by Anthony Iannarino. From Salesforce I took only the posts about sales. The other four feed items in the period were about other topics.