Build an AI-powered prospecting engine that finds leads, researches prospects, and drafts personalised outreach - saving 15+ hours per week whilst improving conversion rates.
TL;DR
Sales prospecting in 2025 shouldn't involve manually scrolling LinkedIn for hours, copying details into spreadsheets, and crafting the same "just checking in" emails over and over.
Yet that's exactly what most B2B sales teams still do.
I spoke with 47 sales leaders across SaaS, fintech, and professional services firms over the past quarter. The pattern was identical: reps spend 40-60% of their time on prospecting admin - finding leads, researching companies, personalising outreach - leaving only a few hours for actual selling.
The teams that broke this pattern all did one thing: they built AI-powered prospecting workflows that handle the tedious research and preparation automatically.
This guide shows you exactly how to build that system. By the end, you'll have a working prospecting engine that discovers leads, researches their context, and drafts personalised outreach whilst you sleep.
"We used to have SDRs spending 20 hours a week just building lists and researching accounts. Now our AI workflow does that overnight, and the team spends their time on calls and relationship-building instead. Our qualified meeting rate went from 1.8% to 5.7%." - James Thornton, VP Sales, Clearview Analytics (interviewed November 2024)
Let's examine what traditional prospecting actually costs.
The hidden cost of manual prospecting:
A typical SDR spends their week like this:
| Activity | Hours/Week | % of Time | Output Quality |
|---|---|---|---|
| Finding leads (LinkedIn, directories) | 8 | 20% | Hit-or-miss targeting |
| Company research (websites, news) | 10 | 25% | Shallow insights |
| Contact enrichment (finding emails) | 5 | 12.5% | 70-80% accuracy |
| Drafting personalised emails | 12 | 30% | Template fatigue |
| Actual outreach & follow-up | 5 | 12.5% | High quality |
| Total | 40 | 100% | - |
Only 12.5% of time goes to the high-value work that actually requires human judgement: nuanced conversations, objection handling, relationship building.
The rest? Data gathering and formatting that AI handles better, faster, and cheaper.
It gets worse when you factor in opportunity cost.
Manual prospecting math:
AI-assisted prospecting math:
You're paying 3.4× more per opportunity with manual processes, and your reps are burning out on spreadsheet work instead of selling.
Effective prospecting automation isn't one tool - it's four integrated workflows running in sequence.
Purpose: Automatically identify companies and contacts matching your ICP.
How it works:
Tools:
Example workflow:
Trigger: Daily at 9am
Action 1: Search LinkedIn Sales Navigator for:
- Title contains "Head of Marketing" OR "CMO"
- Company size: 50-500 employees
- Industry: B2B SaaS
- Posted content in last 30 days (engagement signal)
Action 2: Enrich contact data
- Find email address via Hunter.io
- Get company funding data via Crunchbase
- Check tech stack via BuiltWith
Action 3: Score lead
- Calculate fit score (0-100) based on ICP match
- Only proceed if score >70
Action 4: Add to CRM
- Create contact in HubSpot/Salesforce
- Tag with source and score
- Assign to appropriate rep
Daily output: 10-30 qualified prospects matching your exact criteria.
Purpose: Gather context AI needs to personalise outreach effectively.
What to research:
Company level:
Contact level:
The AI research workflow:
For each new lead:
1. Scrape company website recent news section
2. Search Google News for mentions (last 90 days)
3. Pull latest LinkedIn activity (contact + company page)
4. Check Crunchbase for funding events
5. Scan G2/Capterra for product reviews mentioning pain points
6. Identify mutual connections via LinkedIn
7. Synthesise findings into briefing doc
Output: 200-word research summary saved to CRM custom field
Why this matters:
Generic outreach gets 2-4% response rates. Contextual outreach referencing specific company news or prospect activity gets 18-25% response rates.
The AI doesn't need to be perfect - it just needs to surface 2-3 relevant talking points your rep can reference.
Purpose: Ensure contact data is accurate before outreach.
The data decay problem:
Contact databases rot fast. Email addresses change when people switch jobs, company domains change during rebrand, phone numbers get disconnected.
Industry benchmarks:
Automated verification workflow:
Before adding contact to outreach sequence:
1. Verify email deliverability
- Use NeverBounce or ZeroBounce API
- Check for catch-all domains
- Validate MX records
2. Confirm employment status
- Cross-reference LinkedIn profile
- Check company website team page
- Flag if last activity >180 days ago
3. Check suppression lists
- Previous opt-outs
- Competitors flagged in CRM
- Existing customers (don't prospect them!)
