Jeff Bullas

Yes — AI can speed you to trustworthy sources, but only if you use it like a smart assistant, not the final judge.

Start with a clear question, let AI find candidate sources, then verify those candidates with simple checks. You’ll get fast wins: a focused reading list, notes on credibility, and a short bibliography you can trust.

What you’ll need

• Your exact research question or topic (one sentence).
• Access to an AI chat tool (web-enabled is best) and a web browser or university library access.
• Basic judgment criteria: publication date, peer review, author expertise, citations, publisher reputation, and potential bias.

Step-by-step: use AI to find and verify trustworthy sources

1. Clarify the question. Write one clear sentence summarizing your research aim.
2. Ask AI for candidate sources. Paste the prompt below (copy-paste) into the AI tool and request an annotated list of 8–12 sources with why each is relevant and a trustworthiness score.
3. Verify each source. For every suggested paper/book/website, check: peer-reviewed status, publication date, author affiliation, number of citations, and whether the publisher is reputable.
4. Retrieve full text. Use your library, Google Scholar, or databases like PubMed/JSTOR to get the full articles. If behind paywalls, request via interlibrary loan or contact the author.
5. Create a short annotated bibliography. Keep 6–10 best sources, with 2–3 sentences on each: main finding, reason to trust, and how it fits your paper.

Copy-paste AI prompt (use as-is)

“I am researching: [insert one-sentence research question]. Please list 10 high-quality sources (peer-reviewed articles, books, or authoritative reports) published in the last 10 years. For each source give: full citation, 2-sentence summary, reason it’s trustworthy (peer review, publisher, citations), and one sentence on how it helps answer my question. Prioritize recent reviews and empirical studies. If a source is not openly accessible, note how I can access it (library, database, or author).”

Prompt variants

• Quick review: Ask for 5 recent review articles and why they’re central.
• Grey literature: Ask for authoritative reports and policy papers and how to assess their bias.
• Counter-evidence: Ask AI to list 3 strong sources that disagree with the main view.

Common mistakes & fixes

• Trusting AI without checks — fix: verify peer-review, citations, and authorship yourself.
• Accepting old or irrelevant sources — fix: set a date range and ask AI to prioritize recent reviews.
• Confirmation bias — fix: ask for opposing evidence and include it in your review.

Action plan (30–90 minutes)

1. Write your one-sentence question (5 min).
2. Run the main AI prompt and pick 10 candidates (10–20 min).
3. Verify top 6 sources via your library or Google Scholar (15–30 min).
4. Draft a 1-page annotated bibliography (15–30 min).

Small, consistent steps win. Use AI to find, then your judgement to verify. That combo gives you a trustworthy, time-saving research workflow.

Nice focus — automation plus accuracy is exactly the right problem to solve. Here’s a practical, low-friction way to turn meeting notes into Jira or Trello cards so you get quick wins without heavy engineering.

Quick checklist — do / do not

• Do: start with a simple meeting-note template (Action, Owner, Due, Context).
• Do: use an integration tool (Zapier, Make, or n8n) or direct API if you’re comfortable.
• Do: validate extracted tasks before creating cards (one-click confirm).
• Do not: assume free-form notes will always map cleanly—use prompts and rules.
• Do not: push many cards without deduplication or priority rules.

What you’ll need

1. Meeting notes stored in one place (Google Doc, Notion, or email).
2. An LLM service (ChatGPT-style) or a simple text-extraction rule set.
3. An integration platform (Zapier/Make/n8n) or direct Jira/Trello API access.
4. A brief template for note-takers and a small validation UI (email or Slack confirmation).

Step-by-step

1. Start: Ask note-takers to use a tiny template: “Action:, Owner:, Due:, Context:”.
2. Trigger: When the note is saved, the integration sends the text to the LLM for parsing.
3. Transform: LLM extracts tasks and maps fields to Jira/Trello schema (title, description, assignee, due date, labels).
4. Validate: Send a summarized preview to the meeting owner for quick confirm or edit.
5. Create: On approval, integration creates cards in Jira or Trello and returns links to the team.

