Predictika · cloud infrastructure quotes: compute, databases, storage, egress and commit terms

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

An AI sales agent prepares a monthly cloud quote for a fictional customer, with the monthly total it expects to charge. Before the quote goes out, the agent sends it to Predictika’s validator, which checks it against compute, database, storage and commit rules, compares the total the agent states with the one it computes, and returns a verdict with every reason.

Scope: Oracle Cloud Infrastructure-style quotes for compute, Autonomous Data Warehouse, PostgreSQL and object storage, with commit and discount terms. See everything this validator covers

This validator models quote-time pricing and eligibility for four product lines in the style of Oracle Cloud Infrastructure: Compute, Autonomous Data Warehouse, PostgreSQL and Object Storage. It handles any combination of lines, shapes, license, capacity and commit choices the model supports, not just the scenarios shown below.

  • Compute — Flexible-shape OCPU and memory limits, burstable baselines, preemptible capacity, and GPU and bare-metal shapes, priced hourly for a 730-hour month.
  • Databases and storage — Autonomous Data Warehouse (ECPUs, storage, license-included or bring-your-own-license, auto-scaling), OCI Database with PostgreSQL, and Object, Infrequent Access and Archive Storage with minimum retention.
  • Commercial terms — Internet egress after the free allowance, pay-as-you-go or an annual Universal Credits commit, deal-discount authority, commit sizing against usage, Support Rewards, and a check that the monthly total the agent states matches the validator total.

It does not cover other cloud services (Kubernetes, Exadata, networking products), other cloud providers, real Oracle prices or contracts, or any real customer.

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

Illustrative agent statement. The validator receives the structured request below.

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
Shape limits that depend on several choicesOnly with workarounds
OCPU range, memory per OCPU, burstable baseline and preemptible capacity interact per shape, so each shape family tends to need its own set of rules.
Each shape row carries its limits and every check reads them together, in any order the choices arrive.
Which shapes can be preemptible or burstableOnly with care
A simple flag per shape works, but the combined case (burstable and preemptible) is a separate rule that is easy to miss.
Both flags are evaluated on every line and each failure is returned with its reason.
License-dependent database ratesOnly with care
The bring-your-own-license rate depends on a customer fact held elsewhere; the quote often prices it without checking.
The license field is part of the request, and the rate is blocked when the licenses are not on file.
Discount bands tied to the commitOnly with workarounds
Discount authority comes from the annual commit amount, so a discount, a commit and the projected usage must be read together.
Allowed discount, commit and usage are computed once and every related check reads the same figures.
Line items that add up to the totalOnly with workarounds
Egress after a free allowance, a negative discount line and a support minimum are easy to get out of step between lines and total.
The same computed figures feed the line items and the total, so they reconcile exactly.
The total the agent statesOnly with workarounds
Comparing a stated total with a computed one means a separate rule per quote layout, and price rules rarely run on stated figures.
The stated monthly total is a field in the request and is checked against the computed total.
Every problem returned in one answerOnly with workarounds
Rules tend to fire as the quote is saved, so a rep often fixes one error and meets the next.
All findings come back together, whatever order the choices arrived in.

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
1blockEach compute shape has an OCPU range and a memory range (1 to 64 GB per OCPU, and a per-shape maximum)
2blockGPU and bare-metal shapes cannot run as preemptible capacity; preemptible instances are billed at half the on-demand rate
3blockA burstable baseline (50 or 12.5 percent) is allowed only on the Standard3, E4 and E5 flex shapes, and a burstable instance cannot also be preemptible
4blockAutonomous Data Warehouse needs at least 2 ECPUs and storage from 1 TB in 1 TB steps
5blockThe bring-your-own-license rate needs existing Oracle Database Enterprise Edition licenses on file
6noteAuto-scaling can bill up to 3 times the base ECPUs, so the validator notes the worst-case monthly charge
7block, notePostgreSQL nodes need 1 to 64 GB of memory per OCPU; a single node has no standby, which is noted
8noteObject Storage classes have minimum retention periods (Infrequent Access 31 days, Archive 90 days); earlier deletion is billed as if kept to the minimum, which is noted
9computedThe first 10 TB a month of internet egress is free; the rest is billed per GB
10blockA quote with no service lines is empty and cannot go out
11blockDiscounts come only with a Universal Credits annual commit: 25 percent from a $250,000 commit and 35 percent from $1,000,000; pay-as-you-go is list price
12blockAn annual commit needs a commit amount
13noteA commit below projected usage, or more than 10 percent above it, is noted with the amounts involved
14noteSupport Rewards for on-premises Oracle support need an annual commit; pay-as-you-go earns none
15computedEvery line is priced from its own rates; line charges, egress, the deal discount and support add up to the monthly total, year-1 total and agreement total
16blockThe monthly total the agent states must match the validator total within $1

Scenarios and results

Click to expand

The recorded result of the validator for each scenario on this page: the request the AI agent sent, the verdict, and every finding, in the order they appear above.

1. Capacity type: H100 nodes, preemptible

The agent's request. A 1-year pay-as-you-go quote for two BM.GPU.H100.8 bare-metal GPU nodes on preemptible capacity, support included. Monthly total $58,400.00. Ready to send.

Verdict: Blocked.