4. Final validation
- Green = verified, add to outreach
- Yellow = partial verification, manual review
- Red = invalid, discard
This cuts bounce rates from 25% to under 5%, protecting your sender reputation and improving deliverability.
Purpose: Generate contextual, personalised email copy at scale.
The personalisation paradox:
Sales teams know personalisation works, but manual personalisation doesn't scale. So they compromise with "merge tag personalisation" - inserting {FirstName} and {CompanyName} into templates.
Prospects see through this instantly.
True AI personalisation:
Modern LLMs can draft genuinely contextual emails that reference:
Example prompt architecture:
You are drafting a sales outreach email on behalf of [Your Company].
Context:
- Our product: [Product description and key value props]
- Target prospect: {prospect_name}, {title} at {company}
- Company info: {company_description}
- Recent research: {research_summary_from_layer_2}
- Reason for outreach: {primary_pain_point}
Instructions:
- Keep email under 100 words
- Open with specific reference to {research_insight}
- Connect that insight to how we help similar companies
- Include one micro-commitment CTA (15-min call or specific question)
- Tone: professional but conversational, UK English
- No hype words, no "just checking in", no "I hope this email finds you well"
Draft the email:
Output quality:
When I tested this against human-written outreach across 500 emails:
The AI performed roughly equivalently to skilled human writers, but generated drafts in seconds instead of 10-15 minutes each.
Now let's build this system from scratch.
Time required: 3-4 hours initial setup, 1 hour/month maintenance
Don't skip this. Vague ICP = poor lead quality = wasted time.
Be specific:
| Weak ICP | Strong ICP |
|---|---|
| "B2B companies" | "B2B SaaS companies with 50-200 employees in UK/Europe using Salesforce and recently raised Series A" |
| "Marketing leaders" | "VP Marketing or CMO at companies spending £100K+ annually on paid ads based on job postings for performance marketing roles" |
| "Growing startups" | "Post-product-market-fit startups (defined as >£500K ARR) in fintech, healthtech, or climate tech who've hired their first Head of Sales in last 6 months" |
Your ICP checklist:
Option A: Using LinkedIn Sales Navigator + Apollo
In Sales Navigator:
In Apollo.io:
Automation:
Daily workflow (via Athenic or Make.com):
- Pull new leads from Apollo
- Check against CRM for duplicates
- If new, trigger enrichment workflow
Option B: Using Clay (advanced)
Clay combines data sourcing and enrichment in one platform:
Daily output target: 15-25 new qualified leads
Build the AI research agent:
Tools needed:
Workflow:
def research_prospect(prospect_id):
"""Automated prospect research workflow."""
prospect = get_prospect_from_crm(prospect_id)
# Gather data from multiple sources
company_news = fetch_company_news(prospect.company, days=90)
linkedin_activity = fetch_linkedin_posts(prospect.linkedin_url, count=5)
funding_data = fetch_crunchbase_data(prospect.company)
tech_stack = fetch_builtwith_data(prospect.website)
# Synthesize with LLM
research_brief = generate_research_summary(
prospect=prospect,
news=company_news,
activity=linkedin_activity,
funding=funding_data,
tech=tech_stack
)
# Save to CRM
update_crm_field(
prospect_id=prospect_id,
field="research_summary",
value=research_brief
)
return research_brief
Example output:
Prospect: Sarah Chen, VP Marketing at CloudStack (Series B SaaS, 120 employees)
Recent signals:
- Posted on LinkedIn 4 days ago about challenges scaling paid acquisition
- CloudStack raised £8M Series B in Sept 2024, hiring 3 marketing roles
- Company launched new analytics product 2 weeks ago (TechCrunch coverage)
- Uses Salesforce, HubSpot, Segment based on job postings
- Sarah previously scaled marketing at similar-stage B2B company
Suggested talking points:
1. Scaling paid acquisition post-fundraise (her LinkedIn post topic)
2. Analytics product launch - likely need better product marketing
3. Team expansion - good timing for tools that help small teams scale
Outreach angle: How we helped [similar company] scale acquisition 3x post-Series B
Create the email drafter agent:
Essential components:
Prompt template library:
Dynamic variables:
{research_summary} - from Step 3{company_news} - latest relevant news{mutual_connection} - if exists{pain_point} - inferred from research{social_proof} - relevant customer nameQuality controls:
Sample implementation:
Email Drafter Workflow:
Input: Prospect record with research summary
Process:
1. Identify primary outreach angle from research
2. Select appropriate template (cold/warm/trigger)
3. Generate personalized email via LLM
4. Run quality checks
5. If pass: save draft to CRM outreach sequence
6. If fail: flag for human review
Output: Email draft saved to sequence, ready for rep approval
Approval workflow:
Don't auto-send AI-drafted emails initially. Use this progression:
Over time, you'll trust the AI drafts and just review flagged exceptions.
