An outbound playbook for 2025
The four pillars powering Workflows.io to $2M ARR and beyond
A conversation with Elena Verna, the Head of Growth at Lovable, produces more insights-per-minute than perhaps anyone else in tech. When she’s speaking on a podcast or at an event, I know it’s something I won’t want to miss.
Elena is joining Metronome on December 10th to unpack lessons from Lovable around how to price AI. She’ll cover what makes AI pricing so complex, why traditional SaaS pricing models fall short, and how Lovable is experimenting with new monetization strategies. Grab you spot today.
👋 Hi, it’s Kyle and welcome to Growth Unhinged, my weekly newsletter exploring the hidden playbooks behind the fastest-growing startups.
Many have predicted the demise of outbound. Well, it’s alive and kicking. GTM leaders say intent-based outbound is the #2 channel where they plan to invest the most going into next year based on data from the 2025 State of B2B GTM report.
I wanted to explore the modern outbound playbook for 2025. For help I turned to Fivos Aresti, co-founder of Workflows.io. He shared an extremely tactical guide including first-hand learnings and recipes you can steal. (PS: This post may be cut off by your email provider; read it in full here.)
We hit $1M ARR three months after launching Workflows.io, an AI-first growth studio for B2B tech companies. Now we’re on track to reach $2M ARR by the end of the year.
LinkedIn content has been our main acquisition channel. But here’s the problem with lead generation through content: you can’t choose who’s going to book a meeting with you.
Outbound, instead, lets you choose precisely who you can go after. With proper qualification and scoring, you can pick the accounts you’d love to work with and know you can provide the most value to.
Ten years ago, most outbound activities were completely manual. Today, everyone is talking about automated outbound with AI.
The truth? You need both.
This is our outbound playbook to add $1M in ARR and finish the year beyond $2m ARR. Spoiler: It’s NOT just automated outbound.
The pillars of modern outbound
We see four pillars of successful outbound in 2025. There’s cold calling (yes, it still works!), email campaigns (automated sequences), LinkedIn campaigns (signal-based engagement), and manual prospecting for Tier 1 accounts.
When running outbound, we needed to choose where to turn the dial between signal-based plays versus cold outreach. And we needed to choose between where to invest in automation versus a manual, more personalized approach. Here’s where we landed.
Manual cold outreach for our ICP target accounts
Signal-based outbound is great, but realistically only a fraction of our target market was showing signals at any given point of time. We still needed to convert a cold audience in our ICP. We run cold outreach continuously, with micro and signal-based campaigns layered on top.
For our cold outreach program, the first step was to decide exactly who we were going to target. We defined our ICP and created a scoring model around it. We used this framework for ICP modeling and here’s what the end result looked like.
The second step was to map out our total addressable market (TAM), capturing every single SaaS company we could before filtering down to our target account list where we layer on more AI account research. By going broad before condensing down, we minimized the risk of missing accounts by relying on standard database filters.
We used five sources to create the initial list:
BuiltWith (to identify HubSpot users)
GetLatka (to identify SaaS companies)
Exa (to crawl websites with AI scraping)
Crunchbase (to identify funded companies)
Apollo (to get firmographic information)
We pulled CSVs from all five providers, uploaded them into one Clay table, and normalized and deduped everything. Initially, we found 66,000 companies. But that was a very broad list of SaaS companies that COULD fit our ICP.
Next, we eliminated as many irrelevant companies as possible. Here’s the exact prompt we used to qualify if the company was actually a SaaS:
#CONTEXT#
You are an AI-powered web researcher and extractor. Your task is to determine whether a company is a SaaS business by visiting its official website and credible public pages, then return a structured JSON verdict based strictly on on-page evidence.
#OBJECTIVE#
Assess: “Is “ + + + “ a SaaS company?” and output a JSON object with fields is_saas (true/false), confidence (”high”/”medium”), and reasoning (concise evidence).
#INSTRUCTIONS#
1) Identify the company:
If a domain is available on the company site or via branded search, prefer using that domain. Use the concatenation of and as the company identifier for searches. Do not infer any other columns.
2) Primary research targets (public pages only; no logins/paywalls):
- Company homepage
- Pricing page (keywords: pricing, plans, subscription)
- Product/features page (keywords: features, product)
- Free trial/demo/signup pages (keywords: free trial, demo, start, get started)
- Screenshots or product UI pages (keywords: screenshots, dashboard, interface)
- About pages that clearly state “platform,” “software,” or “app”
3) Evidence checks for SaaS (SAAS = Software + Cloud + Subscription):
- Look for explicit subscription pricing with monthly/annual plans.
