CIRAS | Iowa State University
AI Fluency / Cycle 01
Investigate the Need
Experiment: Separate Evidence from Explanations
Two short business questions. Focus on what the evidence supports, what is uncertain, and what to check next.
How to Use This Page
Everything needed for both experiments is on this page. Use the two short evidence cards and copy the prompts. All eight original company documents remain in the expandable library below, but opening or uploading them is optional.
- Follow the instructor’s three AI prompts (15 minutes).
- Investigate a different concern using the second card (20 minutes).
- Compare what the evidence proves with what AI merely suggests (5 minutes).
Guided Experiment – Was the Customer Ever Answered?
Business question: Why might referrals lack dependable follow-through?
Our job: We are a cross-functional employee team investigating the support operation, not outside consultants. We are looking for one uncertainty worth checking, not solving the whole operation in 15 minutes.
- MGT-01 Management says some customers contact support more than once. No frequency has been measured.
- OPS-01 A department replies through CRM, notifying the CSR, who contacts the customer. Case closure is recorded inconsistently.
- ACT-1003 Engineering response logged: Yes. Customer response logged: No. CRM status: Referred.
- ACT-1010 Parts response logged: Yes. Customer response logged: Yes. CRM status: Referred.
- INT-C A CSR says Referred can mean still waiting, already answered, or simply not updated.
Move 1 – Read the evidence
Look specifically at ACT-1003 and ACT-1010. Both say Referred, but their documented customer responses differ.
Move 2 – Separate fact, report, and unknown
Start a new AI chat. The first copy button includes both the evidence and the prompt, so no document uploads are required. Read just the three rows AI returns.
Show exact prompt and source evidence
You are helping an internal improvement team at a fictional manufacturer. Use ONLY the evidence below. GUIDED CASE: Customer referral follow-through [MGT-01] Management says some customers contact support more than once. No frequency has been measured. [OPS-01] A department replies through CRM, notifying the CSR, who contacts the customer. Case closure is recorded inconsistently. [ACT-1003] Engineering response logged: Yes. Customer response logged: No. CRM status: Referred. [ACT-1010] Parts response logged: Yes. Customer response logged: Yes. CRM status: Referred. [INT-C] A CSR says Referred can mean still waiting, already answered, or simply not updated. Return exactly three rows in a compact table with columns: Category | Finding | Source. Rows: (1) one fact visible in the selected records, (2) one employee report, (3) one important unknown. Maximum 75 words total. Distinguish missing documentation from missing customer action. Do not recommend solutions.
Move 3 – Challenge an appealing conclusion
See whether the records actually support the claim that a Referred status means the customer never received an answer.
Show exact follow-up prompt
A colleague says: "Every activity still marked Referred is waiting for a customer answer." Does the evidence support that statement? Identify the specific record that tests it and offer one other reason the status might remain unchanged. Answer in exactly two brief sentences, 55 words maximum.
Move 4 – Decide what to verify
Ask for two specific checks. Point out which record or person could help us verify the answer. Do not redesign the workflow.
Show exact follow-up prompt
Suggest exactly two small verification checks we could perform before concluding that referrals lack follow-through. Use a two-row table with columns: Specific check | Person or record to consult. At least one check must examine whether ACT-1003's customer actually received a response. Maximum 60 words total. No rates, savings, or software recommendations.
End with: one unsupported assumption we avoided and two practical checks to perform.
Independent Experiment – Why Does CSR Training Take Months?
Your question: What might explain the company’s reported three-to-six-month learning curve for new CSRs?
Use the same fictional company, but explore a different problem. You only need the five observations below. The fuller company library is optional.
- MGT-02 Management estimates that new CSRs need 3 to 6 months to work reasonably independently. This is not a measured baseline.
- OPS-01 New CSRs primarily learn by shadowing experienced colleagues.
- INT-A An experienced CSR says identifying older product generations requires learned cues; some detailed answers stay in personal email.
- INT-B Another CSR thinks routine inquiries may become manageable in about 3 months; there is no organized sequence of practice cases.
- INT-C A CSR reports that earlier customer conversations and their outcomes can be difficult to reconstruct.
Step 1 – Read the evidence
Notice that three to six months is an estimate. Also notice that the CSRs do not entirely agree about what makes someone independent.
Step 2 – Explore two possible explanations
Start a new AI chat. Copy the evidence and the first prompt together. Keep the answer to two rows.
Show exact prompt and source evidence
You are an employee investigating how new CSRs learn to work independently. Use ONLY the evidence below. INDEPENDENT CASE: New CSR learning curve [MGT-02] Management estimates that new CSRs need 3 to 6 months to work reasonably independently. This is not a measured baseline. [OPS-01] New CSRs primarily learn by shadowing experienced colleagues. [INT-A] An experienced CSR says identifying older product generations requires learned cues; some detailed answers stay in personal email. [INT-B] Another CSR thinks routine inquiries may become manageable in about 3 months; there is no organized sequence of practice cases. [INT-C] A CSR reports that earlier customer conversations and their outcomes can be difficult to reconstruct. Give TWO plausible explanations for the training time. For each, name one source-supported observation. Use a two-row table: Possible explanation (UNPROVEN) | Supporting report and source ID. Maximum 85 words total. Do not propose a training program or buy a tool.
Step 3 – Ask what would distinguish them
Ask AI for two small checks, then consider whether either would actually help answer the question.
Show exact follow-up prompt
We cannot yet tell which explanation matters more. Suggest TWO small checks that could distinguish them. Use two rows: Check | Person or record to consult. Maximum 60 words total. Do not treat the 3-to-6-month estimate as a verified measurement.
Step 4 – You decide
Choose the one check you would do first. You may ask AI one question of your own to make it more practical.
1. One observation and source ID
2. Two possible explanations (unproven)
3. One check and who or what you would consult
4. One measure of improvement worth tracking
Need help choosing a check?
Consider asking a recently trained CSR what types of inquiries they can handle independently, or comparing how a new CSR handles a routine inquiry versus an older-model identification problem. In either case, avoid assuming the answer.
Compare / Reflect
Use the final five minutes to discuss two questions:
- What did the evidence actually establish, and what was only a plausible explanation?
- Which small check would most improve our confidence before recommending a change?
See a short checkpoint after your attempt
Guided: ACT-1010 records a customer reply even though its status remains Referred. ACT-1003 does not document a customer reply, which does not prove no reply occurred. Check the originating CSR’s correspondence or call history and review CRM status/closure history.
Independent: MGT-02 provides an estimate, not a training measurement. INT-A emphasizes historical product identification; INT-B suggests routine competence may come sooner. One useful check is to sample recent trainees’ performance by inquiry type, including accuracy and appropriate escalation.
Full Company Library – Optional
Everything in the original company library remains available here. You do not need to open these files for the guided or independent experiment. The two five-observation cards above package what you need for the 15- and 20-minute activities.
View all eight original documents and datasets
Full business evidence and data
Original fictional management, interview, CRM, phone and operating records. Useful for deeper investigation and later cycles.
EXCELSupport Operations SampleRaw records and working notes.
CSVCRM Activity SampleSelected CRM activity export.
CSVPhone Extract SampleSelected phone activity export.
Original extended exercise materials
Earlier, longer versions of the Cycle 1 activity. Keep for reference, but use the simplified exercises above in class.
Source note: All observations are fictional and derived from the existing company source materials. A selected record is not a measured company-wide rate.