Copy-paste AI prompt (use as-is)

You are an assistant that converts meeting notes into task cards. Extract each actionable item and return a JSON array with: title (short), description (one paragraph), owner (name or “unassigned”), due_date (YYYY-MM-DD or “none”), priority (low/medium/high), labels (comma-separated). If an item is not actionable, skip it. If the owner is unclear, set owner to “unassigned” and priority to “medium”. Keep titles under 60 characters.

Worked example

Meeting note: “John to draft Q3 ad copy by May 10. Sarah to review budget numbers next week. Action: update landing page UI.”

LLM output (cards):

• Title: Draft Q3 ad copy — Owner: John — Due: 2025-05-10 — Priority: High
• Title: Review budget numbers — Owner: Sarah — Due: 2025-05-16 — Priority: Medium
• Title: Update landing page UI — Owner: unassigned — Due: none — Priority: Medium

Mistakes & fixes

• Ambiguous owners — require confirm step before card creation.
• Duplicate tasks — add a simple dedupe check (same title + owner).
• Wrong due dates — parse dates conservatively and ask for confirmation.

Action plan — 3 quick wins

1. Create a two-line note template and ask the team to use it for one week.
2. Build a Zapier flow: Notes → LLM extraction → Preview message → Create card.
3. Run for a sprint, collect feedback, tighten prompts and validation rules.

Small steps win. Start with a template and a confirmation step — you’ll get reliable automation fast and improve it as you go.

Good point — focusing on finding trustworthy sources first is the right move. AI can speed that up, but you still need a simple verification routine.

Quick context: AI is best as a smart research assistant: it suggests where to look, proposes search queries, summarizes findings, and flags likely trustworthy sources. It can’t replace your judgment — but it can make your search faster and smarter.

What you’ll need

• Your research question or topic, clearly written.
• Access to an AI chat tool (ChatGPT, Bard, etc.) and at least one academic search (Google Scholar, JSTOR, PubMed, your library).
• Simple evaluation criteria: currency, relevance, authority, accuracy, and purpose (CRAAP).
• A note-taking place (document, spreadsheet, or reference manager).

Step-by-step plan

1. Define the scope: one-sentence thesis, date range, study types you want (surveys, RCTs, reviews).
2. Ask the AI for a search strategy and recommended databases. Use the prompt below (copy-paste).
3. Run the search queries in Google Scholar and a library database; collect top 10 results and PDFs.
4. Evaluate each source quickly with CRAAP: note date, author credentials, citations, methods, funding/conflict of interest.
5. Ask the AI to summarize the 3–5 best sources and draft a short annotated bibliography.

Robust AI prompt (copy-paste)

“I am writing a research paper on [insert topic, date range]. Suggest a short search strategy and 6 precise search queries I can use in Google Scholar and library databases. Then list the top 10 types of sources I should prioritize (peer-reviewed articles, reviews, books, reports) and provide a 1-paragraph checklist (Currency, Relevance, Authority, Accuracy, Purpose) to quickly evaluate each source.”

Example (worked)

• Topic: impact of remote work on employee productivity, 2018–2024.
• Sample query: “remote work” AND “employee productivity” AND (survey OR longitudinal) 2018..2024
• Expect to find: meta-analyses, large surveys, and employer reports. Prioritize peer-reviewed meta-analyses and government/NGO datasets.
• Quick eval: Author: university professor with publications in organizational psychology; Method: national survey of 5,000 workers; Conflicts: funded by government — low commercial bias.

Common mistakes & fixes

• Do not trust a source because it appears first in a normal web search — use academic filters.
• Do not assume every PDF is peer-reviewed. Fix: check the journal and DOI.
• Do not skip checking funding or conflicts. Fix: scan the acknowledgements and author bios.

30–60 minute action plan

1. Write your one-sentence thesis (5 min).
2. Paste the AI prompt above and get a search strategy (10–15 min).
3. Run 3 top queries in Google Scholar and save 6–10 PDFs (20–30 min).
4. Evaluate sources with CRAAP and note 3 strongest for your paper (15 min).

Start small, verify carefully, and let AI handle the grunt work of finding and summarizing. You’ll get to reliable sources faster and keep control of quality.

