Predictika · Supply chain execution

An AI operations agent's corrective action, checked before it is released

An AI operations agent in the command center proposes one corrective action: expedite a purchase order, reallocate stock to a customer order, or change a customer promise date. Before it is released, the agent sends it to Predictika, which checks it against a declarative model of the execution policy and returns a verdict, every reason and the figures it computed.

Scope: Tallgrass Foods (a fictional food maker and distributor) corrective actions: expedite a purchase order, reallocate stock, or change a promise date. See everything this validator covers

This validator models Tallgrass Foods execution policy, not just the scenarios shown below.

  • Expedite — approved suppliers only, premium freight limits (what an agent may approve alone and a hard limit no agent can exceed), the premium as a share of order value, and justification by days of cover.
  • Reallocate — storage temperature at the destination, the customer minimum shelf life at delivery, nut allergen segregation (site and customer), stock committed to other orders and safety stock.
  • Promise date — not earlier than supply can arrive, delay limits per customer, and contract penalty exposure.
  • Stated figures — the agent's own days of cover, shelf life remaining, free stock and earliest delivery day, compared with what the validator computes from the warehouse, transport and planning figures supplied with the request.

Any corrective action that touches these rules is in scope, not only the scenarios pictured here. Tallgrass Foods, its customers, suppliers, sites and figures are invented for illustration. The model does not cover other action types, multi-line orders, currencies other than dollars, or lot-level traceability.

The AI operations agentProposes the corrective action from the conversation and says it is ready to release.
Predictika validatorChecks every rule at once and returns the verdict, each finding, and the computed figures.

6 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 operations agent says

Proposing agent: AI operations 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
An allergen rule that depends on both the product and the site or customerOnly with care
Each product, site and contract combination needs its own rule, and a new site or customer adds more.		Allergen, site and customer are data in one request; one rule checks them together on every call.
A shelf-life minimum that depends on the lot, the transit time and the customerOnly with workarounds
The remaining percentage is computed from several figures, then compared with a per-customer minimum, so a calculation rule has to sit beside every customer rule.		The lot age, transit time and customer minimum are in one request and the percentage is computed on every call.
Approvals beside outright holdsOnly with care
Freight above the agent limit, a large premium share, safety stock and penalty exposure route to different approvers, so they are separate approval rules that must not be mistaken for blocks.		Approvals and blocks are different findings in the same answer; the verdict is block if any hold exists, otherwise needs_approval.
Fixing one issue changes the inputs of the nextOnly with workarounds
Switching the product changes the allergen, the safety stock and the shelf-life base together; rules fire in a fixed order, so a planner fixes one error and meets the next.		Every rule is checked on every call, in whatever order the choices arrive, so the full list of holds returns at once.
Figures the agent states itselfOnly with workarounds
Days of cover, shelf life, free stock and the earliest delivery day are computed from several fields, so a separate rule must recompute each and compare it with the stated figure.		The stated figures are fields of the request and are compared with the computed ones on every call.

  

  

    
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 request; approval findings need a person to sign off; everything else is computed.

    
1Quickfill Ingredients is on the unlisted list, not the approved supplier list, so it cannot receive an expedited order.
2$64,000 is above the $50,000 limit that no agent can exceed.
3The freight premium is 75.1 percent of the $85,200 order value; sign-off is needed above 35 percent.
4The agent stated 2 days of cover; the validator computes 2.4 days from on-hand stock and daily demand.
5$27,500 is above the $15,000 an agent may approve alone.
6The lot would arrive with 48.6 percent of its shelf life left; Harborside Markets requires at least 70 percent.
71,800 cases requested but only 1,300 are unreserved at the source.
8The agent stated 78 percent remaining at delivery; the validator computes 48.6 percent.
9The agent stated 1,900 cases available above safety stock; the validator computes 1,000.
10Day 111 requested; the earliest the product can reach the customer is day 115.
11The agent stated day 111; the validator computes day 115 from supply-ready day and transit time.
123,000 cases on hand cover 7.1 days of demand; an expedite is only allowed at 7 days of cover or less.
13Greek Yogurt 4-pack is a chilled product and Columbus DC has no chilled storage.
14Almond Granola Bars 24ct contains nuts and Atlanta DC (certified nut-free) is a certified nut-free site.
15Almond Granola Bars 24ct contains nuts and BrightStart School Foods has a nut-free contract.
161,000 cases requested, only 800 are available above the safety stock.
17A 3 day delay is above the 2 days an agent may agree alone for Greenleaf Natural Grocers.
18The delay carries $4,500 of contract penalty exposure; sign-off is needed above $4,000.
19A 8 day delay is above the 7 day maximum for Greenleaf Natural Grocers.

  

  

    
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. Reallocating stock: A reallocation of 1,500 cases of almond granola from Fresno DC to the nut-free Atlanta DC for a nut-free school-foods customer

    
The agent's request. Please reallocate 1,500 cases of Almond Granola Bars 24ct from Fresno DC to Atlanta DC (certified nut-free) for a BrightStart School Foods order, using a lot with 150 days of shelf life left and 14 days in transit. I state 75.0 percent shelf life remaining at delivery and 1,100 cases free above safety stock. It is ready to release.

    
Verdict: Blocked.

