News · Nov 07, 2026 · 9 min
AI in sales this week, November 7. Vercel cuts its SDR team from 10 to 1
Vercel cuts lead qualification from 10 people to one, Rocket Mortgage closes deals three times more often with an AI assistant, Supermetrics runs ads through a conversation with Claude, and CarMax hands calls to a voice agent. How it works, the numbers, my opinion.
Once a week I collect cases where a sales or marketing team did something real with AI and showed a result. This week a company moved almost all lead qualification to one agent. A mortgage lender gets three times more closed deals with an AI assistant. There is also a conversation with ads through Claude, and calls answered by a voice, not a person. For each case I wrote what a dealership can take from it.
Vercel handed lead search to an agent and cut the team from 10 people to one
Vercel from the US builds a platform for developing and running websites. Its COO Jeanne DeWitt Grosser talked about this work with AI agents at the SaaStr AI conference. SaaStr published a write-up of the talk on November 6.
A go-to-market engineer watched the company's top salesperson for a few days. He logged every data source she used. LinkedIn, BuiltWith, the client's website, the CRM, and Slack history. Each step became part of the agent's workflow.
One engineer built the first version over a weekend. For six weeks the agent ran in the background while the same salesperson reviewed every answer and gave corrections. Once there were no more corrections, she stepped out of the loop. The support agent, Vertex, was built the same way by three engineers over two months, after an off-the-shelf product did not satisfy the company.
Grosser says lead qualification went from 10 people to one. That person runs the US market, plus a fifth of another person's time for Europe and Asia. The agent costs under $5,000 a year to run, plus a fifth of an engineer's time to maintain. Grosser counted the return as 32 times the cost of the old team.
The agent works 24 hours a day, and its quality, she says, matches the human team. The support agent, Vertex, handles 93% of cases for about $150,000 a year. Similar projects at other companies, Grosser says, use teams of 150 engineers.
My opinion. The AI here did not invent a new way to find customers. It copied the path of the top salesperson, step by step, without skipping steps or getting tired.
I would start a dealership the same way. I would sit a manager next to the top salesperson for a week. I would write down what she checks before a call. Which pages the customer viewed on the site, whether they visited for service, what they looked at in used cars. Then I would hand those steps to an agent and keep the salesperson checking every answer at first.
Source SaaStr
Rocket Mortgage closes deals three times more often with its AI assistant
Rocket Mortgage from the US lends mortgages, and its chat and voice agent is built by Sierra. The details and numbers come from Rocket Mortgage's page about the Sierra partnership, on Sierra's own site.
The digital assistant answers the customer in chat and by voice. It asks questions, pulls credit history, shows a rate for that person, offers loan options, and helps with payments. The assistant guides the customer from first interest to preapproval, then hands them to a banker. The project started as a pilot in May, with no year given.
Rocket Mortgage says customers who start with the digital assistant close a deal three times more often than those who do not use it. Customers who go through both chat and a banker close four times more often for refinancing and for home purchases. The assistant runs more than 400,000 successful chats and more than 1 million outbound calls a month, and both numbers are growing.
My opinion. The gap between a 3 times and a 4 times conversion rate answers the argument about whether AI will replace a salesperson. Chat prepares the customer, the person closes the deal, and together they work better than either alone.
A dealer has the same setup with financing and trade-ins. I would let an agent take the first credit application, calculate the payment, and show options. I would leave the final talk about the rate and terms to a finance manager. The customer arrives with numbers already in mind, not starting from zero.
Source Sierra
Supermetrics put a marketer in conversation with ads through Claude
Supermetrics from Finland collects marketing data from different ad platforms. The customer story about its connection to Claude from Anthropic is on the Claude customers page.
A marketer asks Claude why a campaign did not perform. Claude calls Supermetrics tools and gets data with its source, time, and how it was calculated. Claude then explains the cause and suggests an action.
If something needs to change, Claude can create or edit a campaign on Google Ads, Meta, Microsoft Advertising, TikTok, LinkedIn, or Snapchat. But it creates the campaign paused. A person approves it on a separate approval page, and the client sets the approval level.
