EDITOR’S NOTE: One of the industry’s top enterprise architects from ConvergentIS took the time to talk with me and fill in any gaps from my research on the relationship between ERPs and GenAI. This will be the first in a series of posts that will be published today and throughout December.
Do ERP systems use horizontal GenAI models or vertical GenAI models?
MY RESEARCH
ERP systems (Enterprise Resource Planning systems) can leverage GenAI (Generative AI) models in two primary ways: using horizontal GenAI models or vertical GenAI models. The choice depends on the application and the business needs within the ERP context. Here’s how they compare in relation to ERP systems:
Horizontal GenAI Models in ERP Systems
Definition: Horizontal GenAI models are broad, general-purpose models trained on a wide variety of data sources across industries.
Examples: Models like GPT (OpenAI), Bard (Google), or other large language models (LLMs) that can perform tasks such as summarization, text generation, or question answering across domains.
Vertical GenAI Models in ERP Systems
Definition: Vertical GenAI models are tailored to specific industries or domains, trained on specialized datasets relevant to those fields.
Examples: Models fine-tuned for specific ERP use cases, such as inventory optimization in retail, production planning in manufacturing, or financial analytics in banking.
Key Differences Between Horizontal and Vertical GenAI Models in ERP Systems
Feature
Horizontal GenAI Models
Vertical GenAI Models
Scope
General-purpose across industries
Industry-specific and domain-focused
Training Data
Broad datasets across multiple domains
Specialized datasets tailored to specific industries
Use Case Flexibility
Versatile for general ERP applications
Precision for industry-specific needs
Implementation Time
Faster to deploy and adapt
Requires fine-tuning for specific use cases
Accuracy
Moderate for specific tasks
High accuracy for domain-specific scenarios
Current Trends
Many ERP vendors are integrating horizontal models first due to their scalability and ease of deployment. Over time, they pivot to vertical models for high-value, industry-specific applications.
For example:
SAP uses AI services like SAP AI Core for industry-specific AI, while incorporating broader capabilities via LLMs for general tasks.
Oracle employs machine learning (ML) in Oracle Cloud ERP, focusing on industry-specific automations like revenue recognition in financial services.
Conclusion
ERP systems use both horizontal and vertical GenAI models to deliver broad functionality and domain-specific precision. Horizontal models handle generalized tasks like natural language interaction and basic analytics, while vertical models provide tailored insights and automation for industry-specific processes. Many ERP vendors adopt a hybrid approach, leveraging the strengths of both types of models for a comprehensive AI-powered solution.
ConvergentIS
In my experience with SAP, they have evolved pretty much as you have outlined above. They have enabled an AI framework with AI-Core that allows integration with multiple LLMs initially in a horizontal model and have hinted at “roadmaps” for enabling their own domain or industry specific models through BTP in the future (although I have yet to see any of this materialize). The embedded AI use cases within their applications are more ML thus far than GenAI, and the GenAI use cases released into their commercially available software are in the early stages (e.g., generating job descriptions in SuccessFactors – things similar to what most people do with ChatGPT pretty much every day).
IMHO, innovation in the vertical area will be driven by smaller niche vendors and customers developing their own focused use cases (which could then potentially be productized). For the most part, ERP vendors don’t have the agility to keep pace nor the focus to deliver something specific enough to deliver value and meet the reliability tolerances needed for customer production usage.
Using a GenAI-based orchestration engine (like our RIO) to facilitate automated orchestration of the models based on the context of the prompt makes a lot of sense – especially if you can use that engine to compare models during development and ongoing testing so that the orchestration of the models can actually evolve over time too. This would only be possible by using something like RIO instead of programmatically defining the model based on the application functionality.
MY TAKEAWAY
What do you see in the graphic below?
Here is what I see:
Best of Suite – In theory, end-to-end back-end platforms consisting of multiple Best of Breed front-end bolt-on ProcureTech solutions. (Think Mendocino Project, DUET, and Microsoft Office concept)
Best of Breed – Bolt-on systems transformed to front-end Apps (up to 75% of which will be gone by the end of 2025 via acquisition, absorption, or closing down)
Build Your Own Suite – The early stages of the Agent-based AI operating systems will be extended to interface with external 3rd Party Stakeholder Apps and platforms, e.g., ERP systems, suppliers, courier companies, and customs. (Watch this video for more details) – https://bit.ly/40fJxNY
NEXT POST IN THE SERIES: How do you integrate horizontal and vertical GenAI models seamlessly?
