AI Lead Generation for Beginners: Find & Qualify Prospects Fast

A lead is a person or company that could plausibly buy what you sell, and for whom you have enough information to take a responsible next step. That “next step” might be a first message, a phone call, an invite to a webinar, or a request for an introduction. The key is plausibility plus actionability.

A lead is not just a name in a directory. It’s not a random LinkedIn profile you bookmarked. It’s not “anyone with a job title that sounds right.” Those are raw candidates. Treating raw candidates as leads is how beginners end up blasting messages to the wrong people, damaging their reputation, and wasting days in follow-ups that were never likely to convert.

To keep it simple, define three fields that must exist before something earns the label “lead” in your system:

• Identity: a person or a company you can point to (name + company, or company + role if name is unknown).
• Fit signal: at least one reason they might need you (industry, size, tool they use, hiring pattern, public pain point, etc.).
• Reach path: one way to contact them ethically (email you’re allowed to use, contact form, LinkedIn, referral path).

This is your first piece of engineering judgment: set the bar high enough that you don’t store junk, but low enough that you can move quickly. A common mistake is demanding perfection (full address, direct phone, personal email) before saving anything. That slows you down. Save leads early, then enrich later as needed.

AI is useful here because it can turn your “fit signals” into consistent notes. But the decision boundary—what counts as a lead—should be yours and should match your sales motion. If your offer is $99/month self-serve, your lead definition can be lighter than if you sell $50k/year contracts.

People use these words inconsistently, which creates confusion inside teams and inside your own workflow. Here is a beginner-friendly set of definitions you can reuse throughout the course:

• Prospect: a target you might pursue, but you have not verified fit or contactability yet. Prospects live in the “maybe” pile.
• Lead: a prospect with enough fit and contact information to justify outreach (as defined in Section 1.1).
• Qualified lead: a lead that meets your basic qualification rules (for example: correct industry, minimum size, and a plausible need).
• Opportunity: a qualified lead that has engaged and is now in an active sales conversation (reply, meeting, demo, inbound request, etc.).

This creates a clean lead journey you can map on one page: Stranger → Prospect → Lead → Qualified Lead → Opportunity. In practice, you will move people forward with small actions: research, enrich, message, follow up, schedule.

What matters is not the labels themselves, but that each stage has a clear “entry rule.” Example entry rules you can start with:

• Prospect entry: company appears in your target market list.
• Lead entry: you have a decision-maker role identified and one ethical contact route.
• Qualified lead entry: meets 2–3 ICP rules (industry + size + trigger).
• Opportunity entry: they respond positively or accept a meeting.

A common mistake is skipping stages: treating strangers like opportunities (“Can we hop on a call?”) or treating opportunities like leads (sending generic introductions after they already showed interest). Your message quality improves when you match the ask to the stage.

AI fits best in the transitions: turning strangers into prospects faster (finding candidates), prospects into leads faster (summarizing and extracting fields), and leads into qualified leads faster (scoring against simple rules). AI does not replace the need for you to define the stages.

In this course, think of AI as a fast assistant for three jobs: research, writing, and sorting. If you ask it to “do lead generation,” it will often produce vague lists. If you give it structured tasks, it becomes extremely practical.

Research: AI can help you brainstorm lead sources and targeted search ideas, especially when you provide your ICP in plain language. A strong prompt pattern is: “I sell X to Y because Z. Give me places where Y hangs out and keywords that indicate Z.” You can then use those ideas in a browser to find real prospects. AI is not the web—treat it as a planner for your searches, not the source of truth.

Writing: AI can draft first-touch messages and follow-ups in your voice if you give it constraints: who you are, who they are, why you’re reaching out, what you’re asking for, and what you must avoid (overpromises, sensitive claims, aggressive tone). You’ll still review for accuracy and compliance, but you’ll avoid staring at a blank page.

Sorting: Beginners often keep “messy lead notes” in screenshots, tabs, and long copy-pasted paragraphs. AI is excellent at turning that mess into clean contact fields without needing a spreadsheet first. For example, paste a block of notes and ask: “Extract company, contact name, role, website, location, evidence of fit, and suggested next step. Return as a simple table.” You can store the result in a notes app, a lightweight CRM, or even a document—what matters is consistency.

Good judgment: never let AI invent facts to fill missing fields. Tell it explicitly: “If unknown, leave blank.” Another common mistake is using AI to write messages before you’ve decided what qualifies a lead. If you write to everyone, you will drown in low-quality outreach. Research and sorting come first; writing comes after you know who deserves attention.

AI lead generation fails in predictable ways. If you know the risks upfront, you can move fast without creating a mess you’ll regret later. Keep this short risk list visible as you work.

Risk 1: Wrong information. AI can confidently output incorrect details (wrong job title, wrong company size, invented tech stack). The fix is process-based: require a “proof link” for key fields when possible (company website, LinkedIn page, job post, pricing page). If you can’t verify, store it as an unverified note, not as a fact.

Risk 2: Privacy and compliance. Lead generation involves personal data. Your rules depend on region and channel, but your baseline behavior should be conservative: collect only what you need, store it securely, and respect opt-outs. Avoid scraping personal emails in ways that violate platform terms. If you use AI tools, understand what data they retain and whether your inputs are used for training. When in doubt, don’t paste sensitive personal data into a tool that isn’t approved for it.

Risk 3: Over-automation. Beginners often automate the wrong thing: they auto-send messages before they have quality control. Automation should come after you have a repeatable checklist and you can reliably detect bad leads. A safe early approach is “human-in-the-loop”: AI drafts, you review, you send. AI scores, you decide.

Risk 4: Generic outreach that harms deliverability and brand. AI-written messages can sound polished but empty. If your first touch doesn’t reference a real reason you chose them, you’ll get ignored (or marked as spam). Your safeguard is a required personalization field: one sentence of evidence of fit.

If you treat these as engineering constraints—verification, data minimization, human review—you’ll keep the benefits of speed without the usual downside of low trust.

You do not need a complex tech stack to start. For beginners, the best “minimum tools” setup is: a chat AI, a browser, and one reliable place to save leads. The goal is low friction and repeatability.

1) Chat AI: Use it for prompts, extraction, rewriting, and scoring assistance. Create a reusable template prompt you can paste daily, such as: “Here’s my ICP. Here are 5 candidate companies. For each, summarize fit signals, risks, and the best outreach angle. If anything is unknown, say unknown.” Save this prompt in a notes doc so you don’t reinvent it.

