This AI Sales Agent Outperformed a Human Outbound Team. Here Is How It Was Built.
By Aaron Reid, Head of VIP at Instantly.ai ·
Most people think cold email is spam that clogs inboxes and never works. Aaron Reid, who leads the VIP managed service team at Instantly.ai, argues the opposite: cold email is the only channel that lets you target very specific people, qualify them, and send hyper-relevant messaging to millions a month. In this GTM Clawllective session, joining live at midnight from Bali, he made the case for a single idea that decides whether outbound wins or fails, then walked through the Instantly platform and the AI sales agent his own team now relies on. This post covers his offer framework, how the platform fits together, the agent that he says outperformed his human team, and his answers on deliverability, signals, and data governance.
Why Cold Email Still Works
The reason cold email deserves attention, in Reid's view, is that no other channel gives you the same control. You can target really specific people, qualify them against a long list of data points, and give each one extremely specific messaging, at a scale of millions of contacts a month. Paid channels and inbound cannot match that combination of precision and volume.
Where teams go wrong is treating outbound like a pitch. Pitching your product or asking a stranger for a meeting works if you are Forbes offering an interview, or a known brand people already trust. For everyone else, the person receiving the email has no reason to care who you are. The fix is not better copy on the same bad ask. It is a fundamentally different thing to put in front of them.
What Makes an Offer People Can't Refuse?
The core of the session was the offer. Reid's position is blunt: the best way to turn a stranger into a paid customer is to lead with an offer they cannot refuse, not a request for their time.
He uses a simple test to check whether an offer is strong enough. If you sell a product in the 2K to 3K range, ask whether someone would pay $100 for the offer itself. If the answer is no, it is probably not strong enough. This is why he avoids the term "lead magnet." Lead magnets have come to mean white papers, "top 10 tools" lists, and recycled content people used to pay for. A real offer carries genuine standalone value.
The building blocks come from what he calls outbound assets: case studies, proof, data, and anything else you already have that strengthens the offer before you send it.
| Element | What it means |
|---|---|
| The $100 test | Would someone pay $100 for the offer alone? If no, it is too weak |
| Outbound assets | Case studies, proof, and existing material that make the offer credible |
| Value before the ask | Give something real before requesting a reply or a meeting |
| Not a lead magnet | Avoid recycled white papers and generic list-style downloads |
To make it concrete, he described the offer that drove roughly 50K to 60K in revenue in about 14 days when he ran his own agency serving marketing agencies. Instead of claiming to be the best, he offered to actually book meetings for a specific type of company and show the numbers, something competitors talked about but never proved. He ran a campaign, the client got meetings, and within a day they were asking where to send the payment.
Service businesses have it easier here, but the principle extends to product companies. For an HR tech tool that only shows value once installed, he suggested building a partner scheme: run outreach for partners, give them the tool, let them install it, and let the value pull them in. Even in awkward cases, there is almost always an angle to give something valuable first. The worst case, as he put it, is that you tried to give value to someone who did not want it, and then they forget about you.
What Is the Instantly Platform?
Instantly started as an email sequencer and has grown into a full outbound stack. Reid walked through the pieces in the order a user would touch them.
| Component | What it does |
|---|---|
| Instantly Copilot | An in-workspace assistant that answers questions, drafts outbound emails, scrapes domain and email lists to buy, and surfaces best-performing campaigns. Reports can auto-send to Slack or Teams |
| Campaign sequencer | The original product: add leads, build sequences, use AI to generate them, plus a Spintax checker and spam checker and templates |
| SuperSearch | A lead database with waterfall enrichment across multiple top data providers, full-profile enrichment, and AI copy variants |
| Email account setup | Buy domains or upload a CSV, get mailboxes added automatically, then run them through warmup |
| AI agents | The sales agent, reply agent, and deliverability agent that automate the work end to end |
| Visual automations | An n8n-style builder with triggers and actions to push leads into other tools |
A few of these are worth expanding.
How does SuperSearch replace a multi-tool stack?
