Why this memo.
Epicor's Prism AI roadmap is real and the team behind it is competent. The mid-market specialty shop running Kinetic with two senior estimators and a $34M annual run-rate is not the average Epicor customer. The leverage available to that shop is in custom CPQ AI built on top of Kinetic, calibrated to the shop's part library, the shop's pricing rules, the shop's customer history.
This memo is the architecture. Below: the Kinetic surface area we touch, the workflows we ship, and what we don't build.
The Kinetic surface area we touch.
Epicor Kinetic exposes the Epicor REST API and the Kinetic Functions framework, both authenticated and stable. Reads/writes against Quote, Job, Part, Customer, Operation, Resource Group, and the Engineering Workbench. For the AI layer, we read against the BAQ (Business Activity Query) layer, which is faster than table-by-table reads.
For shops with strict data-residency requirements, we deploy entirely inside the shop's Azure or AWS tenant. Epicor data does not leave the shop's environment.
Workflow I: Custom CPQ AI on the shop's actual part library.
The five-hour-quote-becomes-eleven-minutes workflow. The senior estimator at a $34M custom metals shop spends 30 minutes on the actual judgment call and 5 hours on parsing the RFQ, looking up part history, pulling current material costs, checking machine capacity, and formatting the response.
The custom AI version: reads the inbound RFQ (PDF, email, customer-portal upload), extracts part specs, looks up matching prior jobs in Kinetic, validates capability against current Resource Group state, drafts pricing on the shop's actual rules, formats the response in the customer's expected format. Senior estimator validates the judgment call, signs off, sends.
Recovery: median customer goes from 6-hour quote turnaround to 11-25 minutes. Win rate uplift on speed alone runs 18-25%.
Workflow II: Spec parsing and capability validation.
The "we can't do that" detection workflow. Every shop has capability constraints (tolerance, material, finishing process, machine envelope) that an inexperienced estimator may not catch on the first pass. The custom AI cross-references the RFQ specs against the shop's actual capability matrix in Kinetic.
Catches "we can't hold that tolerance on Mazak 5" before the quote goes out, not after the order arrives. Recovery range: 4-9% lift in win rate from spec-mismatch reductions.
Workflow III: RFQ triage and walk-away recovery.
Most shops walk away from 15-30% of inbound RFQs because the estimator can't get to them. The custom AI triages the inbound, drafts a fast no with a referral or a delayed-quote offer, captures the relationship for a future RFQ, surfaces the high-priority RFQs to the senior estimator first.
Revenue uplift on the walkaways alone: 8-15% in audited shops.
Workflow IV: Estimator-archive RAG.
Twenty years of quotes, part histories, customer-specific rules, and "the way we price that family of jobs." Custom retrieval over the Kinetic + document archive turns a junior estimator into a senior contributor for retrieval-heavy quoting.
What we don't build.
We do not replace Epicor Kinetic. We do not build a competitor to Epicor Prism. We do not build a generic chatbot the floor queries; the leverage is in CPQ workflow integration. If the shop wants Epicor's roadmap configured well, Epicor Pro Services is the right answer. If the shop wants the workflows above, on top of Kinetic, ours is.