2) Browser: Your browser is where truth lives: company sites, LinkedIn, directories, review sites, job boards, app marketplaces, community forums. AI can propose search queries like “site:linkedin.com/company ‘HVAC’ ‘50-200 employees’” or “intitle:‘case study’ ‘industry keyword’ ‘tool name’” to help you find better prospects faster.

3) Place to save leads: Pick one: a simple CRM, Airtable/Notion, Google Sheets, or a plain document with a consistent table format. The tool matters less than the fields. Start with these fields:

• Company
• Website
• Contact name + role (or “unknown”)
• Source (where you found them)
• Fit evidence (1–2 bullets)
• Status (Prospect/Lead/Qualified/Opportunity)
• Next step + date

The common mistake is spreading leads across places: some in email, some in bookmarks, some in DMs. Choose one home and commit for the duration of the course. You can migrate later once your process is stable.

This toolkit supports the chapter’s workflow: AI helps you plan searches and clean notes, the browser verifies and enriches, and your storage keeps the lead journey visible from stranger to qualified prospect.

If you don’t measure anything, AI “feels productive” even when it isn’t. Your first measurable goal for the course should combine volume, quality, and time saved. This prevents the classic failure mode: generating lots of leads that never convert.

Use a simple weekly metric you can track in your lead storage tool:

• Volume: number of new leads saved (not prospects—leads that meet your definition).
• Quality: percentage of those leads that become “Qualified Leads” using your basic rules.
• Time saved: estimated minutes per lead from research → saved record → first draft message.

Pick a target that matches your schedule. Example beginner goal: 20 new leads/week, 60% qualified, and <12 minutes per lead end-to-end. Adjust if your market is narrower (higher minutes, lower volume) or if your offer is broad (higher volume, lower minutes).

Here’s the practical checklist for “quality” that works for most beginners and pairs well with AI assistance:

• Matches ICP industry (yes/no)
• Matches size band (yes/no/unknown)
• Has a trigger or pain hint (yes/no)
• Has a reachable contact path (yes/no)
• Has a personalization hook (yes/no)

Ask AI to score leads against this checklist, but keep the final call. The judgment you’re building is the ability to say, “This is worth a message now” versus “This needs more research” versus “This is not a fit.”

When your metrics improve, you’ll know AI is helping in the way that matters: you’re moving faster from stranger to qualified prospect while keeping your outreach relevant and responsible.

Section 2.1: Start with outcomes: who you help and what changes

Beginners often start an ICP with demographics (“SaaS companies in New York”) and forget the most important part: the outcome your buyer wants. Outcomes make your targeting sharper because they connect to urgency, budget, and decision-making. Start from first principles: what changes after someone buys from you? This is also the foundation for your problem-to-solution story.

Write two sentences using this formula: “We help [role/company type] achieve [measurable change] by [your method], without [common pain].” Example: “We help small B2B service firms generate qualified sales calls by turning their website traffic into booked meetings, without hiring a full-time SDR.” The point is not marketing polish; it’s clarity.

Now build the simplest “problem-to-solution story” you can reuse in messaging and in AI prompts:

• Before: What situation are they stuck in? (symptoms)
• Cost: What does it cost them (time, revenue, stress, risk)?
• After: What does “better” look like in observable terms?
• Proof mechanism: Why is your approach credible (process, asset, guarantee, track record)?

Engineering judgment matters here: choose outcomes you can actually influence. “Double revenue” is rarely something you can promise; “increase qualified demos per month” is something you can directly affect. Another common mistake is writing an ICP around what you want to sell (“AI automation”) instead of what they want to buy (“faster lead follow-up,” “less manual research,” “more qualified pipeline”).

Practical outcome: by the end of this section, you should have a tight definition of success that you can hand to AI tools and to your future self when you’re tired and tempted to chase every lead.

Section 2.2: Firmographics and basics: size, location, industry

Once outcomes are clear, add basic firmographics. Firmographics are not the “real” reason someone buys, but they are a fast filter that saves you hours. For B2B lead gen, the most useful basics are: company size (employees or revenue), industry, and location/time zone. If you sell to individuals (B2C), translate this into life stage, income range, and geography.

Keep this as a one-page ICP section with ranges, not exact numbers. Example ranges: “10–200 employees,” “$1M–$20M revenue,” “English-speaking markets,” or “US/Canada time zones for live onboarding.” Add constraints tied to your delivery model: if you run live training at 2pm ET, it matters. If your pricing requires a procurement process, it also matters.

Workflow tip: define minimum viable fit and ideal fit. Minimum viable fit prevents you from discarding good leads too early; ideal fit helps you prioritize when the list gets large.

• Minimum viable fit: “Has a sales team of at least 1 person; can pay $X/month; sells B2B.”
• Ideal fit: “Has inbound traffic; has a CRM; sells high-margin services; founder is active on LinkedIn.”

Common mistakes: (1) making the ICP so broad that you can’t personalize (“any business”), (2) making it so narrow you can’t find leads, and (3) copying a competitor’s ICP without checking whether your offer and pricing match. Use your outcomes to choose firmographics that support the buyer’s ability to act. If your solution requires content creation, companies with no marketing owner may struggle—even if they “need leads.”

Practical outcome: you now have filters you can apply before you ever read a person’s profile, which is crucial when AI tools surface hundreds of possibilities.

Section 2.3: Buying signals: triggers that suggest real need

Buying signals are observable hints that a lead is likely to have an active problem and the ability to prioritize it. Think of them as “triggers” that move someone from theoretical fit to practical urgency. Without buying signals, you’ll contact people who match your firmographics but have no reason to change now.

Start by listing 8–12 signals you can detect from public info (website, job posts, LinkedIn updates, reviews, ads, newsletters). Tie each signal to your problem-to-solution story. If your offer improves speed-to-lead, a signal could be “running paid ads” (because delays waste spend). If your offer improves outbound, a signal could be “hiring SDRs” (because they’re investing in pipeline).

• Growth signals: hiring for sales/marketing, new office, recent funding, new product launch
• Pain signals: negative reviews mentioning support delays, site shows broken booking flow, inconsistent messaging
• Intent signals: visiting pricing pages (if you have access), downloading lead magnets, attending webinars
• Tooling signals: uses a CRM, uses scheduling software, active email list, pixels installed

Engineering judgment: prefer signals that are (1) easy to verify, (2) strongly correlated with purchase, and (3) relevant to your specific solution. “Active on social media” is weak unless your offer depends on social distribution. Another mistake is confusing vanity activity with intent; a company posting daily can still be broke, misaligned, or unwilling to change.

