Why didn't AI SDRs fix outbound?

AI SDRs did not fix outbound because they automated the wrong part. Adoption went up and reply rates went down, because most teams used the tools to send the same mediocre message to more people, faster. Volume was never the bottleneck. Data quality, deliverability and knowing something true about the person were, and AI made all three easier to skip.

I said this in a post on my LinkedIn, and it is the thing nobody wants to hear. This is the longer version, and what to do about it.

What was the real bottleneck in outbound?

The real bottleneck in outbound was never how many emails a team could send. It was three things that come before sending: data quality, so the right people are on the list; deliverability, so the messages reach an inbox; and relevance, actually knowing something true about the person you are writing to.

Each one is unglamorous work.

• Data quality means the companies fit, the contacts are the right people, and the emails are verified. Most lists fail here first.
• Deliverability means the domains and mailboxes are set up and warmed, and the sending stays within what they can carry.
• Relevance means there is a reason to write to this person this week, and the message says what it is.

None of these is solved by sending more. All three get worse when you do.

Why does sending more make outbound worse?

Sending more makes outbound worse when the message is the same mediocre one, because every extra send to the wrong person costs something. It teaches inbox providers your mail is unwanted, it burns through the market you could have reached properly, and it trains buyers to ignore anything that looks like it.

Those two trends, more teams running AI SDRs and lower reply rates, should not move in opposite directions. But they are, and this is why. More volume on top of weak data and a generic message does not average out. It compounds.

The uncomfortable part is that AI did not create any of these problems. Bad lists, tired domains and generic messages existed long before. What changed is that the tools made it cheap to skip the hard work and still feel productive, because the dashboard shows thousands of sends.

What are the teams still converting doing differently?

The teams still converting are not running the fanciest stack. They are the ones who did the unglamorous work first: real signals, real data and a real reason to write, all in place before a single sequence went live. The tools came after the targeting, not instead of it.

In practice that looks like:

1. A list built from who actually buys, not from a brainstorm. We start from closed-won deals, which I wrote up in finding your ICP in your closed-won deals.
2. Contacts found and verified properly, with several data providers in sequence rather than one. Mike covers that in what waterfall enrichment is.
3. A signal behind every message, such as a new hire or a funding round, so the first line says something true. The signal-based list post covers which ones count.
4. Sending infrastructure that can carry the volume, set out in how many sending domains outbound needs.

That gap between the teams who did this work and the teams who skipped it is only getting wider.

Where does AI actually help in outbound?

AI helps in outbound wherever it saves time on work that is already pointed in the right direction. It is good at sorting and cleaning lists, finding patterns in deal data, researching a company's website and turning a true fact into a short, natural message. It is not good at deciding who to target or why.

AI is a lever. It is not a strategy. A lever multiplies the force you put on it, so the question is always what it is attached to.

We use Claude every week for list work and deal analysis, and Clay for enrichment. Both make good outbound faster. Neither makes bad outbound good. Building the parts that come first, the data, the signals and the infrastructure, is what we do in GTM engineering.

What should you check if your numbers are flat?

If your numbers are flat despite adding tools this year, check the three things that come before sending. The tool is probably not the problem. Look at the list, the inbox placement and the reason behind each message, in that order, because each one limits everything after it.

• The list. Pull twenty contacts at random. Would you be glad to talk to each one? Are their emails verified?
• The inbox. Are your messages landing in the inbox, or in spam? Are your domains and mailboxes healthy?
• The reason. Read the first line of ten sent messages. Does each say something true about that person, or could it be sent to anyone?

Whichever fails first is the one to fix first. Adding another tool before fixing it adds cost, not replies.

What should you try first?

Take one week's worth of sequences and stop sending to anyone without a reason. Keep only the contacts where you can name a signal or a specific fact about their company, and rewrite the first line around it.

Send less, to fewer people, with something true to say. Then compare the replies with the week before. What is actually working for your team right now? Tell me on LinkedIn.