Best of luck — go find the evidence. — Jeff

Short answer: Yes. With a few simple prompts and a light-weight “answer library,” AI can slash hours from comparing and responding to RFPs—without you becoming technical.

Why this works: RFPs are structured documents. AI is great at extracting requirements, building checklists, comparing options, and drafting clear, on-brand language—if you give it boundaries.

What you’ll need

• A general AI chat tool that accepts documents or pasted text.
• A simple spreadsheet (for the compliance matrix).
• 3–6 reusable documents (your “answer library”): Company facts, Differentiators, Past Performance/Case studies, Security & Certifications, Implementation & Support, Standard Terms.
• Sanitized content only; remove client names, prices, and sensitive details unless using an approved, secure environment.

Do / Do not

• Do feed the RFP text in parts and ask for a compliance matrix.
• Do give AI your answer library and tell it to only use those facts. Ask it to leave placeholders when unsure.
• Do demand tables, bullet points, word counts, and evidence.
• Do save prompts you like; reuse them on the next RFP.
• Do not paste confidential or pricing data unless your tool is approved for that use.
• Do not let AI invent capabilities, dates, or certifications. Require citations to your docs.
• Do not overpromise or ignore “must-have” items. The matrix is your guardrail.

Step-by-step: from intake to draft in under an hour

1. Intake summary (10 minutes)
   • Paste the RFP overview and key sections.
   • Use this prompt:

Copy-paste prompt: “You are a proposal analyst. Read the RFP text below. Produce: 1) a one-page brief (client goals, scope, timeline, budget signals, evaluation weights); 2) a compliance checklist with Must/Should/Could; 3) a list of missing information to clarify; 4) a go/no-go quick view (Fit, Effort, Risk, Likelihood to win) scored 1–5 with justification. Use the buyer’s language where possible. If unsure, mark as ‘Unknown’. RFP text: [PASTE].”

1. Build the compliance matrix (10 minutes)
   • Ask for a CSV-ready table with columns: ID, Section, Requirement, Must/Should/Could, Response Owner, Evidence Needed, Due Date, Status, Risk Notes.
   • Then copy it into your spreadsheet and assign owners.
2. Set win themes (5 minutes)

• Tell AI your top 2–3 win themes (e.g., fast go-live, risk reduction, proven in your industry). These will shape every answer.

1. Create a mini answer library (10 minutes)
   • Paste short documents for: About Us, Differentiators, 2 Case Studies with outcomes, Implementation timeline, Support model, Certifications.
   • Then run this constraint prompt:

Copy-paste prompt: “Use ONLY the facts in my Answer Library to draft RFP responses. If a fact is missing, write ‘[PLACEHOLDER: NEED INFO]’. Do not invent data. When you include a claim, add (Source: [Doc Name]). Answer Library: [PASTE DOCS]. Acknowledge with ‘Library loaded.’”

1. Draft targeted responses (15 minutes)
   • Work section by section. Enforce word counts. Ask for evidence and client outcomes, not fluff.

Copy-paste prompt: “Draft the response for Requirement IDs [LIST]. Constraints: 1) 250 words max; 2) Client-first language, reading level Grade 9–10; 3) We will statements (implementation steps); 4) Proof points with (Source: Doc); 5) Mirror the buyer’s terms (e.g., ‘WCAG 2.1 AA’, ‘go-live’). End with 3 assumptions and 3 risks with mitigations. If any claim isn’t in the Library, use [PLACEHOLDER].”

1. Compare multiple RFPs fast (10 minutes)
   • Feed AI the one-page briefs from each RFP and ask for a fit/effort/risk vs. reward comparison table with a recommended priority.

Insider trick: Win Themes before Words — give AI a 3-line positioning message up front, and it will thread it through every answer. Also ask it to echo the client’s phrases; evaluators reward familiarity.

Worked example (condensed)

RFP snippet: “City requests a website redesign. Must meet WCAG 2.1 AA. Go-live in 90 days. Include hosting, CMS training, and content migration of 200 pages. Evaluation: 30% experience, 30% approach, 20% price, 20% references.”

