The last few posts addressed changes in the tech world driven by the AI revolution. Everything from staffing to usability and ultimately to business valuation has been thrown into chaos. We are seeing fundamental changes in the way software companies operate that result in very different cost structures, and we are seeing investors focus on different financial metrics that reflect what a good AI-driven business should look like. For software companies, SaaS metrics are still relevant but not necessarily sufficient to truly understand the health of a business. There is a lot of inertia in a large-scale SaaS company that can mask the impact AI is having on their business. Recurring contracts represent the majority of SaaS revenue, and this flywheel has historically made SaaS businesses so valuable. However, the recurring revenue inertia may be masking shifting fortunes as AI permeates the market. Replacement cycles are shrinking, add-ons are becoming less valuable, and as customers reduce their staff, the number of users is declining. We can see the reduced expectations in lower public market valuations for the largest SaaS vendors (Salesforce.com, HubSpot, etc.).
With these headwinds, companies need to assess their approach to embracing AI tools. Many companies have taken a 1+ approach to AI. They start with all of their existing staff and costs, and add AI tools and capabilities on top of their current expense base. Like a street drug, initially the AI costs are small, and the benefits seem magical. As adoption grows, the costs quickly grow, and the margins start to decline. AI tools make everyone more productive, but they also make them more expensive. If an engineer can become more productive using AI, but the AI costs are equivalent to adding another human, unless the company budgeted for the extra costs, there is going to be a problem if the productivity does not translate into increased revenue. We are seeing layoffs and redundancies across the board as companies recognize that productivity gains come at a cost, and prudent financial management requires companies to right the ship.
The real challenge on the horizon is the squeeze businesses are facing as they deliver natural language and AI-driven interfaces to their customers. We have all become enamored with conversational application interfaces. We can simply ask a question and the interface will figure out what it means and how to answer it - magic. Behind the scenes, the AI engine is burning through tokens that cost the vendor money. Traditional licensing models are based on fixed pricing that is tied to something like number of users, and behind the scenes the vendor’s computing costs to deliver the results are well understood. Buyers want predictable costs that they can rely upon to secure budget approval for purchases, vendors want predictable computing costs, and CFOs on both sides do not like unpredictable expenses that fluctuate wildly. However, vendors do not have a good mechanism (yet?) to predict the AI costs of their customers’ random use of AI-based interfaces. The result is that vendors are caught between a rock and a hard place, needing to offer fixed, recurring contract pricing that is predicated on randomly fluctuating costs to deliver.
The beautiful SaaS model of committed recurring revenues that was the foundation for valuing SaaS vendors, is now creating an impediment to predictable profitability and business health. The answer is going to require a shift in application pricing. Back in the days when SaaS first appeared, recurring license models introduced a shift from one-time up-front perpetual licenses with small fixed annual maintenance fees to an annual recurring license cost. Vendors initially met significant resistance from business buyers, but eventually CFOs accepted the new reality, and the SaaS model became the standard. So too, we are now going to see a shift toward consumption-based or transaction pricing. It is currently evolving as a hybrid price structure. Vendors are charging a SaaS-like fixed license fee for their platform that includes some level of consumption or usage, but beyond the included activity, the buyer will be charged for consumption. On the vendor side of the equation, the fixed license fee will fund the operational costs of the business, and the consumption fees will cover the variable AI engine costs (plus a margin).
How long the hybrid model lasts will depend upon how the winds blow in the marketplace, and how AI pricing eventually settles down. The historic progression of pricing models in financial institutions provides a picture of this type of dynamic. When the cost of stock trades was high, brokers required minimum 100 share transactions, and the fee for trading was their primary source of revenue. As the cost to trade decreased, discount brokers permitted trades in any increment, and full-service brokers shifted from trading fees to asset management fees, with ‘free’ trades included. A primary driver underlying the shift was the declining back-end cost to execute a trade. A similar shift occurred in cell phone pricing. Initially, minutes were costly and consumer contracts were comprised of a base fee that included some minutes, and a per-minute charge for overage. When competition drove down the cost per minute on the backend, consumer contracts became fixed with unlimited minutes. Similarly, the AI cost of buying and consuming tokens will determine pricing behavior in the application vendor market. If AI costs become de minimus, we will see a return to platform pricing - not unlike cell phones carriers moving to fixed pricing for unlimited usage. If AI transaction charges continue to grow and remain unpredictable, we will see increased pressure to shift toward purely consumption-based pricing.
The focus for application developers must remain on business fundamentals. In particular the guiding metrics should be gross margin and operating margin. Gross margin reflects the revenue collected in comparison to the cost of goods sold. With AI costs fluctuating and consumption pricing evolving, vendors need to stay on top of the blended revenue and the blended COGS to preserve their gross margin available to fund the business. On the operating costs, once again we will have a blended people plus AI cost structure for each functional area of the business that adopts AI-based productivity tools. People costs and AI costs will become fungible, so we need to manage the total operating costs and insure we preserve operating margins. That means harvesting gains in efficiency by offsetting the AI costs against the traditional costs which unfortunately may lead to staff reductions. Companies that make the hard decisions to rebalance their costs to align with their productivity will be positioned to thrive and generate profits. Companies that lose sight of their gross margin and operating margin are doomed to under perform and fail.