Liam Paschall Sales Enablement & Learning Design · Case study
AI-assisted sales onboarding

A self-serve onboarding hub for new sales reps, built in NotebookLM

I designed a complete onboarding source pack for a fictional software company, loaded it into Google NotebookLM, and turned it into cited Q&A, an audio briefing, flashcards, and a Day 30 certification quiz. Then I audited every output for accuracy.

Demo Crewlane and its competitors are fictional. No employer or client material was used.
8source documents written for the hub
56flashcards, each checked against the sources
9question Day 30 certification quiz
24minute audio briefing for Week 1
The problem

New reps can't find answers, so managers repeat themselves

Small sales teams rarely have an LMS or an enablement platform. Product sheets, pricing rules, and objection talk tracks live in scattered decks and shared drives. New reps either guess or interrupt their manager. I wanted to show how a team with no enablement budget can give reps one place to ask "how do we sell this?" and get an answer that points back to the official source.

What I built

Source pack in, onboarding assets out

  1. Step 1 · Sources

    Write the source pack

    • Company overview
    • 3 product sheets
    • Pricing and discount FAQ
    • Objection guide (ACE method)
    • Competitor snapshot
    • Ideal customer and discovery guide
  2. Step 2 · Notebook

    Load and test it

    • All 8 PDFs in one notebook
    • Test chat with real rep questions
    • Check every answer's citations
    • Stress-test the pricing math
  3. Step 3 · Assets

    Generate and audit

    • Deep Dive audio briefing
    • 56 flashcards
    • Day 30 certification quiz
    • QA pass on every output
Sample output 1 · Objection practice

Rep asks, the notebook answers with citations

A prospect says our techs won't use an app. How should I respond?

Handle the objection using Crewlane's ACE method (Acknowledge, Clarify, Evidence)1:

  1. Acknowledge: validate the concern without arguing1. Acknowledge that field adoption is a top concern for many owners.
  2. Clarify: "What happened the last time you tried a new tool with them?"2
  3. Evidence: the app takes about 20 minutes to learn2, and 94% of technicians are active within two weeks2.
  4. Offer the 30-day pilot3 with their least tech-friendly technician2.

The highlighted line has no citation. It doesn't appear in any source; NotebookLM added it to smooth the answer. Teaching reps to trust cited lines and verify uncited ones became part of the hub's QA checklist.

Sample output 2 · Certification quiz

Wrong answers built from real mistakes

Day 30 Certification Quiz · Pricing

An AE quotes an 8-technician contractor on Pro ($89 per tech per month) with annual prepayment (15% discount) and the standard $750 setup fee. What is the total upfront cost?

  • A$9,294.00Forgot the 15% annual discount
  • B$7,262.40Left out the setup fee
  • C$7,899.90Discounted the setup fee too
  • D$8,012.40Correct
8 × $89 × 12 = $8,544.00 → less 15% = $7,262.40 → + $750 setup = $8,012.40
Quality control

What the AI got wrong, and how I caught it

AI-generated training content looks finished before it's correct. I reviewed all 56 flashcards and every quiz question against the sources. These are the issues that mattered.

Math error

A quiz explanation did the arithmetic wrong

The right answer was correct, but one wrong answer's explanation claimed $10,680 minus one month's fee equals $9,930. It's $9,790.

Regenerated the quiz and re-checked every number by hand.

Ambiguity

A question had two defensible answers

"A 15% discount" could mean the standard annual discount or an extra discount needing approval.

Required "an extra X% off" wording, and rewrote the pricing FAQ to state that the AE's 10% stacks on the 15% annual discount.

Invented detail

A question added an age stereotype

One objection question became "My techs are older and won't use an app." Nothing in the sources mentions age.

Deleted the question. Training content shouldn't teach reps to stereotype customers.

Too easy

The first quiz could be passed by guessing

The right answer was usually the longest, most specific option, and several topics repeated.

Rewrote the generation instructions (below) and compared three versions.

Iteration

Three quiz versions, one instruction at a time

Version 1 · Default

Easy to guess

Correct, but obvious answers and a math error in one explanation.

Version 2 · Hard

Real distractors

Believable wrong answers. One question had two defensible answers.

Version 3 · Shipped

Certification-ready

Clear discount wording, new topic coverage, audited and trimmed to 9.

The instruction that made the difference:

10 questions a sales manager would use to certify a new Account Executive at Day 30. Make every wrong answer the same length and as believable as the right one. Don't repeat topics. Cover products, pricing math, discount rules, objections, competitors, and discovery.
What I'd tell an enablement team

Lessons from the build

  • The sources are the product. Consistent numbers across every document are what make the answers and quizzes reliable.
  • Audit the explanations, not just the answers. The math error was hiding in feedback text a rep would trust.
  • Instructions shape quality. One paragraph of guidance turned an easy quiz into a real assessment.
  • It complements an LMS, it doesn't replace one. NotebookLM doesn't track completions. Pair it with live role-plays and a pitch certification.
Try it yourself

Ask the Crewlane notebook a question

Open the live notebook to chat with it, play the audio briefing, and take the certification quiz. Google requires you to sign in with any Google account to view it.

Open the notebook
About me

Liam Paschall

I build sales onboarding and enablement programs. For more than 10 years, I've designed and delivered training at Adobe, Cisco, Parexel, and Cardinal Health, including Microsoft 365 Copilot training for 500+ employees.

  • 2× Brandon Hall Gold Award contributor
  • Microsoft 365 Copilot certifications
  • Google Prompting Essentials
  • Stanford Advanced Program Management
LinkedIn profile