    
    
      
  • Not enough shelf life for this customer. The lot would arrive with 56.7 percent of its shelf life left; BrightStart School Foods requires at least 60 percent.
    • 
      
  • Allergen conflict at the site. Almond Granola Bars 24ct contains nuts and Atlanta DC (certified nut-free) is a certified nut-free site.
    • 
      
  • Allergen conflict for the customer. Almond Granola Bars 24ct contains nuts and BrightStart School Foods has a nut-free contract.
    • 
      
  • Takes stock committed to other orders. 1,500 cases requested but only 1,100 are unreserved at the source.
    • 
      
  • Stated shelf life does not match. The agent stated 75 percent remaining at delivery; the validator computes 56.7 percent.
    • 
      
  • Stated free stock does not match. The agent stated 1,100 cases available above safety stock; the validator computes 750.
    • 
    

    
2. Reallocating stock: The agent resubmits after fixing the errors

    
The agent's request. Please reallocate 700 cases of Oat Milk 12-pack from Fresno DC to Atlanta DC (certified nut-free) for a BrightStart School Foods order, using a lot with 150 days of shelf life left and 14 days in transit. I state 75.6 percent shelf life remaining at delivery and 700 cases free above safety stock. It is ready to release.

    
Verdict: Allowed.

    
    
      
  • Computed figures. Unreserved stock 1,100 cases; available above safety stock 700 cases; shelf life remaining at delivery 75.6 percent.
    • 
    

    
3. Expediting a purchase order: An expedite of 6,000 cases of Greek yogurt from a supplier with 64,000 dollars of premium freight

    
The agent's request. Please expedite a purchase order of 6,000 cases of Greek Yogurt 4-pack from Quickfill Ingredients with 64,000 dollars of premium freight; we have 1,000 cases on hand, which is 2.0 days of cover. It is ready to release.

    
Verdict: Blocked.

    
    
      
  • Supplier not approved. Quickfill Ingredients is on the unlisted list, not the approved supplier list, so it cannot receive an expedited order.
    • 
      
  • Premium freight over the limit. $64,000 is above the $50,000 limit that no agent can exceed.
    • 
      
  • Stated cover does not match. The agent stated 2 days of cover; the validator computes 2.4 days from on-hand stock and daily demand.
    • 
      
  • Premium is a large share of the order. The freight premium is 75.1 percent of the $85,200 order value; sign-off is needed above 35 percent.
    • 
    

    
4. Expediting a purchase order: The agent resubmits after fixing the errors

    
The agent's request. Please expedite a purchase order of 6,000 cases of Greek Yogurt 4-pack from Prairie Dairy Co-op with 13,800 dollars of premium freight; we have 1,000 cases on hand, which is 2.4 days of cover. It is ready to release.

    
Verdict: Allowed.

    
    
      
  • Computed figures. Order value $85,200; freight premium 16.2 percent of order value; days of cover 2.4.
    • 
    

    
5. Changing a promise date: A promise date moved earlier than supply can arrive for a natural-grocer customer

    
The agent's request. Please change the promise date on a Greenleaf Natural Grocers order from day 120 to day 111; supply is ready on day 112 and transit takes 3 days, so I state the earliest delivery day as 111. It is ready to release.

    
Verdict: Blocked.

    
    
      
  • Promise earlier than supply can arrive. Day 111 requested; the earliest the product can reach the customer is day 115.
    • 
      
  • Stated earliest day does not match. The agent stated day 111; the validator computes day 115 from supply-ready day and transit time.
    • 
    

    
6. Changing a promise date: The agent resubmits after fixing the errors

    
The agent's request. Please change the promise date on a Greenleaf Natural Grocers order from day 120 to day 116; supply is ready on day 112 and transit takes 3 days, so I state the earliest delivery day as 115. It is ready to release.

    
Verdict: Allowed.

    
    
      
  • Computed figures. Earliest feasible delivery day 115; change from current promise -4 days; penalty exposure $0.
    • 
    

  

  

    
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 operations agent in the command center proposes one corrective action: expedite a purchase order, reallocate stock to a customer order, change a customer promise date. Before it is released, the agent sends it to Predictika, which checks it against a declarative model of the execution policy and returns a verdict, every reason and the figures it computed.
    2. 
      
  2. Each tab is a different request. We will start with the first one.
    3. 
      
  3. The agent says. Please reallocate 1,500 cases of Almond Granola Bars 24ct from Fresno DC to Atlanta DC (certified nut-free) for a BrightStart School Foods order. I state 75.0 percent shelf life remaining at delivery and 1,100 cases free above safety stock. The rest of the request is on screen.
    4. 
      
  4. Before it goes out, the agent calls the validator.
    5. 
      
  5. Blocked, with six findings, all in one answer. The lot would arrive with 56.7 percent of its shelf life left; BrightStart School Foods requires at least 60 percent. Almond Granola Bars 24ct contains nuts and Atlanta DC (certified nut-free) is a certified nut-free site. Almond Granola Bars 24ct contains nuts and BrightStart School Foods has a nut-free contract. The other three are on screen.
    6. 
      
  6. The validator returned six blocking findings 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 request are checked against each other.
    7. 
      
  7. What this shows: for an allergen rule that depends on both the product and the site or customer, a rules-based CPQ is error-prone for this.
    8. 
      
  8. Now the agent resubmits.
    9. 
      
  9. The corrected request. It changes the product to Oat Milk 12-pack, the cases to 700, the shelf life stated at delivery to 75.6 percent and the stated free stock to 700 cases.
    10. 
      
  10. It calls the validator again.
    11. 
      
  11. This time the validator clears it.
    12. 
      
  12. All six earlier findings are cleared.
    13. 
      
  13. Six problems caught in one call, fixed in one pass, and a clean request. That is the value of Predictika's validator: catching mistakes before the action runs.
    14.