Practical workflow: add a “signal” field to your lead notes. Every time you find a lead, capture one sentence: “Signal observed: hiring SDRs” or “Signal observed: running Google Ads to a generic landing page.” This becomes training data for your own process and feeds directly into AI later when you want it to summarize and score leads consistently.

Section 2.4: Disqualifiers: who you should skip (and why)

Disqualifiers are the fastest way to improve lead quality. A beginner mistake is treating disqualification as “being picky.” In reality, disqualifiers protect your time, your deliverability (fewer spam complaints), and your confidence (fewer dead conversations). They also make AI lead generation dramatically more accurate because the model can exclude obvious non-fits.

Create a short list of “hard no” rules and “soft no” rules. Hard no means you will not pursue the lead. Soft no means you might pursue it only if something else is unusually strong (a major buying signal, a warm referral, or perfect timing).

• Hard no examples: cannot legally serve the region; budget below your minimum; industry you do not support; requires features you don’t have; competitor or conflict of interest
• Soft no examples: too small today but growing fast; no CRM yet but hiring sales ops; early-stage founder still validating offer

Include “why” for each disqualifier. The “why” prevents you (and AI) from applying the rule blindly. Example: “Skip companies under 5 employees because our onboarding requires a dedicated owner for 2 weeks.” Or “Skip regulated healthcare because we can’t sign required agreements yet.” This turns disqualification into engineering judgment rather than prejudice.

Common mistakes: (1) disqualifying based on assumptions (“they won’t pay”) instead of evidence (pricing, role seniority, business model), (2) having no minimum budget and then resenting leads later, and (3) ignoring operational constraints (time zone, language, support). A clean disqualifier list also makes your future messaging more compliant and respectful: you contact fewer irrelevant people and avoid outreach that could be perceived as intrusive.

Practical outcome: you now have the backbone of a yes/no qualification checklist. Disqualifiers become the first questions on the list because they save the most time.

Section 2.5: A 3-level lead status system: cold, warm, qualified

You don’t need a complex CRM workflow to start. You need consistent labels that help you decide what to do next. A beginner-friendly system is three statuses: cold, warm, and qualified. These map to your checklist and help AI summarize or prioritize leads without inventing its own criteria.

Define each status in terms of your yes/no checklist:

• Cold: matches basic firmographics, but no verified buying signal yet. Next action: light research + low-pressure first touch.
• Warm: matches basics and has at least 1–2 buying signals. Next action: personalized message referencing the signal; offer a small, relevant next step.
• Qualified: passes hard disqualifiers and meets your minimum viable fit plus multiple strong signals (or direct intent). Next action: direct CTA to a call/demo; faster follow-up.

Now write your yes/no checklist (8–15 items). Keep it scannable. Example categories: Fit (industry, size), Capability (has a website, has a sales owner, has a CRM), Need (signal present), and Access (you can reach a decision-maker). Each item should be answerable from public info or a first conversation. If it requires deep discovery, it’s not a lead-gen checklist; it’s a sales qualification framework.

Engineering judgment: pick thresholds you can uphold. If “must have 10k monthly visitors” is your rule but you can’t estimate traffic reliably, you’ll misclassify leads. Choose checks you can validate quickly. Also, don’t confuse “qualified” with “will buy.” Qualified means “worth a real sales attempt now.”

Practical outcome: you can prioritize outreach. Instead of working through a list randomly, you work qualified first, then warm, then cold. This is where AI becomes a multiplier: it can help label and summarize, but your definitions must come first.

Section 2.6: Prompt template: teach AI your ICP and rules

AI performs best when you provide (1) your one-page ICP, (2) your problem-to-solution story, and (3) your yes/no checklist with scoring rules. The goal is consistency: the same lead should receive the same status regardless of which day you run the prompt. This section turns your work into an AI-friendly prompt you can reuse for research, cleanup, and prioritization.

Use this template and replace the bracketed parts. Keep it in a notes app so you can paste it into any AI tool.

• Role: “You are a lead qualification assistant for [company/offer].”
• ICP (one-page): outcomes + firmographics + buying signals + disqualifiers (paste bullet lists).
• Problem-to-solution story: paste your 3–5 sentence version.
• Checklist: paste yes/no questions. Mark which are hard disqualifiers.
• Status rules: define cold/warm/qualified thresholds (e.g., “Qualified = no hard disqualifiers + minimum fit + ≥2 strong signals”).
• Input format: tell it what you will provide (raw notes, LinkedIn bio, website text, etc.).
• Output format: request clean fields: Company, Person, Role, Location, Evidence (signals), Disqualifiers, Checklist answers (Y/N/Unknown), Status, Priority (1–5), Next action, and a 1-sentence rationale.

Common mistakes: asking for a score without evidence, not allowing “Unknown,” and not forcing the model to cite which signal triggered the label. Require the AI to separate facts from assumptions by using an “Evidence” line and by permitting “Unknown.” This reduces hallucination and makes your process auditable.

Practical outcome: you can paste messy lead notes (screenscraped text, call notes, website snippets) and get structured contact fields plus a consistent status. This is the bridge to the next steps in the course: generating targeted search ideas, cleaning lead data, and writing first-touch messages that are aligned with your ICP instead of generic.

Lead generation starts with choosing the right ponds to fish in. Beginners often fail by picking a source that doesn’t match their ICP (for example, relying on LinkedIn alone for a niche local service) or by mixing too many sources at once and losing consistency. Pick two primary sources and one backup source for your first week.

Directories work best when your ICP has clear categories (industry, certifications, location). Examples include association member lists, local chambers of commerce, “Top agencies in X” pages, app marketplaces, partner directories, and review sites. The advantage is structured data; the risk is staleness and duplicates. Your job is to treat directories as a starting list, not a truth source.

Communities (Slack groups, forums, Reddit, Discord, niche Facebook groups) are best for finding intent signals: questions, hiring posts, tool complaints, “looking for a vendor” requests. Communities produce fewer but warmer leads. The mistake is extracting names without context—always capture the post/thread that indicates need.

