Great questions, Arun. I’ll answer wearing my Prism PM hat, not as a user, mostly to level set. I also really hope our users with hands on experience will add their perspectives.
Prism is not limited to Epicor’s out of the box tables. Prism’s metrics and list agents can work with UD fields, UD tables, and BAQs, and it is designed for customized environments. The Admin layer is a big part of this. Admins control which BAQs Prism indexes as embeddings, so you are explicitly choosing what data sources are available to Prism when it answers your questions. You can also customize those embeddings by editing how BAQs are described and by adding tags that reflect customer specific language and concepts. That context is used by hybrid search to help Prism identify which BAQ or business object should be used to answer a given question. Today Prism list and metrics agents answer questions from within a single BAQ or BO at a time but we are working on expanding that.
It is also worth calling out that Prism is not a single monolithic thing. There are different agents with different capabilities and boundaries that get called upon by Prism depending on your question. The Knowledge Agent provides how to answers grounded in Epicor help documentation. The ECM agent works with documents stored in ECM/DocStar to answer questions, summarize, and compare content. The Reasoning agent can analyze attachments of many types across multiple files, including orders, inventory, quotes, invoices, spreadsheets, and even dashboard screenshots to extract information or analyze performance. The Metrics and List agents are the ones that use customer specific BAQs to answer questions about your own data.
Prism is not autonomously training itself like a general purpose LLM. It is intentionally grounded and constrained. It is more than a smart query layer, but it still requires setup and refinement. Choosing which BAQs to surface and tuning the automatically generated BAQ summaries is important, so Prism has good signals for when to use which data to answer which questions. We recommend you roll out Prism to specific functional groups (purchasing, AR, etc.) and fine tune the BAQ selection/descriptions for that functional group, as opposed to turning on all BAQs in Prism for everyone.
Prism reasoning agent is particularly good. If you, for example, do a lot of evaluating document data from your customers, how many parts of this type etc. that you would normally count manually, etc. Prism will speed up your work. For those here who have Prism but have not used it lately, we have made a lot of changes, try asking it questions about attachments.
We ship Prism weekly and add new features and agents regularly. Customers who sign up also have recurring group calls where we capture feedback, troubleshoot, and explore new ideas. In one recent call, a customer asked about extracting invoice data from email attachments and updating Kinetic. We were able to prototype that quickly and are now refining it.
You should also be aware that Prism for App Studio is in controlled release. All Prism customers were invited to test the first version of creating ERP customizations directly from a prompt.
We are building Prism in three areas. Prism Answers (currently released), Prism Developer (Prism in App Studio, in controlled release) and Prism Tasks (doing tasks in Kinetic on your behalf, under development)
There is a lot Prism can do, but it does take some digging to understand the full surface area. Real world feedback from customers running it in heavily customized environments is especially valuable here. We’d love to hear from you.
that only plays safely with its standard Epicor schema?