San Francisco – There needs to be clarity on how generative artificial intelligence (AI) will be priced, as supplier offers come in and companies seek to avoid hidden costs.
Businesses are especially concerned about facing billing shock under consumption-based billing models, similar to what’s happening with cloud computing, Tim Dillon, founder and director of Tech, told ZDNET. ResearchAsia.
It’s a problem that vendors like Salesforce will need to address as they expand their generative AI service offerings, he said in an interview on the sidelines of Dreamforce 2023 in San. Francisco this week.
How much will AI cost businesses?
Because the adoption of these tools can grow organically within an organization and, therefore, lead to a lack of control and awareness of their consumption. There are also often no policies guiding the use of generative AI, he said.
Acknowledging that concerns about billing shock are well-founded, Gavin Barfield, VP of Salesforce, said pricing models are still being defined as generative AI services are rolled out.
“We’re in the early stages, so all companies are grappling with these questions,” Mr. Barfield told ZDNET. He noted that the same problems arose when cloud computing services were launched.
Credits imitated to use AI
“As the market and product matures, these issues will be resolved,” he said, adding that market participants will need to find ways to price generative AI services. Salesforce is exploring different pricing models, but has so far settled on a credit-based system for a few services, according to Barfield. The amount of credits consumed depends on how the AI model is used to execute the query.
In July, Salesforce announced that Sales GPT, bundled with Sales Cloud Einstein, is available for $50 per user per month and includes a limited number of Einstein GPT credits. Service GPT, bundled with Service Cloud Einstein, is also priced at $50 per user per month and also includes a limited number of Einstein GPT credits.
Customers of either generative AI service can purchase Enterprise Expansion Packs to earn more credits as usage increases.
The ROI of AI depends largely on its cost
As generative AI services are based on a usage model, it is essential that businesses can monitor their consumption, said Jan Morgenthal, chief digital officer of Singaporean telco M1.
He stressed the need to be able to measure and predict the use of these tools within his organization. M1 is currently using several AI tools from different vendors, including Salesforce, and is also testing generative AI services.
Having a dollar value, for example, will allow it to manage the number of queries that need to be made with these tools.
Mr. Morgenthal noted that depending on the complexity of a particular use case and the AI model needed to automate or generate a response, it may not make sense, in terms of ROI, to power the query with generative AI.
This is a point on which companies will have to be careful, otherwise costs will skyrocket. The automation achieved through generative AI might then not be worth the cost of implementing it, he added.
This also means that organizations need to define the processes, including data availability, required to run a query and achieve the desired result, so they can measure the cost of applying generative AI to the use case.
Let clients create their own prompt
This week, Salesforce previewed new generative AI offerings that its executives said would make it easier for enterprise customers to customize these tools to support their operations.
Among them is the Einstein Copilot, presented as an AI conversational assistant that can be integrated into any Salesforce application, allowing users to ask questions in natural language.
Responses are generated from the company’s proprietary data, powered by Salesforce Data Cloud, formerly called Genie. The data engine brings together all data sets, including customer data, telemetry data, and Slack conversations, to create a unified view of the customer.
Create and test prompts consistent with the company’s brand and communication style
According to Salesforce, Data Cloud currently processes 30 trillion transactions per month and connects 100 billion records per day. The data engine is now natively integrated into Einstein 1 Platform, enabling businesses to apply AI, automation and analytics to every customer experience.
It allows Einstein Copilot to provide options for additional actions beyond the user’s request, such as a recommended course of action after a sales call.
Organizations that want to create generative AI applications with their own prompts, skills, and custom AI models can also do so through Einstein Copilot Studio. It includes the Prompt Builder, which allows users to create and test generative AI prompts consistent with their company’s brand and communication style.
“Competitor analysis”: know everything about a competitor thanks to AI
They can do this without any technical expertise, allowing marketers to ask Prompt Builder to generate a personalized message based on a customer’s purchase or order history.
Einstein Copilot Studio also includes Skills Builder, which lets you create custom AI-driven actions to perform specific tasks. For example, it is possible to create a “competitor analysis” skill that analyzes current market data, sales figures and sends API calls to extract data from external sources.
Additionally, a Model Builder component allows organizations that wish to use their own AI models. They can choose one of Salesforce’s proprietary LLMs (large language models) or integrate their partners’ predictive and generative AI models. They can train them on Data Cloud data without moving or copying the data.
Model Builder should eventually support external LLMs
This means that Einstein Copilot can provide more accurate information and content, tailored to the dynamics of the company’s employees or customers.
Model Builder is expected to eventually support external LLMs, including Amazon Bedrock, Google Cloud’s Vertex AI, Anthropic, and Cohere. For now, it only supports OpenAI.
Einstein Copilot is currently in pilot phase, while Copilot Studio will enter pilot phase later this fall. Einstein Trust Layer enhancements will be available in the Einstein platform starting October 2023. No pricing details are available for the new offerings.
Ensuring user security with Einstein Trust Layer
Data Cloud is currently included free for Enterprise edition customers or higher. It encompasses capabilities that allow organizations to unify 10,000 customer profiles and includes two Tableau Creator licenses.
Additionally, a new Einstein Trust Layer will underpin all Einstein products, providing a secure AI architecture that Salesforce says will ensure its customers’ generative AI responses are powered by quality data verified against potential bias and security and confidentiality standards.
Integrated with Salesforce Data Cloud, Einstein Trust Layer mitigates these risks, checking data against risks, such as hiding personally identifiable information and not retaining customer data.
“Everyone can now be an Einstein”
According to Salesforce, Einstein Trust Layer enhancements will be available in October 2023.
“The reality is that every business is going to undergo an AI transformation to increase productivity, drive efficiency, and deliver incredible experiences for customers and employees,” said Marc Benioff, president and CEO of Salesforce. “With Einstein Copilot and Data Cloud, we make it easier to build powerful AI assistants and integrate trusted AI into the workflow for every job, every company, and every industry. In this new world, everyone can now be an Einstein.”
Based in Singapore, Eileen Yu reported for ZDNET from Dreamforce 2023 in San Francisco, USA, at the invitation of Salesforce.com.
Source: “ZDNet.com”