We let an AI SDR run our outbound for a week. Here's what happened

The short answer: We let an AI SDR run our outbound for a week, purely as an experiment. By day 3 it had been told off in four languages. By day 5 people were blocking it. The problem was behavior, not wording: it did not stop at NO, and it wrote to people with no reason to write. What works for us is signal-based outreach, where a human writes to someone because something real happened, using Bitscale to find the signal and the verified contact.

• Day 3: told off in four languages
• Day 5: recipients were blocking it
• Root cause: no stop rule at NO, and volume without a reason to write
• What we do instead: signal-led messages, a human in the loop, verified data

Why I ran the experiment

Sales leaders keep asking me whether AI in outbound actually works. Most of the pages that try to answer "do AI SDRs work" are lists of AI SDR tools, written by people selling one.

I wanted a first-hand answer instead of another list. So we let an AI SDR run our outbound for a week, purely as an experiment. I'll keep this to what happened, because the details are the useful part.

If you want the broader market view first, we've covered whether AI SDR tools actually replace human SDRs in 2026 separately. This page is the field report.

What happened, day by day

The setup. The AI SDR ran our outbound for one week. We treated it as an experiment, not a rollout.

Day 3: the replies. They were the rudest replies we've ever got. By day 3 the agent had been told off in four languages. I wrote at the time that it was "not even selling but straight-up haunting people."

Day 5: people blocked it. Rude replies are at least replies. Blocking is the market telling you it doesn't want to hear from you again, and a block on a cold channel is hard to undo.

End of the week. The experiment ended, and I didn't need a dashboard to read the result.

I posted about it on LinkedIn on October 1, 2026, and the post found an audience: Buffer recorded 12,185 impressions and 9 comments by October 28. Plenty of people had seen the same pattern from the receiving end.

Why it failed: my take

I don't think the copy was the problem. Two things were.

It doesn't stop at NO

My one-line summary on LinkedIn was "bro doesn't stop at NO." A good human SDR who hears "not interested" moves on, because they know what the next message costs. The agent treated a no as one more step in the sequence. My read is that this is what made the replies so hostile: a follow-up after a refusal tells the person you weren't listening.

This is a guardrail problem, and it is also a legal one. More on that in the checklist below.

Volume without a reason to write

The second failure was quieter. The agent had a list, not a reason. On October 26 I put the difference this way: cold email that works and cold email that doesn't are separated by a signal, "a real reason to write to that person."

Without one, nobody replies, not even to say no. With one, even a refusal is useful. A signal-led email gets answers like "not now, we just signed with X" or "wrong person, try another." Both tell you what to do next.

Personalization doesn't fix it

The usual response to a failed AI campaign is to tweak the prompt. In September I wrote that sales leaders keep asking whether AI personalization works, and my answer was that it doesn't. Teams treat it as a copy issue because tweaking prompts is easier than anything else.

What most teams get wrong is the metric. Ignore open rates and send volume. The only metric that should matter is positive replies from accounts with actual intent. And if you're pulling from a single vendor list, no amount of clever AI writing will save you. Fix your data first.

My co-founder Sanket made a related point on September 29: the biggest software companies are putting engineers on customer sites because AI doesn't work out of the box, and he called that "the exact reason that AI SDRs died." I'd put it a little more gently. The agents run. What they can't do on their own is decide who deserves a message.

There's more on the failure patterns in why AI SDR implementations fail.

What works instead: signal-based outreach

Here is the side-by-side, using only what we saw in the experiment and what we run today.

Job changes: one congrats DM, one pilot

On October 20 I shared a small example. One congrats DM turned into a pilot. The signal behind it was a job change, just one of the intent signals we track.

The logic is simple. People who have used your product and liked it don't just leave; they carry you to the next place. A short, human note when they start a new role is welcome in a way that a fifth automated follow-up never is. We wrote more on this in what champion tracking is.

Job changes aren't the only trigger. The ones I watch beyond hiring are funding rounds, new leadership, champion job changes, competitor engagement, tech stack changes and new product launches.

Three campaigns, not fifty

Sanket shared the other side of this on October 1: "we made $1.5M with just 3 campaigns." His contrast was that most teams run 50 generic campaigns a month and get zero back. In his words, the team "skipped the job titles, targeted actual intent signals instead."

Put our two October 1 posts next to each other and the lesson is hard to miss. The week of volume got us blocked. A few campaigns built on signals are the ones Sanket credits with $1.5M.

If you still want an AI SDR: a guardrail checklist

I'm not against AI in outbound; research and drafting are where it helps most. If you're evaluating AI SDR software, here is what I'd insist on before letting it send a single message.

