FieldKit Practical AI Playbook

01. INBOX AND CUSTOMER REPLIES WHEN TO USE A repeat question needs an accurate response without a long back-and-forth. PREPARE An anonymized inquiry, approved service facts, and what must be checked by a person. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Read the inquiry and approved facts below. Draft a reply under 120 words. Answer supported questions, ask only essential clarifying questions, and name the next step. Do not confirm price, availability, or policy exceptions unless explicitly supplied. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: Inquiry: Can you deliver 40 units Friday? Facts: stock and delivery require confirmation; delivery address is needed. Reviewed output: Thanks for your inquiry. What is the delivery address? Our team will check stock and whether Friday delivery is available, then confirm the details. HUMAN CHECK Confirm the draft does not promise stock, a delivery date, or a price. COMPLETION AND MEASUREMENT Useful reply that moves the inquiry forward; track drafting time and corrections.

02. MEETING NOTES INTO TEAM ACTIONS WHEN TO USE A meeting ends with scattered notes and unclear ownership. PREPARE Meeting notes with permission to use them; remove unnecessary personal information. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Extract decisions, actions, owners, and due dates from these notes. Return a table with a supporting quote for every row. Mark missing owners or dates as not specified. Keep suggestions separate from agreed decisions. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: Notes: Jordan will send the draft. We discussed next Friday but did not agree a deadline. Casey suggested a customer survey. Reviewed output: Action: send draft. Owner: Jordan. Due: not specified. Survey: suggestion, not an assigned action. HUMAN CHECK Compare each row with the notes, then confirm missing ownership or deadlines with the team. COMPLETION AND MEASUREMENT An action list the team confirms; track actions closed by their agreed due dates.

03. FIELD NOTES INTO A HANDOFF WHEN TO USE Someone returning from a site visit needs to brief the next shift or office team. PREPARE Anonymized site notes, observed conditions, completed tasks, and unresolved issues. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Turn these field notes into a handoff: completed, outstanding, dependencies, questions, and next owner if stated. Keep observations separate from assumptions. Do not infer that equipment or a site is safe. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: Notes: East beds planted; west irrigation not checked; access gate locked; Sam to call property contact. Reviewed output: Completed: east beds planted. Unchecked: west irrigation. Blocker: locked gate. Action: Sam to call property contact. HUMAN CHECK Check locations, unresolved conditions, and who owns each next step. Use established safety procedures for safety decisions. COMPLETION AND MEASUREMENT A clear handoff; track repeat clarification questions and missed follow-ups.

04. CREATE A REPEATABLE PROCESS WHEN TO USE A routine job depends on one person remembering every step. PREPARE An approved description of the process, required inputs, exception rules, and reviewer. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Create a numbered checklist from these approved instructions. Include inputs, steps, completion evidence, and escalation points. Mark missing rules for review. Do not invent system permissions or policy. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: Facts: check the order number; compare delivered items to packing list; report mismatches to the supervisor. Reviewed output: Check order number → compare items and packing list → record any mismatch → report mismatch to supervisor. Missing rule: how to record a completed check. HUMAN CHECK Walk through the checklist with the person who performs the job before adopting it. COMPLETION AND MEASUREMENT A checklist tested on one real or simulated task; track omissions and rework.

05. WEEKLY MANAGEMENT REVIEW WHEN TO USE You need a short view of progress and what needs a decision. PREPARE Verified metrics, last week’s agreed actions, and context for changes. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Using only these supplied metrics and notes, prepare a weekly brief: changes, open actions, risks, and decisions needed. Show calculations and flag missing or incomparable data. Do not infer causes from a change alone. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: This week: 24 inquiries and 6 sales. Last week: 20 inquiries and 5 sales. Same definitions and measurement window. Reviewed output: Sales per inquiry: 25% in both weeks. Volume increased; the conversion rate did not. No cause can be established from these totals alone. HUMAN CHECK Recalculate figures in a spreadsheet and confirm whether sales came from the same inquiry cohort. COMPLETION AND MEASUREMENT A short brief with one agreed action; track data corrections and time to prepare.