Events (webinars, meetups, conferences, virtual summits) create time-bound opportunities. Attendee lists, speaker lineups, sponsors, and exhibitor directories are gold because they imply active budgets and current initiatives. A practical method: target sponsors/exhibitors first (they have spend), then speakers (they have influence), then attendees (they have intent).

LinkedIn is the most flexible source for B2B prospecting because you can filter by role, seniority, company size, location, and keywords. Use it for confirmation and enrichment even if you found the lead elsewhere. Common mistakes: searching with broad titles (“marketing”) without industry filters, or saving profiles without a company page match.

• Beginner rule: choose one structured source (directory/event) + one conversational source (community/LinkedIn).
• Capture standard: every lead should have a company, a person/role (or “unknown”), and a reason they belong in your ICP.

Engineering judgement here is about reliability and effort: prioritize sources where you can quickly find 20–50 candidates and where you can verify them with a website or public profile. If the source requires login scraping, paywalls, or questionable data harvesting, skip it for now and stick with public pages and manual review.

AI helps most before you search, not after. Your goal is to generate search angles—different ways your ICP might reveal themselves online—then turn those into targeted queries and filters. Provide your AI with your ICP basics (industry, location, employee range, role titles) plus 2–3 “trigger signals” (e.g., hiring, new funding, new tool adoption, compliance deadlines).

Use prompts that force specificity and output you can copy into a browser. Example prompt:

Prompt: “My ICP: US-based dental groups with 5–50 locations. Buyer roles: COO, VP Operations, Practice Manager. Give me 25 Google search queries and 15 LinkedIn keyword combinations to find companies showing these triggers: hiring patient coordinators, opening new locations, switching scheduling software. Include negative keywords to reduce irrelevant results.”

Good AI output includes variations such as:

• Role + problem keywords: “practice manager” “patient scheduling” “vendor”
• Trigger footprints: “we are hiring” + role title, “new location” + city, “implemented” + tool category
• Industry jargon: acronyms, compliance terms, “near me” phrases for local niches

Then apply judgement: remove queries that are too broad, too spammy, or likely to return job boards and aggregators. Add filters that reflect your qualification rules: geography (city/state), company size proxies (“locations,” “team,” “careers”), and stack signals (“Shopify,” “HubSpot,” “ISO 27001,” etc.).

A practical technique is the “two-pass search.” Pass 1 is discovery: 10–15 queries to identify patterns and sources you didn’t know existed. Pass 2 is extraction: once you find a good directory, event page, or list format, you stop inventing queries and instead systematically work through that source with consistent capture fields.

You do not need a spreadsheet to capture leads cleanly. You need consistent fields and a place to store them—your CRM, a notes database (Notion/Obsidian), Airtable, or even a form that writes to a list. The mistake beginners make is saving random snippets (“seems good”) that can’t be used later to personalize outreach or to de-duplicate.

Start with a minimal template you can fill in under 90 seconds per lead. These fields are enough for most beginners:

• Company name
• Website URL (canonical home page)
• Lead source (directory/community/event/LinkedIn + exact page link)
• ICP fit tags (industry, geography, size band)
• Contact name (or “Unknown”) + role
• Contact channel (LinkedIn URL or email if publicly listed)
• Trigger/proof (1–2 sentences, with link)
• Status (New / Validated / Not a fit / Duplicate)

To “turn messy lead notes into clean contact fields,” use AI as a structuring tool. Paste your raw notes (e.g., copied text from an event page and a LinkedIn snippet) and ask for JSON output matching your template. Example prompt:

Prompt: “Extract lead fields from the text below into JSON with keys: company_name, website, source_url, person_name, person_role, linkedin_url, fit_tags, trigger, status. If something is missing, use null and do not guess.”

The “do not guess” line matters. It prevents AI from hallucinating emails, addresses, or roles. You’re using AI to normalize and format—not to invent facts. Once you have consistent fields, you can sort and follow up without needing complex tooling. Consistency is what makes a small list feel manageable and professional.

Research creates candidates; validation creates leads you can trust. A fast validation pass reduces wrong matches, wasted outreach, and compliance risk. Your goal is not perfect certainty—it’s “good enough to contact” based on public information.

Use a simple website-first checklist. In 60–120 seconds, you should be able to answer:

• Is this the right company? (matching name, domain, and location)
• Do they serve the market you target? (industries, customer types, service area)
• Do they have a plausible need? (services offered, product maturity, growth signs)
• Can you find a real contact surface? (team page, leadership, or a legitimate contact page)

Red flags that often indicate a bad lead or high bounce risk:

• Domain parked, broken SSL, or “coming soon” pages
• Only a Facebook page with no business details (unless your ICP is micro-local and you explicitly target that)
• No clear offering, no address/service area for local businesses, or content that appears scraped/spammy
• Company name collisions (same name in multiple states/countries)

AI can assist by summarizing a homepage or extracting signals, but you should provide the source text yourself (copy the “About” section) rather than asking the model to browse. Example prompt:

Prompt: “Based on the About text below, classify ICP fit (High/Medium/Low) using these rules: [paste your rules]. List 3 supporting reasons and 2 uncertainties. Do not use outside knowledge.”

The key judgement is to stop early. If you hit two major red flags, mark “Not a fit” and move on. Beginners often spend 10 minutes trying to rescue a weak lead; that time is better used validating five new candidates.

Useful lead notes are not a biography. They are a short, evidence-based answer to: “Why are we contacting them, and what should we say first?” The best notes are portable—they still make sense a week later, even if you don’t remember the browsing session.

Write notes in three parts:

• Proof point: a concrete signal with a link (job post, announcement, tool badge, event sponsor listing).
• Context: what that signal implies for your offer (growth, compliance need, workload increase, new system).
• Angle: one personalization hook you can use in a first-touch message (a question, observation, or relevant benchmark).

Example note (tight and usable): “Proof: Hiring ‘Billing Coordinator’ (Careers page link). Context: revenue operations likely under strain during expansion. Angle: ask how they handle denial management and reporting across locations.”

AI can help you rewrite messy notes into this structure. Paste your raw snippets and ask for a compact “proof/context/angle” block limited to 60–90 words. This keeps you from storing paragraphs that never get used.

Common mistakes: copying entire webpages, saving opinions without evidence (“they seem big”), or losing the exact source URL. If you can’t point to where you found the signal, it’s hard to personalize credibly and harder to audit later. Treat notes as lightweight documentation—your future self (and your team) will thank you.