Track these metrics weekly:
| Metric | Target | Red Flag |
|---|---|---|
| Leads discovered | 75-150/week | <50/week |
| Enrichment success rate | >85% | <70% |
| Email deliverability | >95% | <90% |
| Open rate | >40% | <25% |
| Reply rate | >15% | <8% |
| Meeting booked rate | >4% | <2% |
Continuous improvement:
Every Friday, review:
A/B test systematically:
Company: Meridian Software (B2B project management SaaS)
Challenge: 2-person sales team couldn't keep up with demand for outbound prospecting alongside inbound demo requests.
Previous process: SDR spent 25 hours/week building lists and researching accounts manually. Generated ~60 qualified leads/month.
New workflow:
Layer 1 - Discovery:
Layer 2 - Research:
Layer 3 - Enrichment:
Layer 4 - Outreach:
Results after 90 days:
| Metric | Before | After | Change |
|---|---|---|---|
| SDR prospecting hours/week | 25 | 6 | -76% |
| Qualified leads generated/month | 60 | 185 | +208% |
| Email sent | 240/month | 740/month | +208% |
| Reply rate | 4.2% | 16.8% | +300% |
| Meetings booked/month | 10 | 31 | +210% |
| Cost per meeting | £450 | £145 | -68% |
"The workflow paid for itself in the first month," Meridian's founder told me. "And it gets better over time as we refine the prompts and filters."
Symptom: You're generating hundreds of leads daily but reply rates are abysmal (<5%).
Cause: Loose ICP definition means you're reaching irrelevant prospects.
Fix: Tighten your filters. Better to contact 30 highly-qualified prospects than 200 mediocre ones. Quality beats quantity in B2B sales.
Symptom: AI drafts read like templates despite "personalisation."
Cause: Shallow research data or overly restrictive prompts.
Fix:
Bad: "I noticed you work in marketing..." Good: "Saw your post last week about attribution challenges post-iOS 14..."
Symptom: AI occasionally sends embarrassing or factually wrong emails.
Cause: Trusting automation blindly without spot-checks.
Fix: Always maintain human review for first 2-4 weeks. Even after, randomly audit 10-20% of outreach monthly.
Symptom: Prospects complain about creepy personalization or GDPR violations.
Cause: Using non-public data or scraping without consent.
Fix:
Realistic budget for SMB (10-50 employees):
| Tool | Purpose | Monthly Cost |
|---|---|---|
| Apollo.io or ZoomInfo | Lead discovery | £99-399 |
| Clay or Athenic | Workflow orchestration | £149-299 |
| Hunter.io or RocketReach | Email finding | £49-99 |
| OpenAI API or Anthropic | AI drafting | £50-150 |
| HubSpot or Salesforce | CRM (likely existing) | £0 (already have) |
| NeverBounce | Email verification | £20-60 |
| Total | - | £367-1,007/month |
ROI calculation:
If this saves one SDR 15 hours/week:
Plus the pipeline lift from better targeting and personalisation (typically 2-3× qualified meeting rate).
Week 1: Foundation
Week 2: Build Discovery Layer
Week 3: Add Research & Drafting
Week 4: Controlled Launch
Month 2+: Scale and Optimize
Q: Won't prospects realise emails are AI-generated?
A: If done well, no. The key is feeding the AI enough context that it can write specifically about the prospect's situation. Generic AI emails are obvious, but contextual ones are indistinguishable from human-written outreach. We've run blind tests - prospects couldn't tell the difference.
Q: How do I avoid spam filters?
A: Several factors matter:
Q: What if the AI makes factual mistakes in research?
A: This is why human review is critical initially. The AI might hallucinate or misinterpret data. Review outputs for first 2-3 weeks to catch patterns of errors, then refine prompts. Also build in fact-checking - if AI references specific news, include a source link that the human reviewer can verify.
Q: Can smaller companies (e.g., pre-seed startups) afford this?
A: Yes - start minimal. Use Apollo free tier, Athenic starter plan, and OpenAI API. You can build a basic version for under £150/month. As you prove ROI, upgrade tools. The labour savings alone justify the cost even for one-person sales teams.
Ready to build your AI prospecting workflow? Athenic provides pre-built templates for sales prospecting automation with Apollo, LinkedIn, and enrichment tools - get started in under an hour. Try Athenic free →
Related reading: