Predictika · Google Cloud quote-time pricing

An AI sales agent's Google Cloud quote, checked before the customer sees it

An AI agent drafts a Google Cloud quote: VM fleets and commitments. Before the quote goes out, the agent sends it to Predictika, which prices it from a declarative model of Google's own quote-time rules, checks the total the quote states against the computed one, and returns every finding together with the line-item quote.

Scope: Google Cloud compute quote-time pricing: VM fleets, committed-use discounts (resource-based and Flex), volume and discount schedules, percent-of-total lines, and GPU lines. See everything this validator covers

This model covers Google Cloud quote-time compute pricing — VM fleet costs, committed-use discounts and volume or tier discount schedules — not the whole GCP catalog:

  • Committed-use discounts (CUDs) — resource-based CUDs scoped to committed vCPU only, with the remainder priced at on demand; 1-year and 3-year rates by machine family; and Flex CUD spend-based commitments with a sizing advisory against other planned spend.
  • Volume and discount-schedule rules — shared and family-specific volume tiers, range versus slab discount-schedule math, and percent-of-total lines that recompute when a quote is amended. These were also run on this model, but are not shown on this page.
  • Rates missing from the catalog, and GPU pricing — accelerator-optimized families and attached GPUs whose per-family or per-GPU CUD rate is not loaded in the catalog, which the validator leaves unpriced and flags for the account team rather than estimating.

Within this scope the validator handles any combination of VM family, term, commitment size and discount schedule these compute quotes can produce, not just the scenarios shown here. It does not cover GKE, BigQuery, networking or other GCP services, or non-GCP cloud providers.

The AI sales agentBuilds the Google Cloud quote from the conversation and says it is ready.
Predictika validatorChecks every rule at once and returns the verdict, each finding, and the computed figures.

6 scenarios in 4 groups · each group runs from the agent’s first attempt to the outcome · click one to switch

What the agent proposes

The AI sales agent says

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

Proposing agent: AI sales agent (scripted)

Before sending, the agent must call Recorded run

Predictika validator


      

        
        
        
      

      
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) does with the same cases. It is a judgment about effort and reliability, not a claim that nothing can be built.

    
What the validator caughtTypical rules-based CPQPredictika
A commitment smaller than the fleetOnly with care
Split the fleet into committed and uncommitted vCPU and price each part. It takes a chain of price rules, and a missed step discounts vCPU the customer never committed.		The split is part of the model: covered vCPU at the CUD rate, the rest at list.
A stated total that does not match the computed priceOnly with care
A validation rule can compare a stated total with a calculated field, but only if the stated total is a field on the quote, which it rarely is.		The stated total travels with the quote and is compared with the computed one on every call.
The right Flex rate for the right termOnly with care
A rate table by term; a quote that borrows the 3-year rate for a 1-year commitment looks plausible and nothing flags it.		The rate comes from the catalog by term, and the total check catches a borrowed rate.
A rate that is not in the catalogOnly with workarounds
Typically falls back to a default or to zero, so a price appears that nobody can support.		The line is left unpriced, the missing rate is named, and the quote goes to a person.
Commitment sizing against other planned spendOnly with workarounds
A judgment, not a hard rule; usually a note in a runbook.		Raised as a warning on the quote, and a person confirms the sizing.

  

  

    
What this validator catches: findings from its test cases
Click to expand
    
Every rule is checked on every call, in any order the choices arrive; one problem never hides another. Blocking findings stop the quote; approval findings need a person to sign off; everything else is computed.

    
1the quote states a total of $3,125.94; the validator computes $3,650.47.
2the quote states a total of $7,884.00; the validator computes $10,512.00.
3resource CUD rate for machine family A3 (3-year term) is not in the pricing catalog; the validator leaves it unpriced and does not estimate.
4GPU CUD rate for H100 is not in the pricing catalog; the validator leaves it unpriced and does not estimate.
5sizing check: a new Flex CUD should reflect only other planned spend, not vCPUs already committed on the same quote.