AI compliance matrix (excerpt):

• ID 1 | Section 2.1 | WCAG 2.1 AA compliance | Must | Owner: Accessibility Lead | Evidence: Prior audits, tooling list | Risk: Medium
• ID 2 | Section 3.4 | 90-day go-live | Must | Owner: PM | Evidence: Timeline, resourcing plan | Risk: High
• ID 3 | Section 4.2 | Hosting included | Should | Owner: DevOps | Evidence: SLA, uptime, region | Risk: Low

AI-drafted response (excerpt, 170 words):

We will deliver an accessible site that meets WCAG 2.1 AA by using component-level patterns tested with automated and manual checks. Our process includes contrast validation, keyboard-only navigation, and screen-reader testing before each sprint review (Source: Implementation). We cut risk on the 90-day timeline by running discovery and design in parallel with content inventory, then migrating pages in batches using our CMS importer (Source: Implementation). This approach has delivered city websites in 60–75 days with zero missed launch dates (Source: Case Study A).

Hosting is provided on a managed environment with 99.9% uptime and daily backups (Source: Support & SLAs). We train staff via two 90-minute workshops and provide 30 days of hypercare post-launch (Source: Support). Assumptions: 1) Final sitemap approved by Day 10; 2) Migration scope limited to 200 pages; 3) No custom integrations beyond listed items. Risks and mitigations: 1) Content delay → early content freeze; 2) Accessibility defect → pre-launch audit; 3) Scope creep → change-control with impact notes.

Common mistakes & quick fixes

• Mistake: Generic answers. Fix: Force evidence lines and client outcomes; require (Source: Doc).
• Mistake: Missing must-haves. Fix: Work from the compliance matrix first, writing only to IDs.
• Mistake: Hallucinated claims. Fix: Use the Library constraint + [PLACEHOLDER] rule.
• Mistake: Word-count busts. Fix: Set per-section limits in the prompt and ask AI to count words.
• Mistake: Weak questions to the buyer. Fix: Ask AI for clarifying questions tied to each requirement ID.

Action plan (first run)

• Minutes 0–10: Load your Answer Library. Declare 2–3 win themes.
• Minutes 10–25: Intake summary + go/no-go.
• Minutes 25–40: Build the compliance matrix. Assign owners.
• Minutes 40–60: Draft two critical sections + executive summary. Generate buyer questions.
• Next day: Internal review. Replace [PLACEHOLDER]s. Final polish and formatting.

One more high-value prompt (gap and risk scan): “Review our draft against the compliance matrix. List: 1) uncovered requirements; 2) risky assumptions; 3) proof gaps; 4) claims without sources; 5) tone mismatches vs. buyer’s words. Return as a table with severity and a suggested fix.”

Closing thought: Treat AI like a disciplined proposal coordinator. Feed it the RFP, your facts, and your win themes. It will give you structure, speed, and sharper language—so you can focus on strategy, pricing, and relationships.

Try this now (under 5 minutes): Copy one paragraph from a news site, paste it into your AI chatbot, and use the prompt below. You’ll instantly get a simplified version, key vocabulary, and quick questions you can use today.

Copy‑paste prompt: “Rewrite the text below for an English learner at CEFR A2. Keep 120–150 words. Include: (1) a 10‑word glossary with simple definitions and an example sentence, (2) 5 comprehension questions (mix yes/no and wh‑), (3) 5 sentence starters (I think…, Because…, In my opinion…, The main idea is…, One example is…). Use clear, short sentences. Avoid idioms. Text: [paste your paragraph]”

Why this works: AI is great at scaffolding—breaking a task into smaller, doable steps. With the right prompt, you can level texts, pre‑teach vocabulary, and create speaking/writing frames in minutes. You stay the guide; AI does the heavy lifting.

What you’ll need:

• Any AI chatbot (free is fine).
• Optional: a phone or laptop with voice typing and read‑aloud for listening and speaking practice.
• A short text (100–200 words) or a topic learners care about (food, work, travel, health).

