Predictika · Storage-software license deals
An AI deal desk agent prepares a storage-software license deal for Tessellate Data, in dollars: new terms, expansions, pay-per-gig, flash reclaim and tenant commitments. Before it goes out, the agent sends it to Predictika, which checks it against a declarative model of the deal rules, compares the annual value the deal states with the computed one, and returns every finding. Tessellate Data is a fictional company and every figure is invented for illustration.
This validator models one slice of the Tessellate Data deal desk, not just the scenarios shown below.
Any deal that touches these rules is in scope, not only the scenarios pictured here. Tessellate Data is fictional and every price, discount, threshold and customer is invented for illustration; the model does not cover support contracts, hardware, services, other currencies or real vendor prices.
7 scenarios in 3 groups · each group runs from the agent’s first attempt to the outcome · click one to switch
Illustrative agent statement. The validator receives the structured request below.
Proposing agent: AI deal desk agent (scripted)
Before sending, the agent must call Recorded run
What this means
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 caught | Typical rules-based CPQ | Predictika |
|---|---|---|
| Discount floor that depends on term and volume | Only with workarounds The floor moves with two discount tables and the exception band, so a rules-based tool needs a price table for each term and volume tier. | The floor is computed from the term and volume of each deal, and the finding states the discounts it applied. |
| Expansion that must co-term | Only with care Whether an expansion is valid depends on the months left on the customer's existing agreement, which is data outside the deal. | The customer's agreement is part of the model, so the co-term check uses the real months remaining. |
| Fixing one issue changes the inputs of the next | Only with workarounds Moving an expansion to the agreement end changes its term discount and so its price floor, and a new price changes the annual value the agent stated; rules fire in a fixed order, so a rep 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 and the next call shows what the fix changed. |
| Data-reduction guarantees against proven ratios | Only with workarounds The risk of a guarantee depends on the workload, the capacity and the gap to the proven ratio, and the limit between approval and block is a threshold on a computed figure. | The free capacity at risk is computed per deal and compared with the tolerance, so a small gap needs approval and a large one is blocked. |
| Customer concentration | Only with care Share of company revenue after the deal needs the customer's current revenue and the company total, both outside the deal. | Both figures are in the model, and the share after the deal is computed every time. |
| Tenant promises against effective capacity | Only with workarounds Effective capacity depends on the workload's proven ratio and what the customer has already promised. | Licensed capacity, ratio and existing promises are combined in one check, and the finding shows all three. |
| Annual value the agent states | Only with workarounds A rule can compare a stated figure with a computed one only if the stated figure is a field on the deal, which it rarely is. | The stated annual value travels with the deal and is compared with the computed value on every call; a mismatch is its own block. |
Every rule is checked on every call, in any order the choices arrive; one problem never hides another. Blocking findings stop the deal; approval findings need a person to sign off; everything else is computed.
| 1 | $17.0 per TB-month is below the $17.81 floor for this term and volume (12.0 percent term and 8.0 percent volume discount off $22.0 list). |
| 2 | $17.2 per TB-month is below the $17.81 floor but within the exception band. Needs a pricing exception. |
| 3 | After this deal Aurora Grid would be 0.2966 of company CARR, above the concentration limit. Needs CFO approval. |
| 4 | An expansion must end with the existing agreement, which has 44 months left; this one runs 60 months. |
| 5 | Solstice AI has no agreement to expand. Write a new-term license instead. |
| 6 | Pay-per-gig terms run at least 3 months; this one is 2. |
| 7 | Data-reduction guarantees are not offered on pay-per-gig terms. |
| 8 | Promising 1.33:1 on Model checkpoints (proven 1.3:1) risks 2.31 PB of free true-up capacity. Within tolerance; needs approval. |
| 9 | Promising 2.0:1 on Model checkpoints (proven 1.3:1) means up to 53.85 PB of free true-up capacity, above the tolerance. |
| 10 | Flash reclaim needs supported drives shipped in 2,021 or later; the drives named do not qualify. |
| 11 | Reclaimed capacity is priced at the locked $14.0 per TB-month; $13.5 is below it. |
| 12 | Pinecrest Compute has 120.0 PB licensed, which is 156.0 PB effective for Model checkpoints. Tenant promises would reach 190.0 PB. A license expansion is needed first. |
| 13 | Harbor Retail is not in the neocloud program, so it cannot resell capacity to tenants. |
| 14 | The stated annual value of $1.06M does not match the validator value of $1.32M. |
| 15 | A 72-month term is longer than the 60-month maximum. |
| 16 | Name the workload, so the promise can be checked against proven data reduction. |
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.
The agent's request. Please write a 2-month pay-per-gig term for Pinecrest Compute of 10 PB at $24.00 per TB-month, promising a 2:1 data reduction on model checkpoints, stating an annual value of $3.00 million. It is ready to send.
Verdict: Blocked.
The agent's request. Please write a 3-month pay-per-gig term for Pinecrest Compute of 10 PB at $27.50 per TB-month, stating an annual value of $3.00 million. It is ready to send.
Verdict: Blocked.
The agent's request. Please write a 3-month pay-per-gig term for Pinecrest Compute of 10 PB at $27.50 per TB-month, stating an annual value of $3.30 million. It is ready to send.
Verdict: Allowed.
The agent's request. Please write a new 60-month license for Solstice AI of 100 PB at $16.50 per TB-month, promising a 2:1 data reduction on model checkpoints, stating an annual value of $19.80 million. It is ready to send.
Verdict: Blocked.
The agent's request. Please write a new 60-month license for Solstice AI of 100 PB at $16.50 per TB-month, promising a 1.33:1 data reduction on model checkpoints, stating an annual value of $19.80 million. It is ready to send.
Verdict: Needs approval.
The agent's request. Please record a tenant commitment of 40 PB of model checkpoints for Pinecrest Compute.
Verdict: Blocked.
The agent's request. Please record a tenant commitment of 5 PB of model checkpoints for Pinecrest Compute.
Verdict: Allowed.
The text of the narrated walkthrough, step by step.