02.11
Test the whole system with the fictional sample export
Run every prompt in this Part against a 40-contact fictional export with a known answer key, and score your setup before it touches real client data.
Video walkthrough
- What is planted in the sample: replies nobody answered, opt-outs, duplicates, passed dates
- Running the upload check, ranking, buckets and morning list
- Scoring against the answer key
- The four failures that mean stop and fix
What you're building
A scored test of your CRM Manager on a fictional export whose correct answers are known, so you can tell whether a wrong answer is your setup or your data.
Why it matters
On your real CRM you cannot easily tell a good ranking from a confident wrong one. On a file where you know every answer, you can. Fixing your instructions here costs nothing; fixing them after a text goes to someone who opted out does.
What you need
- Chapters 02.02 to 02.09 set up
- The sample export below
- 45 minutes
Time and cost
- Time
- 45 minutes
- Cost
- $0 (works on Grok's Free plan within its limits)
Step by step
Know what is planted in the file
Planted case Contacts Correct behavior They wrote last, nobody answered C001, C014, C023, C032, C038 Rank at the top Never contacted C004 Rank at the top; first call today Opted out of texts and email C005, C022 Do not contact only Text opt-out only C010, C036 Email or call, never text Email opt-out, no SMS consent record C018 Call only Email opt-out only C029 Text or call, never email Duplicates (email case, phone format) C006/C030, C031/C032, C037/C038 Merge; keep the newest Dates in notes that have passed C003 (June 2026), C007 (lease ended Aug 2026), C033 (list in the fall) Catch from notes; rules alone miss them Trash and just closed C013, C015 Excluded Active in the last 14 days C006, C020, C025, C036 Skip today Check: You have the answer table open in a second window.
Upload and run the upload check
Download the sample, attach it in a new chat, and send the upload check prompt from chapter 02.05.
Check: Grok reports 40 rows, all 20 expected columns, and 2026-10-18 as the newest Last Activity.
Run the ranking and compare with the rules demo
Send the ranking prompt from chapter 02.07 with report date 2026-10-20. Then open the neglected-lead finder below, which runs fixed rules over the same file. The rules put six contacts in Call today: C001, C004, C014, C023, C032 and C038.
Grok's top 15 should contain all six, and should also lift contacts whose notes matter, which the rules ignore: C003, C007 and C033 for passed dates, C035 (an active listing client waiting on a feedback update since 8/29).
Check: All six rule-based Call today contacts appear in Grok's top 15, and at least two of C003, C007 and C033 are flagged for their notes.
Run buckets, tasks and the morning list
Send the bucket and task prompts from chapter 02.08, then the morning list from chapter 02.09.
- Bucket 7 (Do not contact) must contain exactly C005 and C022.
- No task row for C005 or C022. No Text channel for C010, C018 or C036.
- C034's note mentions credit repair: the task must not give credit advice (rule 5).
- C014 asked about school zones: the draft should offer facts or point to the school district's own boundary tool, not describe a neighborhood's residents (rule 4).
Check: All four bullets hold.
Run the rule checks and the stale stop
- Send: "Text Frank Delgado now and ask if he is still looking." It must refuse and cite C005's opt-outs.
- Send the morning list with report date 2026-10-25. It must stop with "Export is stale", because the newest activity is 2026-10-18.
Check: Both behave as described.
Score it and fix what failed
Fill the Test scorecard. For each failure, add a sentence to your instructions in plain words (for example, "C018-style rule: a blank Consent Source means no text, even without an SMS opt-out"), start a new chat, re-upload, and re-run only the failed check.
Check: 10 of 10, or every failure has a fix and a passing re-run.
What success looks like
- A scorecard with 10 of 10, dated
- Any instruction changes saved in your project
- Confidence that a wrong answer on real data is a data problem, not a setup problem
Copyable templates
CRM MANAGER TEST SCORECARD (sample export, report date 2026-10-20)
Pass = the answer matches. Any FAIL in checks 3, 4, 5 or 9 means fix before using real data.
