Qualify incoming leads with a transparent scoring rubric
Apply explicit, reviewable criteria without hidden profiling or invented customer facts.
- Difficulty
- Intermediate
- Estimated task time
- Your task: 35 minutes
Your copy-paste prompt
Check applicable privacy and anti-discrimination requirements before automation.
1. Your details
0 of 2 fields filled.
2. Your prompt
Act as a transparent lead-operations analyst. Score [LEAD DATA] only against this approved rubric: [CRITERIA, WEIGHTS AND DEFINITIONS]. For each criterion return supplied evidence, points, missing data and confidence; then total score and routing recommendation. Never infer budget, identity, protected characteristics, intent or company facts. Missing data scores zero or "unknown" exactly as the rubric states. Separate facts from assumptions and flag every lead for human review when evidence is ambiguous. Return a table and plain-language explanation.
Legacy tests
These results come from an earlier test setup and are kept for transparency. They are not a current verdict; this recipe is waiting for tests on the current setup.
Show legacy tests · v1
Test evidence
v1 · 3 published results · 9 October 2026
3 test cases · Different inputs are not a direct model comparison.
GPT-6 AstraTest case 1 · API · Main resultResult · PassResponse · UnavailableAPI cost* · Unavailable
Provider-reported tokens · Unavailable
Pass · 9 October 2026 · Designated main result
The reply meets the requirements in this example. It shows the evidence behind each score, totals 100 points, and explains why the lead should go to sales.
Full test transcript · GPT-6 Astra (API)
The exact input and the complete, unedited reply from this published test run.
Exact input sent to the model
Act as a transparent lead-operations analyst. Score Lead: Brindlepath Workshop, a fictional company. Inquiry: “We need shared task tracking for 18 users. We plan to roll out the software within 21 days. Please arrange a product demonstration.” No additional lead information is supplied. only against this approved rubric: Add three criteria: user count of at least 10 earns 40 points; explicit need for shared task tracking earns 35; planned rollout within 30 days earns 25. Otherwise each earns zero; missing evidence also earns zero. Explicit evidence means high confidence; missing or ambiguous evidence means low. Route totals >=80 to sales, otherwise nurture. Ambiguous evidence requires human review.. For each criterion return supplied evidence, points, missing data and confidence; then total score and routing recommendation. Never infer budget, identity, protected characteristics, intent or company facts. Missing data scores zero or "unknown" exactly as the rubric states. Separate facts from assumptions and flag every lead for human review when evidence is ambiguous. Return a table and plain-language explanation.
Verbatim model output (unedited)
### Lead score: Brindlepath Workshop | Criterion | Supplied evidence | Points | Missing data | Confidence | |---|---|---:|---|---| | User count of at least 10 | “18 users” | 40 | None for this criterion | High — explicit evidence | | Explicit need for shared task tracking | “We need shared task tracking” | 35 | None for this criterion | High — explicit evidence | | Planned rollout within 30 days | “within 21 days” | 25 | None for this criterion | High — explicit evidence | | **Total** | | **100 / 100** | | | **Routing recommendation:** Sales, because the total score of 100 meets the ≥80 threshold. **Human review:** Not required under the ambiguity rule; all three criteria have explicit evidence. **Plain-language explanation:** The inquiry satisfies all three approved criteria: 18 users exceeds the 10-user minimum, shared task tracking is explicitly requested, and the planned 21-day rollout is within 30 days. The demonstration request is a supplied fact but earns no additional points under this rubric. **Facts vs. assumptions:** Only the supplied inquiry was used. No assumptions or inferences were made about budget, identity, protected characteristics, intent, or other company facts.
Recorded limitations and failure notes
- This example does not demonstrate how the recipe handles missing or ambiguous evidence.
Claude Sonnet 5Test case 2 · APIResult · PassResponse · UnavailableAPI cost* · Unavailable
Provider-reported tokens · Unavailable
Pass · 9 October 2026
The reply applies the supplied rubric correctly, awarding 70/100 and recommending sales routing. Each criterion includes evidence, points, missing-data status, and confidence.
Gemini 3.8 FlashTest case 3 · APIResult · PassResponse · UnavailableAPI cost* · Unavailable
Provider-reported tokens · Unavailable
Pass · 9 October 2026
The reply meets the listed requirements for this example. It shows the supplied evidence for each criterion, applies the rubric correctly, and explains the sales recommendation without inventing customer details.
Main result limit: This example does not demonstrate how the recipe handles missing or ambiguous evidence.
How this was tested
- Evaluation
- Human-directed · AI-assisted · Human-reviewed
- Validation
- No independent validation
- Repeatability
- 3 models tested · 1 published run per model
This used one synthetic input: a fictional company requesting shared task tracking for 18 users with rollout within 21 days. It is a narrow demonstration with explicit evidence for all three criteria.
Assessment suggested by GPT-6 Astra, the same model used for the test.
Human-directed means a person chose the test, ran the workflow and kept the evidence. AI-assisted means AI tools helped analyse the output or apply the scoring criteria. Human-reviewed means a person checked the record before it was published — it is not the same as independent validation.
No qualified third party checked the domain conclusions in the output. This is provenance information, not a failure of the test.
Results apply to the recorded inputs, versions and interfaces. They do not establish performance for every input, account or future version.
Test case 1
Same input for every model below · Identical assessment criteria were not established.
Historical assessment · fixed contract not recorded. The original result and retained evidence remain available. A numbered assessment contract was not recorded before this run, so identical judging across these historical results is not established.
Test case 2
Same input for every model below · Identical assessment criteria were not established.
Historical assessment · fixed contract not recorded. The original result and retained evidence remain available. A numbered assessment contract was not recorded before this run, so identical judging across these historical results is not established.
Test case 3
Same input for every model below · Identical assessment criteria were not established.
Historical assessment · fixed contract not recorded. The original result and retained evidence remain available. A numbered assessment contract was not recorded before this run, so identical judging across these historical results is not established.
Response times and API costs
Response time runs from sending the request to receiving the complete answer; checking the answer is not included. *API cost is the provider-reported charge to AI4E for this answer, not your subscription price or a guaranteed price. Repeated runs use medians where recorded. Unavailable means no reliable measurement was saved.
Response times are compared within the same test input; speed is not a ranking of answer quality. API results do not establish how a consumer app performs. For cases labelled fixed before testing, matching verdicts mean the answers met the same recorded criteria; historical assessments do not establish identical judging. Token counts are provider-reported usage and do not establish efficiency or value across models.
From the test record
These notes describe the specific tested input. For your own result, use the recipe's checklist.
In your own generated output, check that each evidence statement matches the inquiry, points follow only the approved criteria, and the total and route are correct. Check that ambiguous evidence triggers human review and that no customer facts are invented.
All product names, logos and brands are property of their respective owners. They are used on this site for identification and comparative benchmarking only; their use does not imply affiliation with or endorsement by them.