Supermetrics says the number of active users of its Claude connector has grown by an average of 250% a month since the February launch. Layer, a 25-person agency in Norway, cut the time to build a client report from 10 hours to 20 minutes. Of those 20 minutes, 17 go to loading data into the client's spreadsheet. Layer's strategist, Morten Kleven, says he has not found a single error or invented number across weeks of checking.
My opinion. The most important part here is not the speed of the report. It is that the campaign does not launch without a person's approval. That removes the main fear about AI in advertising, that it spends the budget on its own.
A dealership's ad work can use the same setup. I would let AI prepare the explanation for why a campaign on a specific model did not work, along with ready fixes. I would leave the decision to use them with the marketer who owns the budget.
Source Claude
At Salesforce, an employee assistant's use grew 21 times in three months
Salesforce from the US said on November 4 how use of Agentforce Coworker has grown. Coworker is an assistant that, using Salesforce data, splits work among agents and workflows and shows the result for a person to review. Salesforce named Fulton Bank and The Adecco Group as customers.
Coworker works within an employee's access rights and does not act without approval where that is required. In sales, it checks the risk that a customer will not renew. It pulls service history, finds open cases, and drafts a follow-up email. It builds a meeting brief and a renewal plan from emails and a calendar. It also turns a top salesperson's playbook into a skill the whole team can use.
Salesforce says Coworker's use grew 21 times in three months, and more than 185,000 people now use it. Fulton Bank turned on more than 20 use cases for 3,000 employees, and a company executive called it a game changer. The Adecco Group is rolling it out in 40 countries so staff can see a candidate's and a client's history in one place.
My opinion. The strongest part here is reusing a top salesperson's playbook. That kind of skill usually lives in one person's head and leaves the company when they do.
I would tell a sales manager to start with one such skill. For example, how the top salesperson leads a customer to renew a service contract, and write it down as steps for an agent. Then everyone on the team can use that skill, not only the person who invented it.
Source Salesforce
CarMax handed calls to a voice agent so staff can focus on the buyer
CarMax from the US, the largest used-car retailer, has used voice agents from Sierra for inbound calls since May. The partnership is described on CarMax's page on Sierra's site.
The agent answers the call and asks questions to find out what the caller needs. It answers common questions itself, such as store hours or whether a car is in stock. When a staff member is needed, the agent transfers the call without making the caller repeat the question or wait. This holds no matter how busy the store is or what time it is. CarMax plans to add booking for appraisals, inspections, and test drives next.
CarMax says the share of calls resolved on the first try has gone up since launch, and the share of unresolved calls has gone down. The company gives no exact figures. CTO Shamim Mohammad says staff are freed from simple questions and can guide buyers through bigger decisions.
My opinion. Without numbers in the source, this is a promise, not a proven result, and I would not repeat it on the company's word alone.
But I would borrow the idea itself. Start with one category of questions, such as store hours and whether a car is in stock. Track how many calls the agent closes without a transfer to a person. If that share grows, add more questions. If it does not, find out why.
Source Sierra
Also this week
- Clay compared its agent to a generic one: 93 times cheaper and better at catching planted deal risks
- SaaStr ran an email deliverability test across 500,000 emails and three vendors, with its own AI agent running the work
- Salesforce agreed to buy Cresta
- Amazon lets a small test group text its AI shopping assistant
- Half of Slack messages are no longer from people
- AI is both Canva's biggest bet and its biggest threat
Read next
- Nov 17, 2026
NewsSales blogs, November 17. How to check your team's resultsFour reads on marketing targets, deal stages, sales metrics and AI value. The main ideas and practical checks for a sales team.3 min - Nov 10, 2026
NewsWhat sales blogs wrote, November 10. Build the AI agent on your best sellerVercel on cutting a 10-person lead team to one, HubSpot on when a team needs an AI SDR, Outreach on AI that saves hours but is hard to measure, Bridge Group on the sales kickoff, and SaaStr on an email vendor test. Short facts, my opinion, a link to the article.12 min