“Scraping” sits on a spectrum. On one end is manual research: you read public pages and copy a few fields for legitimate outreach. On the other end is automated extraction at scale from sites that forbid it, potentially collecting personal data without a lawful basis. Beginners should stay firmly in the first category.

A beginner-safe approach:

• Prefer public, intended-for-discovery pages (company websites, public directories, event exhibitor lists).
• Collect business context first (company + role) before personal data.
• Do not guess or generate emails; use emails only when publicly listed for business contact, or use a contact form/LinkedIn message.
• Respect terms and robots signals; if a site blocks automated access or requires login, treat it as a “manual only” source.
• Rate limit your own behavior: even manual copy/paste can look abusive if you hammer a site.

Where AI fits ethically is in summarizing, structuring, and prioritizing information you already accessed legitimately—not in circumventing access controls. If you later graduate to automation, do it with explicit permission, compliant vendors, and clear data handling policies.

Practical workflow recommendation: spend 30–45 minutes generating angles and queries (AI-assisted), 60–90 minutes collecting and validating 15–25 leads manually, and 15 minutes cleaning fields with AI into your template. This keeps volume realistic and quality high. Your goal as a beginner is not maximum extraction—it’s a dependable pipeline you can repeat without risking your reputation or violating platform rules.

Before you ask AI to clean anything, decide what “clean” means. Clean data is consistent data: the same fields exist for every lead, formatted the same way, so you can sort, filter, score, and message reliably. Start with a beginner-friendly schema that matches how you sell.

A practical minimum for B2B leads is: First name, Last name, Role/Title, Seniority (optional), Company, Company domain, LinkedIn URL (person), Company URL, Location, Source (where you found it), Notes, Last touched date, and Status (New/Qualified/Contacted/Waiting/Closed).

Formatting rules matter more than you think. Pick conventions and stick to them: use Title Case for names and companies; store emails in lowercase; store phone numbers in E.164 if you use phones; store locations as “City, State/Region, Country.” For URLs, keep the full canonical link (including https://) and strip tracking parameters when possible.

• Standardize names: Avoid storing “Sam R.” in one record and “Samuel Reed” in another. Use a dedicated field for nickname or preferred name if you need it.
• Standardize roles: “Head of Growth,” “Growth Lead,” and “Growth Manager” might be equivalent in your qualification rules. Store the raw title and a normalized category (e.g., Function=Growth, Level=Head/Manager).
• Standardize companies: Decide how you handle legal suffixes (Inc., Ltd.). Many teams store display name (e.g., “Acme”) and keep legal name only if needed.

Engineering judgment: don’t create 40 fields on day one. Too many fields lead to empty records and inconsistent usage. Start with what you will actually use for targeting, qualification, and outreach, then expand only when a repeated need appears.

AI is great at taking messy text and returning structured fields—if you constrain the task. Your prompt should specify: the fields you want, the formatting rules, and a “do not guess” requirement. Treat AI as a transformer, not a researcher.

Use a repeatable cleanup prompt whenever you paste in raw notes from LinkedIn, an event list, or a website. For example, ask AI to normalize titles, locations, and URLs into consistent formats, and to keep both the original and the cleaned version when helpful.

Practical prompt template (paste your raw notes after it):

“You are a data cleaner. Convert the raw lead text into a single JSON object with fields: first_name, last_name, title_raw, title_normalized, company_name_raw, company_name_normalized, location_raw, location_normalized (City, Region, Country), linkedin_url, company_url, company_domain, source, notes_clean. Rules: (1) Do not guess missing values; use null. (2) If a URL is not explicit, use null. (3) Remove tracking parameters from URLs. (4) title_normalized should be one of: Founder/Owner, Executive, VP, Director/Head, Manager, IC/Individual Contributor, Unknown. (5) Preserve any evidence in notes_clean.”

Why this works: the model has clear output shape (JSON), clear normalization categories, and a strict policy for unknowns. You can reuse the same template every time you capture leads.

• Common mistake: Asking “clean this list” without specifying fields and rules. You’ll get inconsistent formats that can’t be merged later.
• Tip: If you have multiple leads, request a JSON array and require a stable key like company_domain + linkedin_url when present.

Practical outcome: you can paste raw lead snippets and immediately get records that fit your system—without manually retyping names, title casing, or correcting broken links.

Duplicates are inevitable. You’ll find the same company from different sources: a directory, a webinar attendee list, and a LinkedIn search. If you don’t dedupe early, you’ll double-message prospects or split notes across records—both hurt conversion and credibility.

Start with deterministic keys when you can. In B2B, company domain is the best dedupe anchor. Two records with the same domain are almost always the same company, even if the display name differs (“Acme AI” vs “Acme”). For people, a LinkedIn profile URL is a strong unique identifier.

Then use AI for “fuzzy” cases where keys are missing. Ask it to compare candidates and explain its reasoning using only the text you provide. Example workflow:

• Group leads by similar company_name_normalized (or by domain when present).
• Within each group, have AI flag likely duplicates and propose a merge plan.
• Keep a merge log: what got merged, and why.

Deduping prompt (safe and practical):

“Given these lead records (JSON array), identify duplicates at the company level. Use domain as primary key; if domain is missing, use company name + location + website clues. Output: (1) duplicate_groups as arrays of record IDs, (2) a recommended master record per group, (3) fields to merge with precedence rules: keep non-null, prefer newer source, and append notes with source tags. Do not invent domains or URLs.”

Engineering judgment: accept that dedupe is probabilistic when you lack strong identifiers. In those cases, mark records as Possible duplicate instead of merging automatically. It’s better to keep two records than to merge two different companies and corrupt your database.

Practical outcome: a cleaner pipeline where each company has one “home” record and all related contacts and notes attach to it.

Enrichment means adding context that improves qualification and messaging. Done well, it saves time: you don’t need to open five tabs to remember what the company does or whether it fits your ICP. Done poorly, it creates confident-sounding fiction.

What to add safely (when you have a source in your inputs): company description (from their website “About”), industry category, product keywords, customer segment hints, and any explicit signals like “hiring SDRs” or “recent funding” only if you captured the evidence text or link. You can also add normalized fields derived from known text, such as mapping “Head of Demand Gen” to Function=Marketing and Level=Director/Head.

What to avoid: guessing revenue, headcount, tech stack, or decision authority without a source; inventing email addresses; inferring personal attributes; or “filling in” missing locations because a company is associated with a city in your memory. If you need those fields, collect them from a reputable data provider or direct research, and store the citation.

