The Revenue Engine
Farah closes at 340% of quota. She is not your most charismatic rep and she is not your longest-tenured. She is your most prepared. Before every first contact she spends 45 minutes reading Companies House filings, scanning LinkedIn activity, cross-referencing recent hires against technology signals, and building a list of five questions she already knows the answers to. She never opens with a pitch. She opens with a question that demonstrates she has done the homework the prospect expected nobody to do.
Your other three reps spend four minutes on research. They open LinkedIn, skim the headline, check the company size, and send a template. They average 60% of quota. The gap between Farah and the rest of your team is not talent, charisma, or experience. The gap is preparation time -- 45 minutes versus 4 minutes, multiplied across every prospect, every week. The maths is brutal: 500 accounts at 45 minutes each is 375 hours of research per cycle. Your team has 40 hours of combined research capacity per week. So 460 accounts get the four-minute version.
This chapter gives every rep Farah's research depth in under four minutes. The Revenue Engine is a coordinated set of Claude plugins -- Anthropic's base Sales and Marketing plugins plus the Agent Factory extension -- that produces the same structured intelligence Farah builds by hand. By the end of this lesson you will have installed all three plugins, generated a demo business dataset, run your first prospect research brief, and learned the single most important skill in working with AI sales tools: detecting when the agent is making things up.
Install All Three Plugins
If you have not already set up Cowork, follow the instructions in the Chapter 17 prerequisites first.
Install the three plugin layers from the Cowork sidebar:
- Sales plugin. In the Cowork sidebar: Customize → Browse plugins → find Sales (from
knowledge-work-plugins) → click Install. - Marketing plugin. In the Cowork sidebar: Customize → Browse plugins → find Marketing (from
knowledge-work-plugins) → click Install. - Sales RevOps Marketing plugin. In the Cowork sidebar: Customize → Browse plugins → Personal → click + → Add marketplace from GitHub → enter
https://github.com/panaversity/agentfactory-business-plugins→ find Sales RevOps Marketing → click Install.
The Sales plugin provides six skills and three commands (/call-summary, /forecast, /pipeline-review). The Marketing plugin provides five skills and seven commands (/draft-content, /campaign-plan, /brand-review, /competitive-brief, /performance-report, /seo-audit, /email-sequence). The extension adds fifteen skills that enhance both base plugins with business-specific configuration.
Verify everything is connected by running this prompt in Cowork:
List all my sales and marketing skills and tell me which
plugin each one comes from.
You should see skills from all three layers: six from the Sales plugin (including account-research and call-summary), five from Marketing (including content-creation and campaign-planning), and fifteen from the extension (including prospect-research, lead-scoring, persona-icp, and crm-enrichment). If you see only base-plugin skills without the extension additions, reinstall the Sales RevOps Marketing plugin from the Cowork sidebar.
Connect Your Tools (Optional)
The Revenue Engine works with prompt-provided context alone -- you describe the prospect, the agent researches. It becomes significantly more powerful when connected to your actual sales tools.
| Category | Recommended | What It Enables |
|---|---|---|
| CRM | HubSpot CRM (free tier) | Agent reads/writes your real pipeline from L04 onward |
| Gmail | Draft and review outreach sequences | |
| Calendar | Google Calendar | Pre-call brief timing and meeting prep |
| Optional | Slack | Team alerts for pipeline changes |
| Optional | Notion | Knowledge base for competitive intelligence |
| Optional | Apollo | Contact enrichment and signal monitoring |
If you have a free HubSpot account, the agent works with your real CRM data from Lesson 4 onward. If you do not have any of these accounts, skip this section entirely. Both paths produce the same quality output -- with connectors the agent reads live data, without them you provide context in prompts.
Generate Your Demo Business
NexaFlow Technologies is the company you will operate as throughout this chapter. It is a 38-person workflow automation startup in Karachi, selling to mid-market logistics companies across Pakistan, UAE, and the UK. Copy this prompt and run it in Cowork:
Generate a complete demo dataset for a company called NexaFlow Technologies,
a workflow automation platform for mid-market logistics companies based in
Karachi, Pakistan (38 employees, PKR 180M revenue).
