AI in sales this week, October 31. Lithia's AI service booking adds 10-11% pre-sales

Once a week I collect cases where a sales or marketing team did something real with AI and showed a result. This week there is a car dealer group where an agent sells service work, and a small team that sells almost only with agents. There are also three cases on how agents find customers while the sales team is asleep. For each case I wrote what a dealership can take from it.

Lithia's agent offers extra work when it books service, pre-sales in 10-11% of bookings#

Lithia Motors from Oregon, USA, is the largest new-car retailer in the US. The group has 308 stores in the US, 143 in the UK and 15 in Canada. CEO Bryan DeBoer talked about AI in service on the earnings call on October 21. WardsAuto wrote about it on October 28.

Lithia's UK stores have used a dealer management system from Pinewood.AI for six to nine months. Its built-in agent talks with the customer and books the service visit. While booking, it offers extra work, such as a fluid change, wiper blades or a tire check. Before, this offer waited for the inspection at the service desk. Now the customer hears it before the visit.

DeBoer says the UK stores get "around 10% to 11% additional pre-sales" from AI bookings. He guesses that before the agent this number was close to zero, which is an estimate, not a measured figure. Lithia is moving its US stores to Pinewood.AI and plans to finish by 2028. Its expenses are now 68.8% of gross profit, and the goal is below 60%.

My opinion. The agent does not sell better than a service advisor. It asks every customer, while a person asks when they are not busy or rushing to the next call. The extra sales here come from discipline.

Any dealer can copy this, even without a new system. I would pick 3 jobs based on mileage and car age and tell the booking agent to offer them to everyone. Then I would check each week what share of bookings said yes. The service advisor confirms the price and the work at the visit. AI only prepares the customer.

Source WardsAuto

At SaaStr, 1.5 people do sales, and the inbound agent booked 614 meetings#

SaaStr from the US runs conferences and sells sponsorships. Founder Jason Lemkin looked back at ten months of agents doing marketing and sales at the company. The post is dated October 28 in the blog feed.

SaaStr has 3 people and more than 20 agents in production. All cold outreach went to agents, and they send about 3,200 emails a month. A good human SDR at a company of this size sent 75 to 285. One agent answers inbound questions in the website chat. Another one wrote to about 1,000 leads who had stopped answering.

The agent 10K started as an AI VP of Marketing. It ran 3 deals with people watching, and only the fourth on its own. People decide which work goes to agents, which lead goes to which seller, and when to trust the result. A person still presses publish on LinkedIn and X ads. Agents report on their own work, and people check the report against the CRM.

Lemkin says the inbound agent closed more than $1 million in deals and booked 614 meetings from 442,000 chats. The win-back emails had a 72% open rate. The human sales team went down to 1.5 people through normal attrition, with no layoffs.

Lemkin also gives an example from another company, based on Salesforce. At PayPal, agents now work about 8,000 leads a month that nobody used to call. Meeting conversion went up 50% in 14 weeks.

My opinion. 614 meetings from 442,000 chats looks small. But each of those chats would get no answer if a person only ran the chat from 9 to 6.

What I like most is the launch order. Three deals with supervision, then on its own. In a dealership I would start the agent on old leads where the customer went quiet. In the first week the sales manager checks every message. From the second week the agent writes alone, and a salesperson calls everyone who answered.

Source SaaStr

TaskUs started its AI seller with one rep and now answers every lead#

TaskUs from the US gives companies teams for customer support and AI data work. The second case in the same post is about the Cloud Protection by WithSecure team. WithSecure is a cybersecurity company from Finland. Salesforce described both cases on its blog on October 27.

TaskUs launched an AI SDR called Liz with one rep and one gap where leads got no answer. After the pilot worked, the sales team used it more widely. The agent suggests an action, and a person confirms it.

WithSecure built a longer chain. First, website logs go through Data 360 and ZoomInfo to show which company visited the site. Then the lead score is updated from more than 300 signals, such as page views and email opens. The score goes down when the customer goes quiet.

An AI agent reads every reply to an email campaign and decides if the person wants to buy. It sends ready buyers to a seller with the email thread and company details.