30
Here is the framework reflecting orchestration and intake from my patent document for my ProcureTech Solution in 1999 – at the time, it was called a Metaprise:
Read the above post again and then look at this image – From RAM (1999) Metaprise To RIO (2024) Orchestration and Intake. (NOTE: The Agent-Based Development and Implementation model occurs within the center hub.)
Procurement Insights
What Is The Relationship Between ERPs And GenAI?
Posted on November 30, 2024
0
EDITOR’S NOTE: One of the industry’s top enterprise architects from ConvergentIS took the time to talk with me and fill in any gaps from my research on the relationship between ERPs and GenAI. This will be the first in a series of posts that will be published today and throughout December.
Do ERP systems use horizontal GenAI models or vertical GenAI models?
MY RESEARCH
ERP systems (Enterprise Resource Planning systems) can leverage GenAI (Generative AI) models in two primary ways: using horizontal GenAI models or vertical GenAI models. The choice depends on the application and the business needs within the ERP context. Here’s how they compare in relation to ERP systems:
Horizontal GenAI Models in ERP Systems
Vertical GenAI Models in ERP Systems
Key Differences Between Horizontal and Vertical GenAI Models in ERP Systems
Current Trends
Conclusion
ERP systems use both horizontal and vertical GenAI models to deliver broad functionality and domain-specific precision. Horizontal models handle generalized tasks like natural language interaction and basic analytics, while vertical models provide tailored insights and automation for industry-specific processes. Many ERP vendors adopt a hybrid approach, leveraging the strengths of both types of models for a comprehensive AI-powered solution.
ConvergentIS
In my experience with SAP, they have evolved pretty much as you have outlined above. They have enabled an AI framework with AI-Core that allows integration with multiple LLMs initially in a horizontal model and have hinted at “roadmaps” for enabling their own domain or industry specific models through BTP in the future (although I have yet to see any of this materialize). The embedded AI use cases within their applications are more ML thus far than GenAI, and the GenAI use cases released into their commercially available software are in the early stages (e.g., generating job descriptions in SuccessFactors – things similar to what most people do with ChatGPT pretty much every day).
IMHO, innovation in the vertical area will be driven by smaller niche vendors and customers developing their own focused use cases (which could then potentially be productized). For the most part, ERP vendors don’t have the agility to keep pace nor the focus to deliver something specific enough to deliver value and meet the reliability tolerances needed for customer production usage.
Using a GenAI-based orchestration engine (like our RIO) to facilitate automated orchestration of the models based on the context of the prompt makes a lot of sense – especially if you can use that engine to compare models during development and ongoing testing so that the orchestration of the models can actually evolve over time too. This would only be possible by using something like RIO instead of programmatically defining the model based on the application functionality.
MY TAKEAWAY
What do you see in the graphic below?
Here is what I see:
Best of Suite – In theory, end-to-end back-end platforms consisting of multiple Best of Breed front-end bolt-on ProcureTech solutions. (Think Mendocino Project, DUET, and Microsoft Office concept)
Best of Breed – Bolt-on systems transformed to front-end Apps (up to 75% of which will be gone by the end of 2025 via acquisition, absorption, or closing down)
Build Your Own Suite – The early stages of the Agent-based AI operating systems will be extended to interface with external 3rd Party Stakeholder Apps and platforms, e.g., ERP systems, suppliers, courier companies, and customs. (Watch this video for more details) – https://bit.ly/40fJxNY
NEXT POST IN THE SERIES: How do you integrate horizontal and vertical GenAI models seamlessly?
30
Here is the framework reflecting orchestration and intake from my patent document for my ProcureTech Solution in 1999 – at the time, it was called a Metaprise:
Read the above post again and then look at this image – From RAM (1999) Metaprise To RIO (2024) Orchestration and Intake. (NOTE: The Agent-Based Development and Implementation model occurs within the center hub.)
30
Share this:
Related