Generate the following:
1. CLOSED-WON DEALS (20 records)
For each: company name, industry, employee count, revenue, location,
buyer persona (name + title), trigger event, sales cycle length,
primary pain, deal value, close date (last 18 months)
Mix: 12 Pakistan, 5 UAE, 3 UK
2. TARGET PROSPECTS (5 records)
For each: company name, location, employee count, revenue estimate,
industry, key contact (name + title + background), recent news/signals,
LinkedIn activity, tech stack signals
Include: Meridian Logistics (Leeds, UK, Sarah Chen, VP Ops) as prospect #1
Mix: 2 Pakistan, 1 UAE, 2 UK
3. CURRENT PIPELINE (10 deals)
For each: company name, deal value, stage (Discovery/Demo/Proposal/
Negotiation/Closed-Won), assigned rep, last activity date, close date
target, notes
Include: 3 at-risk (no activity >14 days), 2 likely to close this quarter
4. CAMPAIGN HISTORY (Q1 results)
Channels: LinkedIn (organic + sponsored), email nurture, trade press
For each: impressions, clicks, CTR, conversions, cost, leads generated,
HOT leads, cost per lead
5. COMPETITOR INTELLIGENCE
3 competitors: name, positioning, strengths, weaknesses, where they win,
where NexaFlow wins
Format: structured markdown with clear section headers.
Save the output as demo-data.md in my working folder.
Output: A structured dataset saved as demo-data.md containing 20 closed-won deals, 5 target prospects (with Meridian Logistics as prospect #1), 10 pipeline deals, Q1 campaign results, and competitor intelligence. Save this file -- every subsequent lesson references this data.
Set Up Folder Instructions
Every lesson in this chapter assumes the agent knows who NexaFlow is, where the demo data lives, and what voice to use. Instead of pasting this context into every prompt, set it once in Cowork's folder instructions.
Open the Cowork sidebar → Instructions pane. Paste this block and save:
You are working as a sales and marketing assistant for NexaFlow Technologies,
a workflow automation platform for mid-market logistics companies based in
Karachi, Pakistan (38 employees, PKR 180M revenue).
Always read demo-data.md for prospect records, pipeline data, campaign history,
and competitor intelligence. Always read sales-marketing.local.md for our ICP
definition, company profile, and brand voice.
When generating outreach, research briefs, or content, apply our brand voice:
direct, practical, no jargon — "We speak like operators, not vendors."
Our ICP: 50-400 employee regional 3PL operators, warehousing, or fleet management
companies. Buyer persona: VP Ops / Director Ops. Key markets: Pakistan, UAE, UK.
Once saved, the agent reads demo-data.md and sales-marketing.local.md automatically — you do not need to reference these files in every prompt. Every lesson from L02 onward assumes these instructions are active. If you start a new Cowork session, re-paste them.
Your First Research Brief
You are NexaFlow Technologies. Your top prospect is Meridian Logistics in Leeds — prospect #1 in the demo dataset you just generated.
Run this prompt. It activates the prospect-research and persona-icp skills from the extension to produce a structured brief with ICP scoring:
Read demo-data.md for my company profile (NexaFlow Technologies)
and the Meridian Logistics prospect record.
Research Sarah Chen, VP Operations at Meridian Logistics, Leeds.
I sell workflow automation for logistics companies. Generate a full
prospect research brief with ICP match scoring, timing signals,
and a recommended outreach angle.
If your output is a plain text response without structured sections (ICP MATCH, WHO, WHAT, etc.), the research skill did not activate. Try prefixing your prompt with /prospect-research to invoke it directly — you will learn more about explicit skill invocation in Lesson 9.
What to expect: The agent reads your demo data and produces a structured brief. The exact content varies between runs, but look for these sections:
| Section | What It Contains |
|---|---|
| ICP MATCH | Scores Meridian against your ICP criteria (company size, industry, buyer persona). If the extension is active, this section appears at the top. |
| WHO | Sarah Chen's background — sourced from your demo dataset, potentially embellished |
| WHAT | Company profile for Meridian Logistics |
| WHEN | Timing signals (promotion, contract win, LinkedIn activity) |
| PAIN | Primary pain points the agent identifies for this prospect |
| HOOK | Recommended opening angle based on pain + timing |
The agent combines your demo data with web research. Since Meridian Logistics is fictional, the web research returns nothing useful — the agent fills gaps from your dataset and its own inferences. This is exactly the setup for the next section.
Read the full output. It is structured, specific, and professional. It looks like something Farah would produce after 45 minutes of deep research.