- Presence of free trial, demo signup, or self-serve onboarding.
- Clear product features page describing software capabilities.
- Visible software screenshots or dashboard images.
- Company/self-description using terms like “platform,” “software,” or “app” delivered via the cloud.
4) Exclusion checks (NOT SAAS):
- Agency/consulting selling services instead of software.
- Marketplace connecting buyers/sellers without selling proprietary software subscriptions.
- Hardware-focused business.
- Media/content site.
- Systems integrator/implementer of others’ software (e.g., “Salesforce partner”).
5) Decision and confidence rules:
- HIGH: Found pricing with subscription plans and/or free trial/demo plus clear product features; or very clear language and evidence the product is a cloud software subscription.
- MEDIUM: Software is implied (mentions platform/software/app) but lacks clear pricing/trial or screenshots; early-stage with waitlist only.
Be decisive; avoid “low” unless absolutely no signal (then provide MEDIUM only if software is implied; else false with HIGH if clearly non-SaaS).
6) Data handling and constraints:
- Only use publicly visible web pages. Do not log in, submit forms, or interact with dynamic elements.
- If pages are client-rendered, rely on whatever content is retrievable; do not execute JavaScript beyond basic page load available to the scraper.
- Extract short quotations or page titles to justify the reasoning. Keep reasoning concise (1–2 sentences) citing the exact evidence found (e.g., “Pricing page with monthly plans,” “Free trial button,” “Agency services page”).
7) Output format:
- Return exactly one JSON object with this schema: {”is_saas”: true/false, “confidence”: “high/medium”, “reasoning”: “[what you found]”}
- Examples: CRM with pricing page → {”is_saas”: true, “confidence”: “high”, “reasoning”: “Subscription pricing and product features page”}; Marketing agency → {”is_saas”: false, “confidence”: “high”, “reasoning”: “Agency selling services, not software”}; Early startup with waitlist → {”is_saas”: true, “confidence”: “medium”, “reasoning”: “Platform mentioned but no pricing yet”}
For premium subscribers: The workflows mentioned in this piece have been added to the AI for GTM prompt library, which now includes 50+ AI use cases. Subscribe now.
And these were all the data points we found to qualify companies further:
Headcount (Clay “Enrich Company”)
Location (Clay “Enrich Company”)
Business model (Clay Research Agent)
Industry (Clay Research Agent)
Tech (BuiltWith)
Funding (Crunchbase & Clay Research Agent)
For Q4 we wanted to focus only on VC-funded companies using HubSpot, which narrowed it down to 5,700 accounts. Then we applied our scoring model to tier the accounts based on the criteria we’d set earlier.
We were left with a qualified and scored target account list. Here’s how we broke down tiers and allocated resources:
Dream 150 accounts: strictly manual LinkedIn prospecting
Tier 1: Cold calling and semi-automated sequences
Tier 2: Automated email and LinkedIn sequences
Tier 3: Automated email sequences
Next, we layered in account research for personalization and segmentation. For each account, we captured: (1) who’s their ICP, (2) what data they’d want to find on their ICP, (3) their LinkedIn posting activity, and (4) their sales department structure
The final step was writing the email and LinkedIn copy for our sequences. We tested two to three variants for each building block of the message including the first line, body, and call-to-action. (Here’s a cheat sheet around how to write a winning message.)
Automated, signal-based plays for other ICP accounts
There’s been a lot of noise about signals lately and, realistically, a lot of potential signals. At Workflows.io were considered these to be the most important ones to automate:
Website visitors: Warm outbound campaign targeting (de-anonymized) website visitors.
LinkedIn connections of founders: Outbound campaign to warm connections of founders.
Company page followers: Campaign targeting followers of our company LinkedIn page.
Competitor page followers: Outbound campaign targeting followers of competitor LinkedIn pages.
LinkedIn engagement: Outbound campaign triggered by ICPs who recently engaged with our posts.
Social listening: Campaign targeting prospects who interacted with posts that contain relevant keywords.
LinkedIn profile visitors: Outbound campaign for ICPs who recently viewed one of our team members’ LinkedIn profile account.