According to Salesforce and the companies, TaskUs now follows up on 100% of leads, and sellers focus on qualified deals. Building a newsletter at TaskUs went from 16-20 days to 10-12 days. WithSecure says it identifies 98.5% of companies that visit its site. Two employees built the system in six months. This is Salesforce writing about its own customers.

My opinion. The strongest number here is 100% of leads with a follow-up. Everything else matters less while part of your leads get no answer at all.

I also like how TaskUs started. One seller, one gap, then wider use. Dealer measurements in different countries show that 40% of inquiries come outside working hours. I would start there. The agent answers at night, and in the morning the lead with the full conversation is waiting for a salesperson.

Source Salesforce

Clay catches when a former customer joins a new company and drafts an email for the rep#

Clay from the US makes a tool to find information about customers and companies. On October 27, Clay's growth team described its play "Paid Alumni Job Changers" on the blog.

The idea is this. A leader leaves a company that pays for Clay and joins a smaller company that does not use Clay yet. For the old customer it is a churn risk. For the new company it is a reason to sell. The play has five steps.

First, the Signals tool finds the job change every day and logs it in Salesforce. Then comes the filter. The old company paid and was not on a trial, the person is director level or higher, and the new company is the right size. Then come six checks in a row. For example, the account must have an owner with a booking link, and the person must not have been contacted in the last 30 days.

AI writes the email under clear rules. It never criticizes either company and never invents what tools the old company used. In the last step the person goes into an 8-step sequence in Gong Engage. That is three emails, three calls and two LinkedIn messages. The rep reviews and sends.

Clay's numbers are small and honest. About one new record a day, and 62 people met all conditions in 60 days. Clay does not give deal results. The company shared the play as a ready skill for agents.

My opinion. The key rule here is timing. A new leader chooses tools in the first weeks at a new job. After that the window closes.

A dealership has signals like this too, but nobody catches them. A customer who bought two cars from you moved to a company with a fleet. Or a loyal service customer recently had a baby, and the sedan will soon be too small. I would start with one reason like this and let an agent look for it in the CRM every day. The call stays with a salesperson.

Source Clay Blog

Nooks counted sales reps across its market and found 700 accounts it did not know about#

Nooks from the US makes a platform for calls and for sales reps working with agents. Its case was told by Anthony Goldbloom, founder of Sumble and earlier founder of Kaggle. SaaStr published a write-up of the livestream on October 27.

Nooks sells to sales teams, so it needs to know how many reps each company has. It used Sumble to split accounts between its sellers. An accurate SDR count per company showed 700 accounts Nooks did not know about. Goldbloom says this was enough to double its enterprise sales team. And this came from territory planning alone, before ads or inbound.

Goldbloom explains why accuracy matters more than volume. To count reps in a company, you need five steps in a row. Company data, matching people to companies, fresh records, department, and job level. If each step is 80% right, the end result is right only about a third of the time. The errors multiply.

Sumble aims for 99% per step, which gives about 95% at the end. Goldbloom admits they are not there yet.

My opinion. The first wave of AI in sales sent more emails. Here AI decides who to write to at all, and that brought more than any email campaign.

A dealer has the same math. An agent that writes from a database with duplicates, old phone numbers and wrong cars makes mistakes at every step. Before any outbound agent, I would clean the database. One customer, one record, a current phone number, and the car they drive now. If you do not record every contact with a customer, the agent has nothing to work with.

Source SaaStr

Also this week#

• Databricks keeps its SDR team flat while email reply rates grew to 6-10%
• CarEdge's AI agent contacted dealers 160,000 times for car buyers
• Two AI agents wiped deals and code at SaaStr, and unplanned copies saved them
• DoorDash got only several hundred orders from third-party AI agents
• Walmart opens its store to agents that shop for customers
• Zocdoc lets AI agents find doctors and book visits
• Google expands AI Search ads to all markets
• Advertisers pull Meta ads that AI changed without asking
• Publicis platform Marcel builds agents for more than 90 clients
• Salesforce adds a way to stop failing or costly agents in Agent Fabric