Step‑by‑step: scaffold a lesson in 15 minutes

1. Level the text (I‑Do)Paste your paragraph and use the quick‑win prompt above. Expect a simple version, a small glossary, and five questions.
2. Pre‑teach vocabularyAsk AI for pictures you can search for, plus a bilingual gloss if helpful. Use this prompt:“From the text above, choose the 8 most useful words. For each: give a simple definition, one example sentence, a picture idea I can search (e.g., ‘thermometer on a hot day’), and a translation into [your language] in parentheses. Keep it friendly.”
3. We‑Do guided practiceCreate gap‑fills and sentence frames. Prompt:“Make a 10‑item gap‑fill from the A2 text with a word bank. Then add 8 sentence frames that model the grammar and vocabulary. Provide the answer key.”
4. You‑Do independent practiceGenerate a short writing task with scaffolds that fade. Prompt:“Create a 7‑sentence writing task about the topic with three levels of support: Level 1: sentence frames and word bank; Level 2: only topic sentences and transition words; Level 3: no support. Include a simple 5‑point checklist.”
5. Listening & speakingTurn the text into a short audio script (you can read it or use read‑aloud). Prompt:“Write a 120‑word script based on the A2 text at slow speed. Mark syllable stress with CAPS (e.g., imPORtant). Add a shadowing version on a new line with slashes for phrasing (e.g., ‘I would LIKE / to EXplain…’). Include 5 echo‑reading lines.”

Insider template: one prompt, three levels (saves time)

Use this when your group has mixed levels. You’ll get A2, B1, and B2 scaffolds in one go.

Copy‑paste prompt: “Create three versions of the same text (topic: [topic]). Version A2 (120–150 words): short sentences, high‑frequency words, 8‑word glossary, 5 questions. Version B1 (180–220 words): add 1 short quote, 2 connectors (however, because), 8‑word glossary, 5 questions including 1 inference. Version B2 (220–260 words): add 1 statistic or example, 8‑word glossary with collocations, 5 questions including 1 critical thinking. For each level, include: (a) 6 sentence frames, (b) one short writing prompt, (c) a speaking pair task with roles. Avoid idioms.”

Examples of ready‑to‑use prompts (copy, paste, adapt)

• Error feedback with mini‑lesson: “Here is a student paragraph: [paste]. Mark errors with [brackets], rewrite a correct version, and explain the top 2 patterns (e.g., articles, verb tense) with 3 easy examples each. Finish with a 6‑item micro‑practice and answer key.”
• Role‑play with fading support: “Create a role‑play at A2 about [situation]. Part 1: full cues and sentence starters. Part 2: only key phrases. Part 3: role cards with goals and problems. Include a pronunciation tip and 5 follow‑up questions.”
• Picture talk: “Make a picture description activity about [scene]. Provide 12 guiding questions, 10 useful phrases, and 6 things to notice. Add a 5‑sentence model answer and a challenge task for fast finishers.”
• CEFR‑aware quiz: “Build a 12‑question quiz from the A2 text: 4 vocabulary, 4 grammar-in-context, 4 comprehension. Include an answer key and a 1‑sentence explanation for each answer.”

What to expect (and how to tune it):

• Good first drafts that need light checking. Ask for fewer words if it’s too long.
• Occasional odd examples—just say: “Regenerate with everyday, real‑life examples.”
• Level drift. Fix with: “Keep CEFR A2. Use only the 2000 most common words unless needed.”

Common mistakes & quick fixes

• Too much text at once. Fix: Process 1–2 paragraphs at a time.
• Only translating. Fix: Use bilingual glosses sparingly; keep tasks in English with supports.
• No listening/speaking. Fix: Add echo reading, shadowing, and short role‑plays.
• Unclear success criteria. Fix: Ask AI for a 5‑point checklist or a simple rubric (1–5) for the task.
• Complex prompts. Fix: Use templates above and change [topic], [level], [your language].

Action plan (20 minutes today)

1. Pick a short text (workplace email, news paragraph, or a note from school).
2. Run the 5‑minute leveling prompt.
3. Generate the 10‑item gap‑fill and sentence frames.
4. Create the 120‑word speaking script with stress marks and phrasing.
5. Print or share. Tomorrow, add the role‑play with fading support.

Pro tip: Save your best prompts as “recipes.” Reuse them each week with new topics (health, money, jobs, community). Consistency beats complexity.

