The Four Strategic Questions Facing AI Agencies
Since writing The AI Agency: A Novel GTM for Machine Learning Startups, I’ve been meeting many companies who operate this way. These startups use machine learning to disrupt an industry traditionally dominated by agencies: law, accounting, recruiting, translation, debt collection, marketing…the list is long. I will publish a landscape soon on the area. If you’re operating an AI Agency, I’d love to hear from you.
In meeting many of these innovative businesses, I’ve observed they face four strategic questions.
First, to sell to the agency or to be the agency? This is an early strategic question, perhaps the first strategic question an AI Agency will face. Many startups start out selling to agencies and then run into a wall.
Classic agencies don’t value the software enough to engender pricing power, develop fast sales cycles, or change the operations of their business to maximize the value of the ML innovation. In some cases, agencies sell their time. Increases in productivity don’t imply increases in revenue. Other times, agencies prefer to operate they’ve always done.
Consequently, product market fit is weak. It’s possible to build a business, but hard to achieve lift-off. At that point, do you continue the business you have or compete with your customers?
Most of the successful startups decide to compete with their customers by building an agency. Rather, they masquerade as an agency with a completely different engine, one powered by algorithms. The combination results in a larger market size with better gross margins than a classic agency.
Second, how to price the product relative to the competition? At the outset, many AI Agencies price at a substantial discount to the market for two reasons. Gathering training data is critical and more valuable than initial revenue dollars. They market their technology as differentiation and buyers approach the business with some skepticism, also pushing down the price. These are two short term pricing headwinds.
Ultimately, AI Agencies should price at comparable levels or even at a premium. And they ought to develop that pricing power. They will provide higher quality, more consistent work products faster than the competition.
Third, how to employ the labor force? There are four models: as full time employees, as channel partners, as customer employees, or in a true marketplace. There is no consistent pattern across the startups I’ve met in their decision.
Some hire full time employees to accomplish primarily to control the end-to-end user experience for very high quality. Some go to market with channel partners (BPOs, consulting agencies, marketing agencies) who benefit from the technology, sell into an existing customer base and provide a skill that the agency is unwilling to furnish themselves. Still others create a skilled labor market place of individual contractors who complete the work. The last group uses a team within the customer to work the software.
There is no pattern primarily because each industry has its own evolution and dynamics that govern it.
This labor force question is an important one. It impacts margin, ability to generate feedback data to improve ML models and sales processes. It’s worth testing and figuring out. Many AI Agencies start with one model and then migrate to a second.
Last, what is the optimal type of AE: familiar with the domain or familiar with the technology? Across the companies I’ve met, more seem to succeed with AEs from the domain.
Relationships matter most in agency sales. Buyers either love or hate their agencies. This bipolar nature of the relationship reinforces the benefit of AEs with existing relationships who should ramp faster and drive sales quicker.
In addition, many AI agencies focus on enterprise accounts to drive higher ACVs. This strategy is consistent with the trade of lower gross margins for greater TAM compared to a software vendor. Hiring more senior AEs follows this upmarket strategy.
In this novel GTM, these strategic questions seem to be the most important strategic questions. Different paths will lead to business models, data aggregation and margin structures than can differ significantly. If you’re building one of these businesses and have a comment or another point of view, please contact me.