And at least one claim in it will go beyond your demo data.
Hallucination Detection
Open demo-data.md and read the Meridian Logistics prospect record. Now compare it to the agent's brief, claim by claim.
Your dataset contains specific facts you generated: company name, location, employee count, contact name and title, recent signals. The agent's brief includes these — but it also includes claims that go beyond your data. Revenue estimates, career history embellishments, internal process descriptions, LinkedIn post specifics — details the agent inferred to make the brief sound authoritative.
Since Meridian Logistics is fictional, your demo-data.md is the only source of truth. Any claim not traceable to that file is an agent embellishment. The skill you need before every subsequent lesson is the ability to tell them apart before you act on them.
Claims From Your Data
| Claim Type | Example | Why Trustworthy |
|---|---|---|
| Company name, location, employees | Directly from your prospect record | You generated this data — it is your ground truth |
| Contact name and title | Sarah Chen, VP Operations | Specified in the generation prompt |
| Recent signals you specified | Contract win, scaling challenges | You defined these as part of the prospect profile |
These claims are your ground truth. In a real sales workflow, the equivalent is data from your CRM, LinkedIn, or Companies House — sources you can independently verify. Verified claims become conversation openers: "I noticed you are scaling rapidly after a major contract win — are you rebuilding your operations workflow?"
Claims Beyond Your Data
| Claim Type | Why Suspect | The Tell |
|---|---|---|
| Revenue or financial estimates | Your demo data may not include revenue, or the agent adjusted the figure to sound precise | Private financials stated as estimates. Revenue for private companies is not public unless voluntarily disclosed. |
| Internal processes (e.g. "SLA tracking via spreadsheets") | No source — not in your demo data, not discoverable via web search for a fictional company | Internal process presented as intelligence. The agent inferred a common pain pattern for logistics companies of this size. |
| Specific career timeline or LinkedIn post content | Your demo data has a brief background; the agent may add employers, dates, or post topics it cannot have found | Inference dressed as observation. The agent expanded sparse data into a plausible, specific narrative. |
Look for these in your own output. Identify at least two claims that are not in demo-data.md. For each one, decide: is it a reasonable inference, or is it presented as fact?
The Three Rules
These three rules catch most hallucinated data in prospect research briefs:
Rule 1: Private financials are always suspect. Revenue, ARR, burn rate, runway, unit economics -- if the company has not publicly disclosed the number, the agent fabricated it. This applies to funding round details, valuation estimates, and growth rates unless sourced from a press release or filing.
Rule 2: The more specific the unverifiable claim, the more likely it is hallucinated. "£45-60M estimated revenue" is more suspicious than "mid-market company." "3 posts about operational scaling" is more suspicious than "active on LinkedIn." Agents generate specific numbers because specificity sounds authoritative. Specificity without a source is a red flag.
Rule 3: Inferred connections are not confirmed connections. "SLA compliance via spreadsheets" sounds like inside knowledge. It is a pattern match -- logistics companies at this scale commonly use spreadsheets for compliance tracking. The agent noticed the company size, industry, and growth trajectory and generated a plausible internal process. Treat any internal operations claim as a hypothesis until you confirm it in conversation.
Walking into a meeting and citing a revenue figure that does not exist destroys your credibility instantly. The prospect concludes you fabricate research. Every accurate insight you share afterward is tainted.
The rule is straightforward: verify every claim you plan to say out loud. If you cannot verify it, do not say it. Verified claims become conversation openers. Unverified claims stay in your notes as hypotheses to test during the meeting.