Bottom line: With a handful of targeted prompts, AI becomes your fast scaffolding partner—leveling texts, building vocabulary, and guiding practice without burning hours. Start small, keep it simple, and let the supports fade as confidence grows.

Smart question. Being intentional about demos and discovery is where AI shines—turning scattered prep into a repeatable, high-impact routine.

Why this works

• AI helps you translate product features into business outcomes for each stakeholder.
• You’ll walk in with a clear story, tailored questions, and ready answers to tough objections.
• Expect sharper meetings, fewer surprises, and cleaner next steps.

What you’ll need

• Your AI assistant (any reputable one).
• A one-pager on your product (capabilities, integrations, proof points).
• Public info you can paste: company summary, recent news, job posts, or quotes from earnings calls.
• Names/titles of attendees, meeting goal, and agenda.
• Guardrail: don’t paste confidential or regulated data.

Fast prep workflow (30–45 minutes)

1. Create an account brief (8 minutes)Paste public info. Ask for a one-page snapshot: priorities, risks, initiatives, and what they likely care about from your category.
2. Map stakeholders (5 minutes)Generate likely roles, goals, anxieties, and the top three discovery questions for each person.
3. Value hypothesis (6 minutes)Turn your features into business outcomes with simple metrics and a 90-day win.
4. Demo storyline (10 minutes)Design a short narrative and click-path that proves value fast, with two “wow” moments.
5. Objection handling (5 minutes)Prepare crisp answers for security, integration, change management, and price.
6. Discovery script + close (5 minutes)Time-boxed questions and a mutual action plan to end the meeting with momentum.

Copy-paste master prompt (save as your Deal Copilot)

Use this once per deal. Replace items in brackets.

Prompt:

“You are my enterprise prep coach. Using ONLY the information I paste and what you can infer conservatively, produce a clear, concise plan. Company: [Account]. Industry: [Industry]. Attendees: [Names + titles]. Product: [Your product + 3 core capabilities]. Goal of meeting: [Discovery | Demo | Mixed]. Time: [30/45/60 mins]. Known constraints: [Tooling, budget, compliance].

Deliver in this order, bullet-first, plain English:

• Account brief: business model, 3 plausible priorities, 3 risks (cite which pasted line informed each).
• Stakeholder map: for each attendee, list goals, anxieties, success metric, and 3 tailored discovery questions.
• Value hypothesis: 3 statements that link our capability to their outcome. Include a simple metric and a 90-day win.
• Demo storyline (6 steps max): title each step, what to click/show, proof asset to reference, and the ‘wow’ moment.
• Objections: top 6 likely, with a one-sentence response and a question that advances the conversation.
• Discovery script (20 minutes): opening, 6–8 questions, and a 2-minute summary + next steps.
• Close: propose a mutual action plan with 3 milestones, owners, and dates.
• Constraints: list any assumptions or unknowns you need me to validate in the call.

Format tightly. Avoid generic fluff. If uncertain, offer 2 options and flag assumptions.”

Quick variants

• 10-minute sprint: “Summarize the top 3 buyer priorities from the pasted content, 5 discovery questions, and a 5-step demo click-path. Keep under 200 words.”
• Deep dive: “Build a MEDDICC-aligned discovery plan with required evidence for each element, and a risk pre-mortem with mitigations.”
• Security-first meeting: “Create a security FAQ with crisp answers (SOC, data flow, RBAC, SSO, audit logs), plus 3 questions to uncover their review process and timeline.”

Example (short)

• Scenario: Acme Retail evaluating a customer data platform. Attendees: VP Marketing, Head of Data, Procurement.
• Value hypothesis: “Consolidate customer profiles to lift email conversion by 15% in 90 days via clean segments and triggered journeys.”
• 3 discovery questions:
  • What revenue targets depend on improving repeat purchase or cross-sell this quarter?
  • Where do incomplete or duplicated profiles hurt campaign performance today?
  • Which systems must stay the source of truth, and who signs off on schema changes?
• Demo storyline (5 steps):
  • Problem framing: current fragmented profiles and impact on revenue.
  • Unify profiles: show merge rules on a sample account.
  • Segment build: create “lapsed high-value” in 60 seconds.
  • Trigger journey: publish to email tool; show test send.
  • Prove value: dashboard showing lift assumptions and a 90-day pilot plan.

