In January, Microsoft announced Copilot Pro, a paid version of Microsoft Copilot.
Then, in mid-March, the tech giant extended the application by allowing users to create custom GPTs.
As with OpenAI’s GPT Store, users could hard-code those custom GPTs to perform particular tasks within various applications.
However, just three months later, Microsoft announced that it would remove that capability for consumers starting July 10, 2024.
In a surprise web post, the tech giant noted: “We are continuing to evaluate our strategy for consumer Copilot extensibility and are prioritizing core product experiences while remaining committed to developer opportunities.
To this end, we are shifting our focus on GPTs to Commercial and Enterprise scenarios and are stopping GPT efforts in consumer Copilot.
By making such a statement, Microsoft hints that it has not made enough money through its 365 subscriptions to account for the significant cloud computing costs that come with ChatGPT.
As such, its sales teams are shifting the focus to selling these capabilities to enterprises.
Yet, Christopher Rice, Managing Partner at Refuturing, who has also predicted the shutdown of OpenAI’s GPT Store, questions this strategy.
“The economics don’t work because the costs are too high,” he wrote on LinkedIn. “There’s no sustainable business model for Generative AI (GenAI) as a product.
If Microsoft doesn’t start moving more Enterprise MS 365 and Teams subscriptions with their vision of Recall-style worker surveillance and promises of “productivity?” Hey, remember Cortana? Clippy, anyone?
Similar concerns that the current economics of LLMs don’t quite add up have often hit the headlines ever since reports surfaced last year suggesting that ChatGPT costs $700,000 to run daily.
Earlier this month, that narrative again emerged as OpenAI and Microsoft pitched their GenAI goods to the US military, a strategy that the latter employed when trying to gain some sort of momentum around HoloLens.
Now, that is not to say that enterprise applications of GenAI aren’t sustainable. However, it’s a warning that this space is moving quickly, and even Microsoft is still toying with its strategy.
That gives enterprise GenAI adopters significant food for thought.
A Warning for Enterprise GenAI Adopters?
As the Copilot Pro GPT Builder bites the dust, organizations must reconsider “boil the ocean” GenAI projects – especially those tied to bulky LLMs.
Instead, enterprises should start small and implement only LLM-agnostic GenAI apps to safeguard their implementations from future price hikes, which may happen if the likes of OpenAI and Microsoft fail to get a grip on their expenditures.
An example of such a vendor-agnostic app in the contact center space is the Five9 GenAI Studio, which allows businesses to click-and-customize the LLM and prompts behind each use case.
By following such a model, businesses can switch to smaller, more focused LLMs to lower their costs in the long run while validating the models for accuracy.
Those smaller LLMs are coming, with AWS and Microsoft Azure developing SaaS-based small model gardens to reduce data center costs and deliver the tech at a more affordable, predictable price.
Meanwhile, Google is engineering on-device LLMs to eliminate cloud costs where possible.
Yet, the fall of the Copilot Pro GPT Builder serves as a timely reminder to businesses of the cost economics of GenAI.
As such, CX leaders must embed cost control into the decision-making process behind every implementation, establishing well-defined parameters for roles, metrics, and timelines.
Furthermore, it’s crucial to dashboard all the costs related to GenAI projects, with automated alerts for high resource consumption.
From there, the team can spot where GenAI delivers cost-effective results and strategically expand the go-forward strategy.
Those positive results are happening across the CX space. Consider Klarna. Its GPT-based chatbot reportedly does the work of 700 full-time customer service agents.
Then, there’s Microsoft itself. Its contact center deployment of Copilot resulted in a 12 percent drop in average handling time (AHT).
However, with newfound fine-tuning, prompt engineering, talent, security, and cloud costs, enterprises still have much to consider.