The usual outbound setup chains several tools together: pull leads from one source, push them into an enrichment tool, enrich, then export to a sending platform. SuperSearch collapses that into one place. It talks to the top databases and runs a waterfall across them, which matters because different providers are strong in different regions. US data is the easiest to get; European and other markets are harder, so using multiple providers in sequence maximizes how much valid data you find. Once a contact is found, you can enrich the full profile and then use AI to build relevance.
Relevance is the point he stressed most about copy. The tighter you can define who you are talking to, the better. Rather than "we work with marketing agencies," you narrow to SEO agencies, then SEO agencies in a specific state, then SEO agencies in that state working with a specific niche. That specificity is what makes the email feel written for the reader, and Instantly bundles the tools to do it in one workflow.
Why does email setup and warmup matter?
Setting up cold outreach mailboxes has traditionally been a headache: buy the email, connect it to a domain, wire up the settings, and deal with disconnections. Instantly internalized this. You choose domains or upload a CSV, mailboxes get created and added, and they go straight into a 14-day warmup where the system simulates real mailbox interaction to build reputation. Launch mailboxes cold without warmup and Google and Microsoft will flag them as spam. The platform also lets you diversify sending across Gmail, SMTP, and Azure and match different accounts to different prospects to lift reply rates.
The AI Sales Agent: Did It Really Beat the Human Team?
This was the part Reid called the real stuff. Instantly has analyzed an enormous volume of email, on the order of a billion messages a month sent through the platform, and trained its own AI sales agent on that data. His striking claim: inside his VIP team, the AI sales agent, given the team's framing and guidance, has actually outperformed the human team. He described that as both nuts and a little scary.
Setup is minimal. You give it your domain, it wakes up, learns the business, and starts finding leads. From there it can engage leads and book meetings on autopilot, connecting to your scheduler. It runs in two modes:
| Mode | How it works |
|---|---|
| Human-in-the-loop | The agent finds and drafts, then waits for your approval before leads run or replies send |
| Autopilot | The agent finds leads and engages automatically, within the accounts, limits, and deliverability settings you set |
Inside the agent, a memory holds your business offers, customer profiles, and guidance, which is where you add instructions to improve output. It integrates with Slack for daily output notifications and supports webhooks so you can watch meetings get booked. If you run several offers, such as a partnership offer plus two others, you can point separate agents at different audiences and let them run continuously, with one person owning the system rather than building campaigns by hand.
Two more agents round out the automation:
- AI reply agent: Handles replies on running campaigns. You train it on how to respond, run it in human-in-the-loop mode inside Slack to approve or edit sends, and over time it converges on output you trust, then runs on autopilot with daily and weekly reporting.
- Deliverability agent: The part he says most people neglect and considers the most important. It tracks everything you send, flags anything that will hurt deliverability, and fixes it automatically. It also runs automatic placement tests to produce a live health score, so you can spot burnt-out domains and swap them quickly.
On top of the agents sits an n8n-style visual automation builder with triggers and actions, so you can push leads into multi-channel LinkedIn outreach or into tools like HubSpot. The idea is that you manage the system rather than build individual campaigns, since lead lists and campaigns keep updating on their own.
How Do You Keep Cold Email Deliverable?
Deliverability is getting harder, and Reid frames that as good news: less outreach reaching the inbox means more room for the messages that do land. His tactics for staying in the inbox:
- Spintax everything. Vary the wording heavily so no two emails look identical.
- Spin the sentiment, not just the words. Google and Outlook already know the major outbound tools and look at sentiment to detect cold email. Repeating the same sentiment over and over is a signal. Changing how something is said, not just which synonyms you use, helps. This is part of why the AI sales agent works well, since every email it writes is genuinely unique.
- Diversify your infrastructure. Spread sending across providers so that if one gets blocked, your whole operation does not go dark. He has seen stretches where nothing inboxed for months on a single setup.
- Expect lower reply rates on enterprise. Large companies spend more on security, so those numbers will always run lower.
On the brand and volume concern, that sending endlessly from rotating inboxes can damage how a brand is perceived, his answer was discipline. He typically sends one email and one follow-up. If there is no response, the prospect is either not interested or not in a buying window, so it is better to follow up in about three months with a different offer than to hammer them four, five, or six times. Reply rates tank after the first couple of emails anyway, so aggressive follow-up lowers domain reputation without winning deals. And if the offer genuinely delivers value, few people resent receiving it.
