What I'm Building: Substituting Sales Reps With AI
Our early-stage venture into the unknown
The pain-point
There are B2C services that must include a sales call as part of their client acquisition process. Usually, these are services with a high price or complexity, such as education, medicine, or insurance. People leave their contacts, and the sales rep calls them back. (This is not about spam phone calls; all the leads are inbound)
This means that each one of the businesses has a telesales department. The telesales department is usually the most critical and problematic part of such a company. There are 3 big problems that every such business faces:
Scaling. If you want to increase your sales x2 and have 50 salespeople, you will need to hire 50 people. You have 100 people and want to grow 10x? Hire 900 people.
Transparency. When you have a 100-people sales team, you end up with a black box. 100 people produce around 300 hours of conversation per working day. It’s incredibly hard to control what sales reps say during these calls and almost impossible to steer them.
Motivation. People don’t want to work in telesales. I’ve worked with hundreds of telesales reps: not one of them imagined working in B2C sales as a career. It’s an ungrateful job with virtually no career prospects. People burn out quickly. Fraud is widespread: misleading customers, selling to friends who then refund, selling leads to competitors.
But more fundamentally, having such a complex human system at the heart of your revenue generation shifts focus away from the product to sales. Improving your product to get 10% more revenue is hard. Improving your sales to get 10% more conversion is significantly easier. A human sales team is always severely de-optimized. Optimizing the sales department at the expense of everything else is too attractive of an incentive.
For the past 6 years, I’ve worked in such companies: first, a home cleaning startup, and then two online education providers. I’ve scaled companies from ~$0 revenue to $10,000,000, and I’ve seen inbound telesales teams grow from 0 to 150+ people. And it’s that experience that made me want to build a product that makes human telesales teams obsolete.
The solution
With GPT-4 and the latest speech synthesis AI, you can create fully autonomous sales agents that can close a deal. We’ve been working on this for a few months, and the dialogs that we produce are very close to real-life conversations.
With AI telesales reps, businesses can
Scale immediately. Instead of spending time and money on recruitment and training, they can push a button to add more sales reps. Instead of layoffs with their ethical and legal complexities, they can push a button to scale down.
Control every conversation. You can see which approaches led to more sales and fewer refunds, immediately testing different selling approaches at scale with A/B tests
Automate quality control. Companies can make sure their sales reps don’t mislead a single customer, adhere to the QC checklist and remember all the product details.
And finally, they can focus on the customer rather than trying to fix a complex human system that’s incidental to their product.
The business model
AI sales reps have great unit economics. A sales rep’s fully loaded cost in the US is around $4000-$5000. In Brazil, it’s around $3000, and in the Philippines, it’s around $2000. The main variable costs for an AI sales rep are compute cost (voice generation and text generation are the biggest ones, with voice generation being x3-x6 higher) and the customer success team (a project manager, a sales expert and a data scientist/backend developer team).
Today, the total cost per substituted sales rep should be around $1000. Provided the compute cost should go down, it’s fair to expect a healthy 60-80% gross margin, in line with B2B SaaS benchmarks.
This will surely drive competition — what’s the defense?
The moat is in the switching cost. As we work with the client, we will fine-tune our models based on its products and customers. A competitor will always lose in an A/B test. They also won’t be able to undercut us on price to compensate for the lower conversion rate. As AI-substituted sales departments start to scale, the cost of lead acquisition becomes much higher than the cost of the AI sales rep. So higher lead conversion will always trump agent cost.
AI sales reps, despite the heavy competition, will have extremely high switching costs, on par with CRMs.
What worries me
There are a few risks to our business that I keep thinking about.
What if only humans can sell to humans? Let’s say a customer understands that they’re talking to an AI — and AI sales reps have to identify themselves as AI. Will they continue the conversation? Will they make the purchase decision without talking to a real human?
What if you can easily replicate the conversion rate by exporting calls made by the AI and training a model on them? Price competition will ensue, with clients giving out their sales calls to AI sales reps providers and choosing the cheapest one.
What if a large CRM or telephony provider (Salesforce or Twilio) undercuts everyone by price or even provides the service for free? Will they be able to subsidize this product long enough for startups to run out of money?
Where we are now
My co-founder and I have been working on this for a few months and have 3 ongoing pilots. We’re early in the testing process, but the results are very encouraging. I’ve had conversations with AI that were very lifelike, barring the reply delay (for a similar experience, check out Call Gen-Z GPT by the brilliant Vocode team). And it’s super-encouraging to see clients with whom we have pilot projects being so excited about this.
We think that our AI sales rep will make their first proper sale this month.
If you have any thoughts to share about our startup, especially about the concerns I’ve raised or NOT raised — feel free to reach out to me via comments or directly by email at stas@airs.ai (sorry about the lack of a website, didn’t really have the time to put up anything there).
Are you building an AI startup?
If you’re building an AI startup and want to share your ideas and concerns with our readers — let’s talk! We’d love to chat and make a write-up about your startup in The 1993.