• Good enrichment field: “ICP Fit Reason” (short bullet from evidence: ‘Sells to SMB healthcare; hiring for growth; uses outbound’).
• Risky enrichment field: “Budget: high” without evidence.

Safe enrichment prompt:

“Using only the provided source snippets (website excerpt, LinkedIn ‘About,’ and my notes), add: industry_category (choose from list), value_prop_summary (1 sentence), and icp_signals (up to 5 bullets). For each item, include a ‘source_quote’ from the snippets. If evidence is missing, return null.”

Practical outcome: richer lead records that support better scoring and more relevant first-touch messages—without making up facts.

Even with good prompts, AI will sometimes be uncertain: ambiguous names, unclear titles, multiple possible company websites, or conflicting locations. Your system should make uncertainty visible instead of hiding it. The aim is to prevent “silent errors” that look clean but are wrong.

Add a simple confidence pattern to your workflow. For any normalized field, ask AI to return (1) the cleaned value, (2) a confidence rating (High/Medium/Low), and (3) what evidence it used. If confidence is Low, require the model to propose next steps rather than guess.

Confidence check prompt:

“Review this cleaned lead record against the raw notes. For each field (name, title, company, location, URLs), output: value, confidence (High/Medium/Low), and evidence (exact substring from raw notes). If any field is Low confidence or missing, list ‘follow_up_questions’ I should answer or sources I should check.”

• Common mistake: Treating AI output as final and copying it into your CRM without verification.
• Practical fix: Create a status like “Needs verification” and only move leads to “Ready to contact” once key fields (company, role, URL/source) are High confidence.

Engineering judgment: decide which fields are “must be correct.” For compliant outreach and good personalization, the must-haves are usually: correct name, correct company, correct role/function (at least roughly), and a reliable source link. Everything else can be optional or marked unknown.

Practical outcome: fewer embarrassing mistakes (wrong company, wrong title) and a clear to-do list for the small number of leads that require human review.

“Create a single source of truth” means you choose one place where the latest lead status and core fields live. You can still capture raw notes anywhere, but they must end up in the same system if you want consistent scoring, outreach tracking, and handoffs.

Beginners usually choose one of three storage options:

• CRM (HubSpot, Salesforce, etc.): Best when you’re actively selling and need pipelines, tasks, email logging, and reporting. Tradeoff: CRMs can feel heavy, and bulk cleanup can be annoying unless you import carefully.
• Table tools (Airtable, Notion databases, Google Sheets as a database): Best for flexible schemas, quick filtering, and lightweight enrichment. Tradeoff: without discipline, you’ll create multiple tables and lose the “single truth.”
• Simple docs (Google Docs/Notes): Best for very early exploration. Tradeoff: hard to dedupe, score, and track follow-ups; data becomes unsearchable text.

A practical approach is “table first, CRM second.” Use a table tool as your lead intake and cleaning station (where AI output is pasted as structured records), then push qualified leads into your CRM once they meet your minimum requirements. This keeps your CRM clean and reduces clutter.

To maintain a single source of truth, define one record ID per company and per person, and enforce a rule: status changes happen only in the system of record. If you take notes elsewhere (calls, DMs), copy the essential outcome back into the record within 24 hours.

Common mistake: storing the same leads in three places and updating only one. The fix is a weekly “sync ritual”: dedupe, merge notes, and confirm that each active lead has one current status. Practical outcome: you always know who is next to contact, why they’re a fit, and where the proof lives.

Qualification becomes straightforward when you reduce it to first principles. Every lead decision can be explained with three variables: fit (are they the right type of customer), need (do they likely have the problem you solve), and timing (is it plausible they will act soon). If you can’t articulate these three, any numeric score is just decoration.

Fit usually comes from your ICP rules (industry, company size, geography, tech stack, role/title). Fit is about “can this be a customer?” not “will they buy.” Need is about symptoms: hiring for a role related to your product, complaining publicly about a pain point, adopting a tool that creates a downstream gap you solve, or having a business model that strongly correlates with your use case. Timing is about triggers: recent funding, leadership change, product launch, compliance deadline, a new initiative mentioned in a post, or a job posting that implies active spend.

A practical workflow: when you collect a lead, capture 1–2 pieces of evidence for each category. Evidence beats vibes. For example: “Fit: 50–200 employees (LinkedIn), Need: hiring RevOps manager (job post), Timing: announced new sales region (press release last week).”

Common mistakes: confusing fit with need (“they’re in our industry so they must want this”), treating timing as a guess instead of a trigger, and over-penalizing missing info. Missing info is not always a “no”—it’s often a “research later” signal you’ll formalize in Section 5.5.

Practical outcome: you can explain every outreach decision in one sentence using fit/need/timing, which makes your process teachable and consistent.

Now convert your checklist into a simple score. The goal is not statistical precision; it’s speed and consistency. Use a 0–2 scoring model per factor so you can rate leads in under a minute.

• Fit (0–2): 0 = outside ICP; 1 = partial match (one key mismatch); 2 = strong ICP match.
• Need (0–2): 0 = no clear pain; 1 = weak/assumed pain; 2 = explicit signal (post, job ad, tool usage, stated goal).
• Timing (0–2): 0 = no trigger / long horizon; 1 = possible trigger; 2 = clear trigger within ~30–90 days.
• Access (0–2) (optional but helpful): 0 = no contact path; 1 = indirect path; 2 = direct email/DM and correct persona.

This yields a 0–8 score if you include Access, or 0–6 if you don’t. For beginners, that’s enough. Then define thresholds you will actually use:

• 7–8 (or 5–6 without Access): outreach today.
• 5–6: outreach this week with a softer opener (ask a question, confirm context).
• 3–4: research later bucket (missing evidence).
• 0–2: discard or park (not ICP).

Engineering judgment: keep the model stable for at least two weeks before changing it. Beginners constantly tweak weights after one bad day, which destroys comparability. If you want one “override rule,” make it explicit: for example, “A Fit=0 is an automatic no, regardless of the total.”

Practical outcome: your checklist becomes a usable score you can apply repeatedly without complex scoring software.

Scoring is only half the job. You also need a short explanation that makes outreach easy and keeps you honest about evidence. AI is ideal for converting scattered notes into a compact “fit summary + next steps” block you can paste into a CRM, a Notion table, or even a plain document.