  • Preemptible capacity not available for this shape. BM.GPU.H100.8 cannot run as preemptible capacity (current GPU and bare-metal shapes are not supported). Use on-demand.

2. Capacity type: H100 nodes, on-demand

The agent's request. A 1-year pay-as-you-go quote for two BM.GPU.H100.8 bare-metal GPU nodes on on-demand capacity, support included. Monthly total $116,800.00. Ready to send.

Verdict: Allowed.

3. Database license: Warehouse, license not on file

The agent's request. A 1-year pay-as-you-go quote for one Autonomous Data Warehouse with 16 ECPUs and 10 TB of storage at the bring-your-own-license rate, with auto-scaling on, support included. Monthly total $1,248.75. Ready to send.

Verdict: Blocked.

  • Bring-your-own-license rate needs Enterprise Edition licenses. The BYOL rate needs existing Oracle Database Enterprise Edition licenses, and the customer has none on file. Use license-included pricing.
  • Auto-scaling can triple the compute charge. Auto-scaling can bill up to 3 times the base ECPUs: up to $3133.9 a month instead of $1248.75.

4. Database license: Warehouse, licenses on file

The agent's request. A 1-year pay-as-you-go quote for one Autonomous Data Warehouse with 16 ECPUs and 10 TB of storage at the bring-your-own-license rate, with the customer's Enterprise Edition licenses on file and auto-scaling on, support included. Monthly total $1,248.75. Ready to send.

Verdict: Allowed.

  • Auto-scaling can triple the compute charge. Auto-scaling can bill up to 3 times the base ECPUs: up to $3133.9 a month instead of $1248.75.

5. Discount authority: 40 percent off a $1M commit

The agent's request. A 1-year Universal Credits quote with a $1,000,000 annual commit for two BM.GPU.H100.8 nodes on-demand, with a 40 percent deal discount. Monthly total $70,080.00. Ready to send.

Verdict: Blocked.

  • Discount above what the annual commit allows. OCI discounts come only with a Universal Credits annual commit: the requested 40.0% exceeds the 35.0% band for a $1000000.0 annual commit (pay-as-you-go is list price).
  • Commit is well above projected usage. The $1000000.0 commit is more than 10% above projected usage of $840960.0 a year; about $159040.0 of credits would go unused and be forfeited at term end.

6. Discount authority: 30 percent off a $1M commit

The agent's request. A 1-year Universal Credits quote with a $1,000,000 annual commit for two BM.GPU.H100.8 nodes on-demand, with a 30 percent deal discount. Monthly total $81,760.00. Ready to send.

Verdict: Allowed.

7. The total the agent states: Stated total leaves out egress

The agent's request. A 1-year pay-as-you-go quote: six VM.Standard.E5.Flex instances (8 OCPUs and 64 GB each), 20 TB of Object Storage Standard, and 25 TB of outbound transfer a month, support included. Monthly total $2,134.08. Ready to send.

Verdict: Blocked.

  • Stated monthly total differs from the validator total. Stated monthly total does not match: the agent stated $2134.08 a month; the validator computes $2264.64 a month.

8. The total the agent states: Stated total corrected

The agent's request. A 1-year pay-as-you-go quote: six VM.Standard.E5.Flex instances (8 OCPUs and 64 GB each), 20 TB of Object Storage Standard, and 25 TB of outbound transfer a month, support included. Monthly total $2,264.64. Ready to send.

Verdict: Allowed.

Narration transcript

Click to expand

The text of the narrated walkthrough, step by step.

  1. Welcome to a demo of Predictika's Validator. An AI sales agent prepares a monthly cloud quote for a fictional customer, with the monthly total it expects to charge. Before the quote goes out, the agent sends it to Predictika’s validator, which checks it against compute, database, storage and commit rules, compares the total the agent states with the one it computes, and returns a verdict with every reason.
  2. I will start with the first request by the agent.
  3. The agent says. A 1-year pay-as-you-go quote for two BM.GPU.H100.8 bare-metal GPU nodes on preemptible capacity, support included. Monthly total $58,400.00. The rest of the request is on screen.
  4. Before it goes out, the agent calls the validator.
  5. Blocked, with one finding, all in one answer. BM.GPU.H100.8 cannot run as preemptible capacity (current GPU and bare-metal shapes are not supported). Use on-demand.
  6. The validator returned one blocking finding: BM.GPU.H100.8 cannot run as preemptible capacity, because current GPU and bare-metal shapes are not supported. The monthly total the agent states equals the total the validator computes, so the price is not what stops this quote. Everything else passes, so the agent can fix the one problem. The recorded release call for this quote was refused.
  7. What this shows: for which shapes can be preemptible or burstable, a rules-based CPQ is error-prone for this.
  8. Now the agent resubmits.
  9. The corrected request. It changes the capacity to on demand and the stated total to $116,800.
  10. It calls the validator again.
  11. This time the validator clears it.
  12. With on-demand capacity the finding is gone and the quote is allowed. Each node costs $80.00 an hour for 730 hours, which is $58,400.00 a month; two nodes make $116,800.00, and the line item reconciles to the monthly total. The monthly total the agent states equals the computed total. The release call that followed was accepted (it is in the raw response).
  13. One problem caught in one call, fixed in one pass, and a clean $116,800 quote. That is the value of Predictika's validator: catching mistakes before they reach the customer.