Insider tricks

• Force citations: Ask the AI to tag each claim with the pasted line it came from. Cuts guesswork.
• Set hard limits: Cap bullets (e.g., “6 bullets max”) to avoid verbose output you cannot use live.
• Say-this-not-that: Add a small style guide: “Prefer numbers, avoid jargon, write to a CFO-level reader.”
• Pre-mortem prompt: “List the top 5 ways this meeting could fail and the counter-moves I should prepare.”

Common mistakes and quick fixes

• Generic talk track: Fix by pasting 5–10 lines of company context and forcing citations.
• Feature dumping: Convert features to outcomes with a metric and timeline (“reduces rework 30% in 90 days”).
• Too long demos: Limit to 5–7 steps; identify 2 wow moments and 1 risk-preemption slide.
• Hallucinations: Label assumptions, ask for two options, and validate live.
• Weak close: Always end with a mutual action plan request and a date.

After the call (copy-paste prompt)

“From these notes [paste], produce: 1) a MEDDICC-style summary with gaps; 2) a follow-up email with agreed outcomes, owners, and dates; 3) a risk list with mitigations; 4) the next meeting agenda request. Keep it to 200 words, bullet points only.”

Action plan for your next meeting

1. Paste 8–12 lines of public context into the master prompt; generate brief and stakeholder map.
2. Refine the value hypothesis with one metric and a 90-day win.
3. Build a 6-step demo storyline with two wow moments.
4. Print the 8-question discovery script; time-box to 20 minutes.
5. Prepare a one-slide mutual action plan template to fill live.
6. Run the pre-mortem prompt and patch gaps before you join.

Closing thought

AI won’t replace your judgment—it amplifies it. Use it to do the heavy lifting before you walk in, so you can spend the meeting doing what wins deals: listening, tailoring, and securing next steps.

Nice question — you’re on the right track. Predicting churn and firing timely “save” campaigns is one of the highest-impact uses of AI for revenue retention. Below I’ll walk you through a clear, practical path you can implement quickly, with a checklist and a copy-paste AI prompt.

Quick context: Use historical customer activity to predict the probability each customer will churn, then trigger tailored outreach (email, SMS, in-app) when risk passes a threshold. Start small, measure, iterate.

What you’ll need:

• Customer activity data (purchases, logins, sessions, last activity)
• Engagement metrics (email opens, clicks, NPS, support tickets)
• A labeled churn definition (e.g., no purchase or login in 90 days)
• Basic tooling: spreadsheet/SQL, simple ML model (AutoML or Python with scikit-learn), and your CRM/email tool for automation

Step-by-step:

1. Define churn: pick a clear rule (example: no purchase or login in 90 days).
2. Assemble features: recency, frequency, monetary (RFM), days since last login, support tickets, email_open_rate_30d.
3. Label your past customers using the churn rule to create a training set.
4. Train a simple model first (logistic regression or random forest). Use 70/30 train/test split and check AUC, precision/recall.
5. Pick a risk threshold for action (e.g., probability > 0.65 = high risk). Create buckets: low/medium/high.
6. Automate: when a customer enters high-risk bucket, trigger a save campaign in your CRM with personalized content.
7. Measure lift with an A/B test: control vs. targeted save campaign.

Practical example (worked):

• Dataset: 10,000 customers. Churn label = no purchase in 90 days.
• Model: random forest → AUC 0.82. Threshold 0.7 produces a high-risk group of 800 customers.
• Save campaign: send a personalized email with subject “We miss you — 20% off to come back” to high-risk group. Expected: 12% reactivation vs 4% control.

Checklist — do / do not:

• Do label churn clearly, run small tests, personalize offers, monitor metrics.
• Do not rely on one feature only, ignore data leakage, or blast every churn-risk the same way.

Common mistakes & fixes:

• Bad label definition → fix by testing several churn windows (30/60/90 days).
• Data leakage (using future info) → keep training features only from prior to label window.
• No measurement → run randomized control tests to prove impact.