The Verification Hierarchy
Not all claims carry equal risk. Use this hierarchy to prioritise what to verify before a meeting:
HIGHEST CONFIDENCE
│
│ Cited sources (named article, named event, press release)
│ Public records (Companies House, LinkedIn Jobs, patent filings)
│ Public events (conference talks, published interviews)
│
│ ─── Verification line: above = verify easily, below = treat with caution ───
│
│ Career history (LinkedIn profiles, privacy-dependent)
│ Financial estimates (inferred from benchmarks)
│ Internal processes (inferred from industry patterns)
│ Business relationships (inferred from market proximity)
│
LOWEST CONFIDENCE
Configure sales-marketing.local.md
The extension includes a local configuration template. In your connected working folder, create a new file called sales-marketing.local.md and paste the NexaFlow ICP skeleton:
# Sales & Marketing Local Configuration
## Company Profile
- **Company:** NexaFlow Technologies (Pvt) Ltd
- **Product:** Workflow automation platform for mid-market logistics companies
- **Market:** B2B SaaS — warehouse ops, fleet coordination, SLA tracking
- **Geography:** Pakistan (70%), UAE (20%), UK expansion (10%)
## Ideal Customer Profile (ICP)
- **Company size:** 50-400 employees
- **Industry:** Regional 3PL operators, warehousing, fleet management
- **Buyer persona:** VP Ops / Director Ops
- **Pain points:** Manual coordination at scale, SLA tracking via spreadsheets,
slow staff onboarding, scaling without headcount
- **Buying signals:** New contract wins, VP/Director hired in last 12 months,
LinkedIn posts about scaling challenges, open operations roles
## Brand Voice
- Direct, practical, no jargon. "We speak like operators, not vendors."
Save the file and run the same research prompt again:
Read demo-data.md for the Meridian Logistics prospect record.
Research Sarah Chen, VP Operations at Meridian Logistics, Leeds.
I sell workflow automation for logistics companies. Generate a full
prospect research brief with ICP match scoring.
Compare the output to the first brief. The ICP MATCH section now scores Meridian against NexaFlow's specific profile -- 50-400 employees, 3PL operators, VP Ops persona, scaling pain. The HOOK section references NexaFlow's value proposition instead of generic automation language. The recommended approach filters through your brand voice: direct, practical, no jargon.
This skeleton will be completed in Lesson 2 when you analyse your closed-won deals to build a data-driven ICP rather than an assumed one.
What You Built
- All three plugins installed and verified -- Sales, Marketing, and the RevOps extension
- Demo business dataset (NexaFlow Technologies) generated with 20 closed-won deals, 5 target prospects, 10 pipeline deals, campaign history, and competitor intelligence
- First prospect research brief on Meridian Logistics / Sarah Chen with structured WHO / WHAT / WHEN / PAIN / HOOK sections
- Hallucination detection skill -- you can distinguish claims traceable to your demo data from agent embellishments (revenue estimates, internal process descriptions, career timeline expansions)
- sales-marketing.local.md skeleton configured with NexaFlow's ICP
Flashcards Study Aid
Test your understanding of the key concepts from this lesson.
Try With AI
Setup: Use these prompts in Cowork or your preferred AI assistant with the Sales, Marketing, and RevOps extension plugins installed.
Prompt 1: Reproduce
Read demo-data.md for my company profile (NexaFlow Technologies)
and the Meridian Logistics prospect record.
Research Sarah Chen, VP Operations at Meridian Logistics, Leeds.
I sell workflow automation for logistics companies. Generate a full
prospect research brief with ICP match scoring, timing signals,
and a recommended outreach angle.
What you're learning: How the prospect-research and persona-icp skills structure intelligence into actionable sections (ICP MATCH / WHO / WHAT / WHEN / PAIN / HOOK). Run this prompt twice and compare the two outputs. The structure should match but specific claims will differ between runs — the agent produces different embellishments each time, which is itself a lesson in why verification matters.
Prompt 2: Adapt
Research [pick a second prospect from your generated demo dataset].
Compare the research brief to the Meridian brief. Which prospect
has stronger timing signals? Which has more verifiable data?
What you're learning: Research briefs vary in quality based on how much public information exists. Data-scarce prospects (smaller companies, emerging markets, private companies) produce more hallucinations than data-rich prospects (UK companies with Companies House filings, active LinkedIn presences, press coverage). Recognising this pattern helps you calibrate trust per brief rather than trusting all briefs equally.
Prompt 3: Apply
Research a real prospect from your own network — someone you have
been meaning to contact. After reading the brief, mark every claim as
VERIFIED (you can confirm it), PLAUSIBLE (likely true, not confirmed),
or SUSPECT (cannot verify, possibly hallucinated).
What you're learning: The discipline of evaluating agent output before acting on it. This is the foundational skill for every subsequent lesson in this chapter. A rep who sends outreach referencing a fabricated funding round loses credibility permanently. A rep who verifies first and leads with confirmed intelligence earns trust immediately. Your VERIFIED/PLAUSIBLE/SUSPECT audit becomes a habit you apply to every research brief from this point forward.