Use AI to generate a structured summary that includes the score, the evidence, and a recommended first action. The key is to feed the model your raw facts and demand citations to those facts (not external guesses). A practical prompt template:

Prompt: “You are my SDR assistant. Using ONLY the notes below, produce: (1) Fit/Need/Timing/Access scores (0–2 each) with one sentence of evidence per factor, (2) total score, (3) one recommended next step, (4) one risk/unknown to verify. Notes: [paste LinkedIn snippet, job post excerpt, website notes, your call notes]. Output in 6 bullet points.”

This does two things. First, it forces you to keep evidence attached to the score, which reduces self-deception (“I feel like they need it”). Second, it produces an immediate action, which supports your “today list” creation in the next section.

Common mistakes: letting AI invent facts (“they likely have budget”), pasting too much irrelevant text (the model latches onto noise), and failing to capture unknowns. Unknowns are valuable; they guide micro-research or the wording of your first message (“Curious if you’re already…”).

Practical outcome: every lead has a compact, consistent summary: why they fit, what triggered timing, and what you’ll do next—ready for outreach or handoff.

After scoring, you still need prioritization. Two leads can have the same score but require different effort. Use simple rules to create a “today list” that balances fastest wins and big bets.

Fastest wins are high score + low friction. Examples: you have direct contact info, the persona is correct, and the trigger is recent. These leads should dominate daily outreach because they create quick feedback loops (replies, objections, booked calls). Big bets are high score + higher effort: larger accounts, multiple stakeholders, unclear access, or requiring customization. These are worth pursuing, but not at the expense of daily volume.

Practical prioritization rules you can implement immediately:

• Rule 1 (Daily quota): 70% fastest wins, 30% big bets.
• Rule 2 (Fresh triggers first): prioritize triggers within the last 14 days.
• Rule 3 (Effort cap): if a lead needs >10 minutes of research to send a first message, move it to “research later” unless it’s a strategic account.
• Rule 4 (One-thread focus): contact one persona per account first; expand only after you get signal.

AI can help generate your “today list.” Ask it to sort leads by (a) total score, then (b) freshness of trigger, and finally (c) access quality, and to output the top 10 with one-sentence outreach angles. Keep this list small enough that you will finish it today—completion beats ambition.

Practical outcome: you know exactly who to contact today, and why, without feeling like you need the “perfect” scoring system.

Beginners often treat uncertainty as failure: “If I can’t score it confidently, I’ll ignore it.” That wastes good opportunities. Instead, formalize uncertainty with a research later bucket and clear rules for moving leads in and out.

A lead belongs in “research later” when fit is plausible but evidence is incomplete—typically Need or Timing is unknown, or Access is missing. The point of the bucket is to protect your daily outreach time while keeping promising leads from disappearing.

Make the bucket actionable by attaching a single research question to each lead, such as:

• “Confirm headcount range and whether they sell B2B or B2C.”
• “Find a trigger: funding, hiring, new initiative, or tool adoption.”
• “Identify the right persona and obtain a direct contact path.”

Use AI for micro-research planning rather than long investigations. A good prompt: “Given this lead and my ICP, list the top 5 fastest checks (under 3 minutes each) to confirm need and timing, and what evidence would upgrade the score from 1 to 2.” This keeps research bounded and purposeful.

Common mistakes: turning “research later” into a graveyard, adding leads without a question, and doing deep research before sending any message. Remember: sometimes the first message is your research—ask a crisp question that surfaces need without sounding intrusive.

Practical outcome: uncertainty becomes a managed state, not a blocker, and you maintain momentum while steadily improving lead quality.

Simple scoring systems drift unless you run a weekly review. Quality control doesn’t mean bureaucracy; it means small, consistent checks that reveal where your judgment (or the AI’s summaries) is going wrong. Set a recurring 20–30 minute weekly block.

Start with sampling. Randomly select 10 leads from the week: a few you contacted, a few you skipped, and a few in “research later.” Re-score them from scratch using your rubric. Compare the new score to the original. Large differences indicate unclear rules or inconsistent evidence collection.

Then do spot checks on outcomes. For contacted leads, tag results: replied, no reply, negative, booked meeting. Look for patterns by score band. If your 7–8 leads don’t outperform your 5–6 leads, you are likely over-scoring Need or Timing based on weak signals.

Track error patterns explicitly. Common ones include:

• Industry bias: assuming need because you like the industry.
• Title mismatch: messaging the wrong persona even when the account fits.
• Trigger inflation: treating “posted on LinkedIn” as timing.
• AI hallucination: summaries that imply budget/intent not present in notes.

Use AI as a reviewer too. Provide 5 scored leads with your evidence and ask: “Identify any scoring inconsistencies with my rubric and propose one rule clarification.” Only accept changes that you can state as a simple rule and test for two weeks.

Practical outcome: your qualification stays grounded, your “today list” improves week over week, and your outreach time is spent on leads with measurable, repeatable reasons to believe they’ll respond.

Effective first messages are built from three parts: value, relevance, and a clear ask. Value is the outcome you help with (reduce churn, book more demos, speed up onboarding). Relevance is why you picked them (industry, role, trigger event, tool stack). The ask is a small next step (a 10–15 minute call, permission to send a one-pager, or a quick question). If any of these are missing, the message either feels generic, confusing, or demanding.

A beginner mistake is writing a “feature tour” instead of a first message. Your product details can come later. Another common mistake is hiding the ask inside a paragraph. Make the ask explicit, and make it easy to say yes or no.

Use AI as a drafting partner, but supply the ingredients. Give it your ICP, the prospect’s role, and one reason you believe they might care. Example prompt:

Prompt: “Write a first outreach email to a [role] at a [company type]. Context: they likely care about [pain/outcome]. Relevance: I noticed [trigger]. Offer: [specific help]. Ask: 15-minute call next week. Constraints: 90–120 words, 2 short paragraphs, one question, no hype, no ‘revolutionary.’”

Engineering judgment matters: if you can’t state the relevance in one sentence, don’t send the message yet. Go back to research and capture one more concrete note (a hiring post, a product change, a stated KPI, a tool mentioned). Practical outcome: your first messages become consistent “mini experiments” you can repeat and improve.

Prospects can tolerate short messages; they rarely tolerate fake messages. The easiest way to “sound like you” is to show the AI what “you” sounds like. Create a small voice reference: 2–3 emails you’ve written that got good replies (or a paragraph describing your style: direct, friendly, low-pressure). Then instruct the model to imitate that style while keeping the content compliant and concise.