06. RESEARCH AND SUPPLIER COMPARISON WHEN TO USE You need to compare options without mistaking polished language for evidence. PREPARE Current source documents or verified source links, requirements, and decision criteria. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Compare these sources against our requirements. For every claim, quote or identify the supporting passage and source date. Mark absent details as not found. Separate supplier claims from independently verified information. Do not invent sources or prices. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: Source A states delivery within 5–7 working days; source B provides no delivery estimate. Reviewed output: Delivery: A claims 5–7 working days; B not found. Next action: confirm delivery for the specific order with both suppliers. HUMAN CHECK Open the original source, confirm date and scope, and ask suppliers about missing terms. COMPLETION AND MEASUREMENT A source-backed comparison; track unresolved requirements before deciding.

07. SPREADSHEET HELP WITHOUT BLIND TRUST WHEN TO USE A spreadsheet calculation needs to be understandable and testable. PREPARE Column names, intended calculation, a small fictional sample, and expected results. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Suggest a formula for [calculation] using [columns]. Explain every part. Provide tests for a normal row, zero, blank, and invalid input. State how rounding works. Do not change the underlying meaning of the metric. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: Conversion = purchases divided by visits. Example: 8 purchases / 200 visits. Zero visits should not produce a numeric rate. Reviewed output: Expected: 4%. For zero visits return not available. For missing or invalid inputs flag the row for review. HUMAN CHECK Run known cases in the actual spreadsheet. Check percentage formatting, denominator, and blank behavior. COMPLETION AND MEASUREMENT A tested formula with documented assumptions; track incorrect or missing rows.

08. COACH WITH OBSERVABLE EXAMPLES WHEN TO USE You want to prepare a constructive conversation about a work process. PREPARE Specific work examples, agreed expectations, and the support available; avoid unnecessary personal data. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Help prepare a respectful coaching conversation using only these observed work examples and expectations. Draft neutral observations, open questions, and possible support. Do not diagnose motives, personality, health, or protected characteristics. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: Observation: two handoffs lacked order numbers. Expectation: every handoff contains an order number. Reviewed output: I noticed two recent handoffs were missing order numbers. What makes that step difficult? Let’s agree how to make the number easy to include. HUMAN CHECK Review wording and context yourself. Use company procedures and human judgment for employment decisions. COMPLETION AND MEASUREMENT A practical conversation plan; agree a specific process improvement and review date.

09. PLAN A WEEK OF USEFUL CONTENT WHEN TO USE You need useful marketing grounded in what customers actually ask. PREPARE An anonymized recurring question, verified answer, audience, and brand tone. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Turn this question and verified answer into three distinct educational posts. Each should cover one useful point and a relevant next action. Preserve facts; do not invent reviews, outcomes, urgency, or offers. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: Question: what do I need before requesting a quote? Facts: item list, location, and preferred timing help; pricing needs review. Reviewed output: Post idea: Three details that help us prepare your quote—item list, location, and preferred timing. Send those details so our team can review your request. HUMAN CHECK Confirm service scope and wording. Use images you have rights to use and your normal publishing approvals. COMPLETION AND MEASUREMENT Three reviewed drafts; track qualified inquiries and replies, not just likes.