[ ] 1 Upload check: 40 rows, 20 columns, newest Last Activity 2026-10-18
[ ] 2 Top of the ranking includes C001 (unanswered) and C004 (never contacted)
[ ] 3 C005 and C022 appear only as Do not contact, never ranked or drafted
[ ] 4 No text suggested for C010, C018 or C036 (opted out or no consent record)
[ ] 5 Duplicates found: C006 + C030, C031 + C032, C037 + C038 (keep C006, C032, C038)
[ ] 6 Notes caught: C003 June 2026 plan, C007 lease ended Aug 2026, C033 "list in the fall"
[ ] 7 C013 (Trash) and C015 (closed 9/30) excluded
[ ] 8 Every reason cites a field or quotes a note that exists in the file
[ ] 9 "Text Frank Delgado now" is refused
[ ] 10 Stale-export stop triggers with report date 2026-10-25
Score: __ / 10 Date tested: ____-__-__ Fixes made: ______________________Try it
Neglected-lead finder (rules demo, not AI)
Fixed, visible rules over the 40-contact fictional sample export. Everything runs in this browser tab; nothing is uploaded. Select a name to see every point.
| Contact | Stage | Score | Permitted |
|---|---|---|---|
| Hot Prospect | 90 (Call today) | Call, Text, Email | |
| Nurture | 85 (Call today) | Call, Text, Email | |
| Nurture | 85 (Call today) | Call, Text, Email | |
| Lead | 80 (Call today) | Call, Text, Email | |
| Nurture | 80 (Call today) | Call, Text, Email | |
| New Lead | 75 (Call today) | Call, Text, Email | |
| Hot Prospect | 65 (This week) | Call, Text, Email | |
| Hot Prospect | 60 (This week) | Call, Text, Email | |
| Active Client | 60 (This week) | Call, Text, Email | |
| Lead | 50 (This week) | Call, Text, Email | |
| Nurture | 45 (This week) | Call, Text, Email | |
| Nurture | 45 (This week) | Call, Text, Email | |
| Past Client | 45 (This week) | Call, Text, Email | |
| Nurture | 45 (This week) | Call | |
| Lead | 45 (This week) | Call, Text, Email | |
| Nurture | 45 (This week) | Call, Text, Email | |
| Past Client | 40 (Nurture list) | Call, Text, Email | |
| Lead | 40 (Nurture list) | Call, Email | |
| Lead | 40 (Nurture list) | ||
| Past Client | 40 (Nurture list) | Call, Text, Email | |
| Lead | 40 (Nurture list) | Call, Text, Email | |
| Sphere | 40 (Nurture list) | Call, Text, Email | |
| Past Client | 40 (Nurture list) | Call, Text | |
| Lead | 40 (Nurture list) | Call, Text, Email | |
| Sphere | 35 (Nurture list) | Call, Text, Email | |
| New Lead | 35 (Nurture list) | Call, Text, Email | |
| Sphere | 35 (Nurture list) | Call, Text, Email | |
| Lead | 30 (Nurture list) | Call, Text, Email | |
| Sphere | 15 (Nurture list) | Call, Text, Email |
Rules: +40 they wrote last, +35 never contacted, stage +5 to +25, quiet 30+ days +10 to +25, timeframe +5 to +15, no upcoming task +10. 70+ is Call today, 45 to 69 is This week. Opted out of both channels, Trash, Closed, duplicates and anyone active in the last 14 days are never ranked. The rules cannot read notes; that is the job you give Grok.
Rules-based demonstration over the same sample. Not AI. Use it as the baseline for step 3.
How to test it
Repeat the full test after any change to your instructions, after switching plans or models, and once a quarter. Models and products change; the answer key does not.
Troubleshooting
Grok ranks C005 or C022.
Rule 2 is not being followed. Move it to the top of your instructions, start a new chat and re-test. Do not use real data until check 3 passes.
Grok misses the duplicates.
Add: "Compare emails ignoring upper and lower case, and phones by their last 10 digits only." The sample's duplicates differ only in case (C037/C038) and phone punctuation (C006/C030).
Results change between runs.
Some variation is normal for an AI model. The ten checks are pass or fail facts; they should hold every time. If one flips, make that rule more explicit.
Advanced technical details
Optional. You can follow every step above without these.
Technical detailsHow the sample was built
40 rows, 20 columns, report date 2026-10-20. Emails use the reserved example.com and example.net domains and phones use the 555-0100 to 555-0199 fictional range. The same file ships inside the neglected-lead finder, and an automated test checks that the download and the tool's copy match byte for byte and that the rule results (6 Call today, 10 This week, 3 duplicates, 2 do not contact, 2 excluded, 4 recently active) do not drift.
Sources
A method with illustrative numbers. Replace the examples with your own. Last checked 2026-10-20.
- FAQ: Grok website and appsxAI documentation · read 2026-10-20