Event attendees: Campaign targeting people attending or who recently attended relevant industry events.
MELs (lead magnet downloads): Follow-up campaign for contacts who downloaded resources but didn’t convert into a meeting.
Job changes of champions: Outbound campaign reconnecting with champions who moved to new companies.
Tech stack signals: Companies using specific tools (HubSpot)
Customer alumni: Outbound campaign identifying people who previously worked at our closed-won accounts and re-engaging them at their new companies.
Closed-lost reopens: Campaign re-engaging previously closed-lost deals after a set period.
We used these signals to create initial lists, and layered in AI qualification to avoid spamming accounts that might demonstrate multiple signals.
There are hundreds of signal-based plays we could’ve run; the best ones to start with depend on your GTM motion and level of sophistication. We prioritized five as our favorites:
Play 1: Customer Alumni
Outbound campaign identifying people who previously worked at our closed-won accounts and re-engaging them at their new companies.
We pull closed-won accounts from HubSpot, find alumni who left for new companies using Clay, score those companies with ChatGPT, then automatically route prospects into tiered outreach sequences based on the account-fit score.
Tools: HubSpot, Clay, ChatGPT, Instantly, HeyReach
Play 2: Website Visitors
Warm outbound campaign targeting (de-anonymized) website visitors.
We identify companies and contacts visiting our website using Warmly, enrich the data in Clay, score leads with ChatGPT, then route top-tier accounts to Slack notifications for immediate cold calling (phone numbers via BetterContact) while lower-tier prospects enter automated email/LinkedIn sequences through Instantly and HeyReach.
Tools: Warmly, Clay, ChatGPT, Instantly, HeyReach, Findymail, BetterContact
Play 3: LinkedIn Connections of Founders
Outbound campaign to warm connections of founders.
We export LinkedIn connections natively in the platform without any additional tools. Between me and my co-founder we have 50,000+ followers and 13,000+ connections. Out of those, there were 3,400+ GTM leaders that we could potentially reached out to. The messaging here could be conversational since they’ve warmed up to our content.
Tools: LinkedIn, Clay, ChatGPT, HubSpot, Instantly, HeyReach, BetterContact
Play 4: LinkedIn Engagement
Outbound campaign triggered by ICPs who recently engaged with our posts.
We track engagement with Trigify (likes, comments, reposts) and profile visitors with Teamfluence from our LinkedIn content (for all of our employees), then route the data back to Clay via webhooks. From here we qualify and enroll leads in sequences according to tiers.
Tools: Trigify, Teamfluence, Clay, ChatGPT, HubSpot, HeyReach, Instantly
Play 5: Social Listening
Campaign targeting prospects who interacted with posts that contain relevant keywords.
Clay continuously monitors LinkedIn for specific keywords (our keyword is actually “Clay”), captures who’s engaging with those posts, enriches the contact and company data, uses ChatGPT to qualify fit and then assigns leads to sequences.
Tools: Clay, ChatGPT, HubSpot, HeyReach, Instantly, Findymail, BetterContact
What we’ve learned so far
Depending on the trigger, we’ve found signal-based plays to perform anywhere from 2x-10x better than cold outreach plays.
Our best-performing play so far has been: LinkedIn connections of founders.
Here were the initial results (still ongoing): 370 messages sent and 94 replies.
The reason this campaign got a 25.4% reply rate is because we were capturing the interest me and my co-founder were accruing from over 1.5 years of posting content on LinkedIn. What most people miss about signals is that the most effective ones are those that you create from your other marketing efforts.
Our biggest learnings from building GTM motions in 2025 is that you have to vigorously “un-silo” your channels. We believe in the GTM flywheel:
You run outbound to hand-raisers from content. Your best-performing organic content becomes your ads. Your ads target the same accounts in your outreach lists.
Heading into 2026, we want to double-down on our Tier 1 accounts with “unscalable” tactics. We’re planning a New Year’s gifting campaign and are setting up ABM ad campaigns that will trigger tasks for manual outreach.
New GTM tech is great, but it’s not perfect nor does it replace humans. The most successful GTM teams aren’t automating absolutely everything, but they are combining AI workflows with human expertise.
 | A guest post by
|
Appreciate this breakdown. We are in the middle of building out our GTM strategies and this give us some additional frameworks to add and think through.
Hey guys, thanks for sharing this. I'll be in touch :)