A useful technique is a two-step draft. Step 1: AI produces a neutral draft. Step 2: you ask AI to rewrite in your voice with explicit constraints. Example prompt:

Prompt: “Rewrite this outreach in my voice. My style rules: short sentences, no exclamation points, avoid buzzwords, one concrete detail, one clear ask, polite opt-out. Keep it under 110 words. Here’s a sample of my writing: [paste 100–150 words]. Here’s the draft to rewrite: [paste draft].”

Common mistakes include over-styling (“too clever”), over-apologizing (“sorry to bother you”), and over-personalizing (“I loved your post” with no substance). If AI outputs fluffy lines, delete them. Short, calm, specific beats long and “impressive.”

Practical outcome: you can generate multiple variations quickly (email, LinkedIn message, voicemail script) while keeping a consistent identity. This also helps team members: one shared “voice prompt” reduces randomness across reps and makes results easier to compare.

Personalization works when it proves relevance, not when it proves surveillance. The rule of thumb: personalize with information a reasonable person would expect you to have in a business context (company website, job post, public announcement, published case study). Avoid personal details (family, photos, non-business social posts) and avoid implying you tracked them (“I saw you visited our site”).

This is where your lead notes become powerful. You don’t need a spreadsheet full of fields; you need 1–3 clean notes that justify why you’re reaching out. Examples: “Hiring 2 SDRs,” “Using HubSpot,” “Launching in Germany,” “Complaints about onboarding in reviews.” Then let AI convert notes into one sentence of relevance.

Prompt: “Turn these lead notes into one non-creepy relevance line for an outreach message. Notes: [paste bullets]. Constraints: 1 sentence, business-only, no flattery, no assumptions, no ‘noticed you’re struggling.’ Provide 3 options.”

A common beginner error is inventing pain (“you must be frustrated with…”) instead of offering a hypothesis. Safer language: “Often teams like yours run into…” or “Curious if… is on your radar.” Another error is overloading the message with multiple personalized facts; one sharp detail is enough.

Practical outcome: you can scale relevance. Even with 20 leads, you can craft individualized messages by focusing on one credible trigger and one plausible outcome, rather than writing unique essays for each prospect.

Most replies come from follow-ups, not the first message. Beginners often either spam daily or wait three weeks and restart from scratch. A simple, respectful sequence is better than improvising. Start with 3–4 touches over 10–14 days, then pause. If you have multiple channels available, use them lightly (email + LinkedIn), but don’t “pile on” the same day.

Here’s a beginner-friendly timing pattern: Day 1 first email, Day 3 follow-up, Day 7 follow-up with a useful asset or question, Day 12 close-the-loop message. Each follow-up should add something: a clearer ask, a different angle, or a smaller commitment (e.g., “Worth sending a 3-bullet summary?”).

Prompt: “Create a 4-touch outreach sequence for [ICP] about [offer]. Channel: email only. Constraints: each message under 100 words, low-pressure, one question max, include a polite opt-out line, and each follow-up must add new value (not ‘bumping this’). Provide subject lines.”

• Follow-up #1: Restate relevance in one line and repeat the ask.
• Follow-up #2: Offer a small resource (checklist, short audit, benchmark) or ask one qualifying question.
• Follow-up #3 (close loop): Give them an easy “no,” confirm you’ll stop, and leave a door open.

Track outcomes by tagging replies: Interested, Not now, Not a fit, Wrong person, No response. Then adjust one variable at a time (subject line, ask, relevance trigger). Practical outcome: you build a simple sequence you can reuse, instead of rewriting from scratch every week.

Outreach must be compliant and safe. Rules vary by region and channel, but a beginner-safe baseline is: use business contact details gathered from legitimate sources, clearly identify yourself, make the purpose clear, and provide an easy opt-out. For email, include an unsubscribe instruction (even in one-to-one outreach) and honor it quickly. For SMS or automated dialing, consent requirements are often stricter—avoid until you understand your local rules.

Also think about data handling. Don’t paste sensitive personal data into an AI tool if you don’t have permission or if your tool’s policy doesn’t allow it. Keep your lead notes business-relevant and minimal. If you store notes, protect them: restrict access, avoid unnecessary fields, and delete data you no longer need.

Safety prompt: “Review this outreach message for compliance and risk. Flag any lines that could be misleading, too personal, or non-compliant. Suggest safer replacements. Context: outbound to business contacts, include opt-out, no claims that require proof. Message: [paste].”

Common mistakes include making unsubstantiated claims (“guaranteed results”), implying you have private information, or forgetting an opt-out. Another mistake is “over-automation” that removes judgment; AI can draft, but you are responsible for what you send.

Practical outcome: you reduce legal and reputation risk while increasing trust. Compliance isn’t just a checkbox—it makes your outreach feel more professional and lowers the chance of angry replies.

This chapter’s lessons come together in a weekly workflow you can run in about 60 minutes. The goal is a repeatable system: you research a small batch, capture clean notes, qualify quickly, send human messages, then review replies to improve next week.

Step 1 (15 min) Research: Pick one lead source (directory, LinkedIn search, event list) and gather 10–15 prospects. Limit scope to avoid rabbit holes. Capture one “trigger” per prospect you can cite credibly.

Step 2 (10 min) Capture: Dump messy notes into AI and ask it to normalize fields (Name, Role, Company, Email/URL, Trigger, Hypothesis, Next step). Keep it lean—no spreadsheets required if a simple doc or CRM works for you.

Step 3 (10 min) Qualify: Use your beginner checklist (ICP match, clear trigger, reachable contact, plausible need, no obvious conflicts). Ask AI to score leads 1–5 and explain why, but keep final judgment with you. If the model can’t justify the score using your notes, the lead isn’t ready.

Step 4 (20 min) Message: Draft first touches and queue follow-ups. Use your voice prompt, add one non-creepy relevance line, and include a clear ask plus opt-out. Send fewer, better messages—quality beats volume early on.

Step 5 (5 min) Review: Log outcomes (replied, booked, bounced, opt-out, no response). Paste a few replies into AI and ask: “What objection pattern do you see, and what should I change next week—relevance, offer, or ask?” This is where you learn from replies instead of guessing.

Common workflow failure is skipping review. Without it, you repeat the same weak message and blame the market. Practical outcome: you build a feedback loop that steadily increases reply rates while keeping your outreach respectful, compliant, and sustainable.