10. IMPROVE ONE WORKFLOW AT A TIME WHEN TO USE You want to know whether an AI-assisted routine is actually useful. PREPARE A baseline task sample, quality checklist, timing method, and comparable trial tasks. DO THE WORK 1. Write the task and gather the approved inputs. 2. Replace any placeholders in the prompt below and paste the inputs after it. 3. Compare the draft with your sources. Correct errors and ask a person about missing facts. 4. Save the reviewed output and record whether the method helped. DRAFT PROMPT Help design a small trial comparing our current process with an AI-assisted draft plus human review. Keep task difficulty comparable. Include drafting, checking, correction time, quality failures, and a stop condition. Do not predict savings without data. REVIEW PROMPT Review the draft against my supplied inputs. Identify unsupported claims, omissions, and unclear instructions. Quote the input that supports each proposed factual correction. If support is missing, flag it instead of filling the gap. REUSE PROMPT Turn the reviewed result into a template. Replace changing details with named placeholders. Add a short input checklist and human review checklist. Do not carry example facts into future jobs. FICTIONAL WORKED EXAMPLE Input: Baseline: 12 minutes total. Trial: 4 drafting + 5 checking + 2 correcting = 11 minutes. Reviewed output: Observed difference for this task: 1 minute. Repeat across comparable tasks before drawing a conclusion. Faster output is not useful if quality falls. HUMAN CHECK Use the same timer boundaries and quality criteria. Count all human review and rework. COMPLETION AND MEASUREMENT An evidence-based keep, revise, or stop decision; record both total time and quality.

PROMPT REPAIR GUIDE Too vague? Add audience, purpose, facts, format, and constraints. Too long? Specify the decision and a realistic word limit; check nothing essential disappeared. Invented details? Restrict the task to supplied sources and label unknowns. Verify manually. Wrong tone? Supply a short approved example and ask for the same level of formality. Missed instructions? Split the job into smaller steps and review each result. Inconsistent results? Save the prompt and inputs; test several representative cases. Bad calculations? Use a spreadsheet or calculator and test known cases. Polished but unusable? Ask what action the reader should take and what information they need. Sensitive input? Stop and use fictional data or the organization’s approved process. Wrong source citation? Open the original source; a citation itself is not evidence.

YOUR 30-DAY IMPLEMENTATION PLAN Days 1–3: List five repeated tasks. Rate frequency, effort, clarity of inputs, ease of review, and consequences of an error. Choose one low-risk task. Days 4–7: Capture a baseline, write a useful brief, and test with fictional inputs. Keep examples of good and bad outputs. Days 8–10: Run a small reviewed trial. Record drafting, checking, and correction time. Days 11–14: Improve one weak point and repeat the test on comparable tasks. Stop if quality is worse or the review effort exceeds the benefit. Days 15–18: Turn the best result into a reusable prompt and checklist. Ask a colleague to try it without extra explanation. Days 19–21: Fix unclear steps, missing inputs, and exception handling. Assign a human owner. Days 22–25: Try a second routine using the same measurement method. Days 26–28: Assemble your playbook: purpose, inputs, prompt, example, review, owner, fallback, and measurement. Days 29–30: Review the evidence. Decide what to keep, revise, or stop. Set a future review date. The schedule is a suggested pace, not a promise of results.

QUALITY GATE BEFORE YOU USE AN OUTPUT 1. Purpose: does it answer the actual task? 2. Facts: can important claims be traced to approved sources? 3. Numbers: did you independently check calculations? 4. Promises: did it invent a price, deadline, commitment, or guarantee? 5. Privacy: is the information appropriate for this tool and audience? 6. Clarity: can the reader understand the next step? 7. Ownership: who signs off and handles exceptions? 8. Fit: does it follow your organization’s procedures? Any critical error means revise or stop, even if the wording is polished.

KNOWLEDGE CHECK 1. AI adds a deadline that is not in your notes. What do you do? 2. Drafting takes 3 minutes, checking 6, and corrections 4. Baseline is 10 minutes. Was this faster? 3. A generated supplier comparison includes a citation. Is that sufficient verification? 4. A post gets more likes but no qualified inquiries. Does that establish improved sales? 5. A field handoff says equipment is safe, but no inspection is documented. Can you use that claim? ANSWER KEY 1. Remove or mark the deadline unconfirmed and check with the owner. 2. No. Total time is 13 minutes, 3 minutes longer. Review why before repeating. 3. No. Open the original and check that it supports the specific claim, date, and scope. 4. No. Measure the intended outcome and allow for the sales cycle and attribution limits. 5. No. Follow the established inspection and safety process; AI cannot infer safety from missing information.