  

  

    
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. Commitment too small: Commitment smaller than the fleet

    
The agent's request. Please quote 35 n2-standard-4 VMs in us-central1 with a 1-year commitment on 100 vCPU. The commitment gives 37 percent off, so the fleet comes to $3,125.94 a month. The quote states a total of $3,125.94. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Quoted total does not match. The quote states a total of $3,125.94; the validator computes $3,650.47.
    • 
      
  • Scope reminder. The line cost covers compute (and GPU, where attached) only: persistent disk, snapshots, and egress for this same fleet are priced separately and are never discounted by a resource-based or Flex CUD.
    • 
    

    
2. Commitment too small: The agent resubmits after fixing the errors

    
The agent's request. Please quote 35 n2-standard-4 VMs in us-central1 with a 1-year commitment on 100 vCPU. The commitment covers 100 vCPU at 37 percent off and the other 40 vCPU are at the list price. The quote states a total of $3,650.47 a month. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Scope reminder. The line cost covers compute (and GPU, where attached) only: persistent disk, snapshots, and egress for this same fleet are priced separately and are never discounted by a resource-based or Flex CUD.
    • 
      
  • Quote summary. Total $3,650.47 a month: committed use $2,232.81; on demand $1,417.66.
    • 
    

    
3. Flex commitment rate: Flex commitment at the wrong rate

    
The agent's request. Please quote a 1-year Flex commitment of $20 per hour in us-central1, at 46 percent off. The quote states a total of $7,884.00 a month. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Quoted total does not match. The quote states a total of $7,884.00; the validator computes $10,512.00.
    • 
    

    
4. Flex commitment rate: The agent resubmits after fixing the errors

    
The agent's request. Please quote a 1-year Flex commitment of $20 per hour in us-central1, at 28 percent off. The quote states a total of $10,512.00 a month. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Quote summary. Total $10,512.00 a month: flex commitment $10,512.00.
    • 
    

    
5. Rate not in the catalog: GPU machine with no published rate

    
The agent's request. Please quote one a3-highgpu-8g in us-central1 with a 3-year commitment on all 208 vCPU, GPUs included. It is ready to send.

    
Verdict: Needs approval.

    
    
      
  • Committed-use rate not in the catalog. Resource CUD rate for machine family A3 (3-year term) is not in the pricing catalog: load Google's published committed-use price before quoting; the committed-use cost intentionally left unset, not estimated.
    • 
      
  • GPU committed-use rate not in the catalog. GPU CUD rate for H100 is not in the pricing catalog: load Google's published committed-use price before quoting; the GPU cost intentionally left unset, not estimated.
    • 
    

    
6. Layered commitment: Commitment on top of a commitment

    
The agent's request. Please quote 25 n2-standard-4 VMs with an existing 1-year commitment on 80 vCPU, plus a new 3-year Flex commitment of $40 per hour. The quote states a total of $18,263.08 a month. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Flex commitment may be oversized. Sizing check: this quote's new Flex CUD spend must reflect only other planned spend, not vCPUs already committed on this same quote's compute lines: verify the committed hourly spend was not double-counted against those lines before finalizing.
    • 
      
  • Scope reminder. The line cost covers compute (and GPU, where attached) only: persistent disk, snapshots, and egress for this same fleet are priced separately and are never discounted by a resource-based or Flex CUD.
    • 
      
  • Quote summary. Total $18,263.08 a month: flex commitment $15,768.00; committed use $1,786.25; on demand $708.83.
    • 
    

  

  

    
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 agent drafts a Google Cloud quote: VM fleets and commitments. Before the quote goes out, the agent sends it to Predictika, which prices it from a declarative model of Google's own quote-time rules, checks the total the quote states against the computed one, and returns every finding together with the line-item quote.
    2. 
      
  2. I will start with the first request by the agent.
    3. 
      
  3. The agent says. Please quote 35 n2-standard-4 VMs in us-central1 with a 1-year commitment on 100 vCPU. The commitment gives 37 percent off, so the fleet comes to $3,125.94 a month. The rest of the request is on screen. The quote states a total of $3,125.94.
    4. 
      
  4. Before it goes out, the agent calls the validator.
    5. 
      
  5. Blocked, with one finding, all in one answer. The quote states a total of $3,125.94; the validator computes $3,650.47.
    6. 
      
  6. The validator returned one blocking finding in one answer. A check that reads one field at a time would pass most of this; it shows only when the choices in the quote are checked against each other and against the stated total.
    7. 
      
  7. What this shows: for a stated total that does not match the computed price, a rules-based CPQ is error-prone for this.
    8. 
      
  8. Now the agent resubmits.
    9. 
      
  9. The corrected request. It changes the quoted total to $3,650.47.
    10. 
      
  10. It calls the validator again.
    11. 
      
  11. This time the validator clears it.
    12. 
      
  12. The validator also returns the line-item quote, on screen.
    13. 
      
  13. The earlier finding is cleared, and the validator's total, $3,650.47, matches the quote.
    14. 
      
  14. One problem caught in one call, fixed in one pass, and a clean $3,650.47 quote. That is the value of Predictika's validator: catching mistakes before they reach the customer. Now you can try the other scenarios yourself.
    15.