How Should You Use Buying Signals?
On signal-based, intent-driven outreach, Reid cautioned against over-engineering qualification. Too many enrichments and qualification markers just cut down your addressable market. His approach is to run two motions at once rather than choosing between them.
First, find your best offer at scale. Test around 15 different offers, isolate the winner quickly, and lock that in as a known variable, the way you would isolate variables in an experiment. Then apply that proven offer to signal-based targeting, which he prefers to call inbound intent. Look at your outbound assets first: a large LinkedIn following, web visitors, and similar owned signals. If you do not have those, look further out, for example scraping the followers of leaders in your space and qualifying down to the right fit. Tools like Trigify can surface intent, and one campaign he described simply reached everyone who commented on well-known posts with a specific credit offer, which generated a lot of leads.
The key is to run both the total-addressable-market motion and the signal motion at the same time, segmented across each other. Signals improve your positive reply rate, but broad reach drives daily lead volume, and you usually want both.
What About Data Governance?
A theme that ran through the discussion: pulling data from many providers and scraping across sources sounds powerful but creates real cost and exposure. Waterfall enrichment does not do data governance for you, so the burden falls on the team, and errors burn credits and API spend fast. Reid's own stack uses an email finder first and falls back to vendors like Prospeo when needed, but his favorite data vendor is Better Contact, a single tool that scrapes broadly, verifies catch-alls, and charges only for verified leads. He also floated an idea worth building: prioritizing data providers by the lead's location, since different providers are stronger in different regions, to raise the probability of finding valid contacts. The broader takeaway is that as more of this gets automated, clean CRM data and a real ops function that owns governance become more important, not less.
From Managing Campaigns to Orchestrating Agents
Reid closed on where this is heading. He started his career building a direct mail tool, and even then the offer was king. The medium keeps changing, from physical mail to email to whatever comes next, but the fundamentals hold. What is changing now is the operator's job: from writing manual emails and managing campaigns to orchestrating agents, much like people are learning to orchestrate tools and MCPs with AI assistants.
That shift does not erase the SDR. His argument is the opposite. If an SDR is just sitting by a laptop, the AI can do that job. The agents are good at the repetitive work, but they cannot shake hands, sign a deal, or meet a prospect at 2am in the pub after a conference, which is where a lot of deals actually get done. Automating the boring part frees people to do the parts that need a person, and he thinks that human work only gets more valuable as the rest gets automated.
Key Takeaways for GTM Teams
- Lead with an offer, not a pitch. Run it through the $100 test: if nobody would pay $100 for the offer alone, it is too weak.
- Relevance beats volume of words. Narrow your targeting until the email feels written for one specific reader.
- Warm up mailboxes and diversify sending. Fourteen days of warmup and spreading across providers protects deliverability.
- Spin sentiment, not just words. Detection now looks at how you say things, so genuine variation matters more than synonyms.
- Limit follow-ups. One email plus one follow-up, then revisit in about three months with a new offer.
- Run TAM and signals together. Prove your best offer at scale, then apply it to intent-based targeting, segmented across each other.
- Own your data governance. Automation raises the value of clean CRM data and a real ops function.
Want to Attend the Next Session?
GTM Clawllective is a community for go-to-market operators building with AI, with live sessions where practitioners demo what they are actually shipping. Sessions are not recorded, so the discussion happens live and these recaps are the written record. If you want to join the next one, join the community.
About the Speaker
Aaron Reid leads the VIP team at Instantly.ai, the company's managed service arm, and is part of its executive leadership. His team has helped more than 700 fast-growing organizations around the world scale revenue through outbound. He began his career building a direct mail tool and later ran his own outbound agency serving marketing agencies before joining Instantly, where he had a direct hand in shaping products like SuperSearch and the AI sales agent.
Originally from Scotland and frequently on the move, Reid is active on LinkedIn, where he posts regularly on outbound strategy and AI in sales.
- LinkedIn: Aaron Reid
- Instantly VIP managed service: instantly.ai/vip
Watch the session replay: This AI Sales Agent Outperformed a Human Outbound Team