Predictika · AI infrastructure

An AI sales agent's AI gateway plan, checked before it goes out

An AI sales agent prepares an order for an AI gateway (the proxy that sits between a company's applications and the language model providers): the plan, data region, single sign-on, PII redaction, semantic cache, guardrails, audit log retention, private networking and model access, together with the request rate, token volume and admin seats the customer needs and the rate limit the agent states. Before it goes out, the agent sends it to Predictika, which checks it against a declarative model of typical gateway plan rules and returns a verdict, every reason and the figures it computed.

Scope: AI gateway plan and policy configuration (a generic, illustrative model of the product class): four plans, three data regions, features, retention, private networking, model access, request rate and token volume. See everything this validator covers

This validator models one AI gateway plan order at a time, based on plan limits that are typical of AI gateway products in general. It is a generic example for the product class, not any vendor's plans, and it handles any order built from these options, not just the scenarios shown below.

  • Plans — Starter, Team, Business and Enterprise, each with a request rate limit, admin seat limit, included monthly token volume, audit retention and feature set (illustrative values).
  • Options — data region (US, EU, APAC), single sign-on, PII redaction, semantic cache, guardrails (none, basic, custom), audit log retention (30, 90 or 365 days), private network connectivity, and routing to public model APIs, private models or both.
  • Checks — which plan offers which region and feature, private models needing private networking, EU traffic to public model APIs needing PII redaction, seats, retention and request rate against the plan, and the stated rate limit against the plan limit. A peak above 80 percent of the rate limit, or a token volume above the included amount, goes to a person.

Any gateway order built from these options is in scope — not only the scenarios pictured here. Not covered: prices (none are modeled, so this page shows no price build-up), any particular vendor's real plan names and limits, model-by-model routing rules, per-key budgets, rate limits by team, and on-premises deployment.

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

7 scenarios in 3 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

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
Which plan offers which region and feature, and which features need a higher planOnly with workarounds
Each plan-by-feature combination is a separate rule, and they have to be kept in step as plans change.		Plan, region and features are compared together, in any order.
Privately hosted models only over private networking, and EU traffic to public model APIs only with PII redactionOnly with care
These are cross-field rules that depend on the model access, network and region choices together.		The combined choices are checked on every call.
Seats, retention and request rate against the plan limit, which differ by planOnly with care
It takes a lookup of limits per plan and a comparison for each, and a rep can pick the plan last.		The limits for the chosen plan are looked up and compared on every call.
A quote that states a rate limit the plan does not giveNot caught
A stated figure is free text to a CPQ, so nothing compares it with the plan data.		The stated figure is compared with the computed one.
Every problem returned in one answerOnly with workarounds
Rules tend to fire as the order is saved or a step is completed, so a person often fixes one error and meets the next.		All findings come back together, whatever order the choices arrived in.

  

  

    
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 has 6 admin seats; the chosen plan allows 5 at most.
2The quote has 26 admin seats; the chosen plan allows 25 at most.
3The chosen plan does not offer the selected data residency region.
4With the EU data region, traffic to public model APIs must have PII redaction switched on.
5Single sign-on is not included in the Starter plan.
6PII redaction is not included in the Starter plan.
7The semantic cache is not included in the Starter plan.
8Custom guardrail policies need the Business or Enterprise plan.
9The quote asks for 365 days of audit log retention; the chosen plan keeps 90 days at most.
10The quote asks for 90 days of audit log retention; the chosen plan keeps 30 days at most.
11Private network connectivity needs the Business or Enterprise plan.
12Routing to privately hosted models needs the Business or Enterprise plan.
13Privately hosted models are reached only over private network connectivity.
14The expected peak of 2,401 requests per minute is over 80 percent of the plan limit of 3,000, so a person should confirm the burst headroom.
15The expected peak of 3,000 requests per minute is over 80 percent of the plan limit of 3,000, so a person should confirm the burst headroom.
16The customer needs 3,001 requests per minute; the chosen plan allows 3,000 at most.
17The expected 501 million tokens a month is over the 500 million included in the plan, so a person should agree the overage terms.
18The expected peak of 2,900 requests per minute is over 80 percent of the plan limit of 3,000, so a person should confirm the burst headroom.
19The expected 900 million tokens a month is over the 500 million included in the plan, so a person should agree the overage terms.
20The quote states a limit of 5,000 requests per minute; the chosen plan allows 3,000.