sales automations

sales automations

Jason Lemkin has already replaced his SDR and AE reps for 20 sales AI agents paired with 1.2 people. There is tons of nuance in this conversation; I’m going to uncover his points and provide examples parallel to onboarding a complex enterprise application in support of his arguments.

Initially it felt like we had two classes of sales interactions; low touch, high volume, self serve deals that we can upgrade verses high touch, low volume multi-million dollar deals that are typically categorized under the enterprise plan. Let’s take a similar analogy; today we have engineering managers that often run a single team that builds features end to end typically comprised of multiple engineers, a single product manager and a single designer. We’re seeing these roles collapse where a single person PMs the product and its features , designs based on existing styleguides or on demand, builds, ships, tests and monitors features and defines whether it was a success.

We can derive a similar analogy on a single person compressing into doing these tasks typically done by junior hires; generating leads from demand gen lists or new signups, outbound emails, lead qualification, responding to inbound requests, drip follow ups and outreach all through AI agent automations. These tasks already are automated by many teams using SaaS workflows. But unlike existing software that is brittle, and doesnt’ understand nuance, these AI agents can be trained on your past ‘cultural’ behavior. You might not want to reach out to customers on a friday or after work hours in the customers timezone; but instead of defining rulesets, the agent learns that through past behavior implied in your current processes simply by reading your existing email candences and studying your segmentation. This is the kind of fuzzy learning human are really good at learning. He’s advocating that sales managers are going to be managing an army of sales reps that are essentially agents.

He’s advocating for sales leaders to learn how to build, train and deploy these themselves as well as encouraging uprising sales hires to take the opportnity to essentially become the AI sales engineer; a new type of role that might work well with sales leaders to unlock these superpower to gain a set at the table. He’s not convinced that a traditional sales agent can do this role and makes the case sales or growth engineers are better suited to this role.

An understated point is the shift in integration timelines; as the base models become more capable, the speed of integration across enterprises even for complex tasks will drop down to days. In that case even with a human in the loop, who’s optionally available to you if you are so inclined, you’re requesting an enterprise-grade agent with all compliance checks and customizations you can be live in a day. If complex deals drop to days, there is no distinction between the low touch high volume and high touch low volume interactions. Is there?

The current limitations that I do see are the upfront work on training time which can’t be eliminated regardless of plan size. To get trials on pre-trained agents on your own enterprise data you would need the cost of the models to be very cheap and that will continue be the bottleneck for the forceable years. The simplest way is hire agents based on trust through case studios or lower tiers that aren’t trained on your entire corporate corpus.

Agents in this context are agent harnesses that have a base model with some custom functionality; defining tools and actions, training on industry domain and custom data but what happens when the enterprise starts to run multiple agent harnesses quarterly to audit the best agents? And in real time switches out agents the way developers are replacing programming harnesses? Will your enterprise sales cycle still hold up when you can get replaced next month?

Something to think about.