1. A hard stop on NO. Any reply that says no, not interested, remove me or wrong person ends the sequence for that contact immediately. Test this before launch with real negative replies.
2. Opt-out handling that meets the law. In the US, the FTC's CAN-SPAM guide says every commercial email must include a clear and conspicuous way to opt out, you must honor opt-outs within 10 business days, and the message must include a valid physical postal address. The FTC also states that "the law makes no exception for business-to-business email."
3. You own what the agent sends. The FTC is explicit that you "can't contract away your legal responsibility" by hiring someone else to send email, and that both the company promoted and the sender may be held responsible. Each violating email can draw penalties of up to $53,088.
4. Objections in the EU end the processing. Under GDPR Article 21, a person can object at any time to processing for direct marketing, and after an objection the data "shall no longer be processed for such purposes." Your agent's suppression list has to carry that across every sequence.
5. Signal triggers, not lists. Only start a sequence when a defined signal fires: a job change, funding, a new leader, a tech-stack change. No signal, no message.
6. Human review before the first touch. A person approves who gets contacted and why. The agent can draft and research; the human decides.
7. Verified data. Single-source lists put the wrong person in front of your agent. Verify emails before sending, because bounces and complaints both cost you deliverability.
8. Watch spam complaints, not opens. Google's sender guidelines ask senders to keep spam rates reported in Postmaster Tools below 0.3%. A campaign that gets people blocking you will find that line quickly.
9. Score on positive replies from intent accounts. Drop open rate and send volume as success metrics.

If you're weighing an agent against a hire, our AI SDR vs human SDR comparison and SDR salary breakdown cover the cost side.

Where Bitscale fits

Our conclusion from the week was the line I ended the LinkedIn post with: "human + Bitscale is the combo GTM teams love." The agent was missing judgment and a reason to write. Those are the two things we built Bitscale to supply.

• Signals that give you a reason to write. Bitscale tracks 15 buying triggers, including exec hires, tech-stack changes, funding and hiring posts, and they can be stacked. The ready-made Champion Job-Change Tracker playbook watches for the kind of move that produced our congrats-DM pilot.
• Verified data so the right person gets the message. Bitscale runs 18+ contact-data providers in one email and phone waterfall, with 95% email and phone coverage and bounce rates below 0.5%.
• Research before the first touch. The AI research agent pulls company details, decision-makers, tech stack, funding and hiring data into the list, so the human writing the message knows why it matters now.
• CRM sync. Two-way HubSpot sync from Growth and Salesforce from Booster write back only changed fields and flag job changes and bad emails.

Pricing is per account with unlimited actions on paid plans: Free with 200 credits, Growth at $349 a month ($314 billed annually) with 15K credits, Booster at $799 ($719 annually) with 50K credits, and Enterprise from $15K a year. Credits roll over up to 3x.

Frequently asked questions

Do AI SDRs work?

In our one-week experiment, no. By day 3 the AI SDR had been told off in four languages and by day 5 people were blocking it. In my view AI works for research and drafting, while a person decides who gets a message and why.

Why did your AI SDR get such rude replies?

It didn't stop at NO and it had no reason to write to most people. It kept following up after refusals, which I described as haunting people rather than selling. Rude replies and blocks followed within five days.

What should I look for in AI SDR software?

A hard stop on negative replies, compliant opt-out handling, signal-based triggers, human review before the first touch, verified contact data and reporting on positive replies from intent accounts rather than opens or send volume.

Is cold email from an AI SDR legal in the US?

CAN-SPAM covers business-to-business email too. Each email needs a clear opt-out and a valid postal address, opt-outs must be honored within 10 business days, and you stay responsible even when a tool or agency sends for you.

What is signal-based outreach?

Writing to someone because something real happened, such as a job change, funding round, new leadership or a tech-stack change. The signal gives the message a reason, so even a no tends to be useful, like "wrong person, try another."

How does Bitscale help with AI SDR guardrails?

Bitscale supplies the inputs an agent lacks: 15 buying triggers, verified contacts from an 18+ provider waterfall and an AI research agent. A person then decides who to contact, so messages go to people with a reason to hear from you.

Sources

First-hand posts (LinkedIn, via Buffer export)

• Abhinay, "the rudest replies we've ever got" (the AI SDR experiment), October 1, 2026
• Abhinay, on AI personalization and the metric that matters, September 15, 2026
• Abhinay, "1 congrats DM turned into a pilot," October 20, 2026
• Abhinay, on signal-based cold email, October 26, 2026
• Sanket, "we made $1.5M with just 3 campaigns," October 1, 2026
• Sanket, on vendors deploying engineers and AI SDRs, September 29, 2026

Compliance and deliverability (opened October 2026)

• FTC: CAN-SPAM Act compliance guide for business
• GDPR Article 21: right to object
• Google Workspace: email sender guidelines

Bitscale

• Bitscale pricing
• Bitscale playbooks
• Bitscale waterfall enrichment