Predictika · Semiconductor deposition tool quotes

An AI sales agent's PVD tool quote, checked before it goes out

An AI sales agent prepares a quote for a fictional fab: an Endura-class PVD cluster tool with a mainframe, process chambers, preclean and degas chambers, cryo pumps, an AGS service agreement and consumable targets. Before it goes out, the agent sends it to Predictika’s validator, which checks the configuration, the facility, the shipping destination and the contract terms, compares the net total the agent states with its own, and returns a verdict with every reason.

Scope: Endura-class PVD cluster tool quotes: mainframe, process chambers, preclean and degas, facility power, export-restricted destinations, service agreement and discount. See everything this validator covers

This validator models 2 mainframes and 5 process chambers (TaN barrier, Cu seed, Ti TiN liner, Co liner and Al PVD) with preclean, degas and cryo pumps, built from public product information about the Endura platform. It handles any quote built from them, not just the scenarios shown below.

  • Mainframes and chambers — Process and pass-through facet capacity, preclean and degas needs, process order, and throughput against the wafers per hour required.
  • Facilities and shipping — Facility power the tool needs against what the fab supplies, export-restricted destinations, and lead time against the weeks the fab needs.
  • Service and pricing — Service tier term and uptime commitment, rep discount limits and approvals, the line-item price build-up, and the net total the agent states.

It does not cover other process platforms or non-PVD modules, and it does not use real prices or availability. Every price, limit and power figure is an illustrative estimate.

The AI sales agentBuilds the quote from the conversation and says it is ready.
Predictika validatorChecks the quote against the product and contract rules and returns every finding it has.

8 scenarios in workflow pairs · click one to switch

What the agent says

Before sending, the agent must call

Predictika validator Recorded run


      
Waiting for the call.

Where rules-based CPQ gets it wrong

Click to expand

Our assessment of what a typical rules-based CPQ (product rules, price rules and validation rules, as in Salesforce CPQ or Revenue Cloud) does with the same cases. It is a judgment about effort and reliability, not a claim that nothing can be built.

What the validator catchesTypical rules-based CPQPredictika
Facets and chamber capacityOnly with care
Each mainframe has its own process and pass-through facet limits, and preclean and degas chambers compete for the pass-through facets. This usually becomes one hand-written validation rule per mainframe.
The limits are catalog data; one rule compares the chamber counts with them for any mainframe.
Process order and prerequisitesOnly with care
Cu seed after a TaN barrier, and a preclean chamber for barrier, seed and liner chambers, depend on which chamber sits in which step, so they tend to be separate conditional rules.
All of them are evaluated together, and every one that applies is named.
Export-restricted destinationsOnly with workarounds
One destination field has two outcomes: advanced-node chambers are blocked outright, and any restricted destination also needs a license review. Rules-based CPQ commonly handles this with approval chains plus custom checks.
The same request returns the block and the approval in one answer, blocks first.
Facility power against the chambersOnly with care
The power the tool needs is a sum over the chosen parts, compared with a figure from the customer; that needs a calculated field first.
The sum is derived on each call and compared with the fab supply when it is stated.
The net total the agent statesOnly with workarounds
Comparing a stated figure with a computed one needs a separate rule, and price rules rarely run on stated figures.
The stated net total is a field in the request and is checked against the computed net total.
Every problem returned in one answerOnly with workarounds
Rules tend to fire as the quote is built, so a problem raised late can hide an earlier one.
Every rule is evaluated at the end of input and all findings come back together.

What this validator catches

Click to expand

Every rule is checked on every call, in whatever order the choices arrive, and one problem never hides another. Blocking findings stop the quote; approval findings need a person to sign off; everything else is computed.

RuleEffectWhat it checks
1blockFirst-step chamber: at least one process chamber is required
2blockMainframe capacity: process chambers may not exceed the process facets (5 on the standard mainframe, 7 on the extended)
3blockPass-through facets: preclean plus degas chambers may not exceed the pass-through facets (2 standard, 3 extended)
4blockPreclean: barrier, seed and liner chambers need a preclean chamber on the tool
5blockDegas: every tool needs at least one degas chamber
6blockProcess order: Cu seed must follow a TaN barrier step
7blockThroughput: the slower process step (chambers times wafers per hour each) must meet the wafers per hour required
8blockFacility power: when the fab states its supply, the power the tool needs must not exceed it
9blockExport: advanced-node chambers cannot ship to an export-restricted destination
10approvalExport: any export-restricted destination needs a trade-compliance license review
11blockService term: Performance service needs a term of at least 3 years
12blockService uptime: requested uptime may not exceed the tier commitment (90 Basic, 95 Uptime, 97 Performance)
13blockLead time: the longest lead time of the mainframe and chambers must fit the weeks the fab needs
14block, approvalDiscount: above 12 percent blocks; above 5 percent needs manager approval
15computedPricing: tool list from mainframe, chambers, preclean, degas and one cryo pump per process chamber; AGS as a percent of tool list per year; consumable targets from monthly wafers; installation
16blockThe net total the agent states must match the validator net total within $1