Copy-paste AI prompt (use this with an LLM to help build features, SQL and playbook):

“Act as a marketing data scientist. Given a customer table with columns: customer_id, last_purchase_date, total_purchases, avg_order_value, last_login_date, support_tickets_90d, email_open_rate_30d. Provide: 1) SQL to compute recency, frequency, monetary, days_since_last_login; 2) a churn definition and how to label historical data; 3) a recommended modeling approach for ~10k rows and expected evaluation metrics; 4) a sample save-email template personalized with risk score and 1-line subject.”

Action plan (next 7 days):

1. Export 90 days of data and label churn candidates.
2. Create basic RFM features and train a quick model.
3. Define risk buckets, craft a single save email, and run an A/B test on the high-risk group.

Closing reminder: Start small, measure impact, and iterate. A simple model plus a well-timed personalized offer will often beat waiting for a perfect solution.

Quick win: Copy 8–10 recent support tickets into a chat with an AI and ask it to suggest 3 tags per ticket. You’ll see useful tags in under 5 minutes — enough to prove the approach.

Nice point to start from: you’re thinking about small teams, where simplicity beats complexity. That’s the right mindset—start small, measure, then expand.

Why this works: modern language models can read short ticket text, extract intent, and map to categories and tags. For small teams, the goal is not perfect automation but reliable assistance that saves time and reduces manual work.

What you’ll need

• Access to your ticket data (export or copy a sample).
• An AI tool (Chat-style LLM or built-in helpdesk AI) or an automation platform (Zapier/Make) if you want live tagging.
• A simple tag taxonomy (5–12 tags to start).
• A place to store tags (your helpdesk, spreadsheet, or CRM).

Step-by-step: set it up in one day

1. Define 8–12 tags you care about (e.g., Billing, Technical – Login, Feature Request, Refund, Shipping).
2. Quick test: pick 8–10 real tickets and run the AI prompt below to get tags and category suggestions.

   Expect: 70–90% sensible suggestions. Don’t trust it blindly—review.

3. Create a simple automation: when a ticket arrives, send the subject + first 200–400 characters to the AI, get tags back, and write them to the ticket fields.
4. Monitor for 1–2 weeks: sample 20 tagged tickets daily and log accuracy. Adjust prompts or tags where it fails.
5. Gradually add rules: fallback rules for very low-confidence predictions, and escalation for risky categories (security, legal, refunds).

Sample mapping (example)

• “I can’t log in after the update” → Category: Technical, Tags: Login Issue, Urgent
• “I was charged twice for my order” → Category: Billing, Tags: Duplicate Charge, Refund
• “Would love a CSV export of reports” → Category: Feature Request, Tags: Reporting, Product Idea

Common mistakes & fixes

• Mistake: Too many tags. Fix: Reduce to the top 8–12 and merge similar ones.
• Mistake: Trusting AI 100%. Fix: Add a human review step for low-confidence tags.
• Mistake: No monitoring. Fix: Sample accuracy weekly and refine prompts/taxonomy.

Copy-paste AI prompt (use as-is)

“You are a support categorization assistant. For each ticket below, return a short JSON list with: category (one of: Billing, Technical, Feature Request, Account, Shipping, Other), tags (max 3 tags from this list: Login, Payment, Refund, Bug, Setup, Reporting, Integration, Shipping, Performance, Cancellation, Feature Idea, Other), and confidence (low/medium/high). Ticket format: n[ticket id] – [ticket text]. Tickets:n1 – I can’t log in after the app updated and it keeps saying invalid password.n2 – I was billed twice for last month, please refund the duplicate.n3 – Is there a way to export reports to CSV?”

Action plan (next 7 days)

1. Today: pick tags and run the quick win test with 8–10 tickets.
2. Day 2–3: build simple automation to add tags to new tickets (or manually copy AI results into tickets).
3. Day 4–7: monitor accuracy, refine prompt/tags, set rules for low-confidence cases.

Keep it iterative. A small team that starts with a simple AI-assisted workflow will cut manual tagging time dramatically — then you can scale accuracy and automation as confidence grows.