The short version
If you're under roughly €200k ARR, or still finding out which message gets answered. At that stage you need the learning more than the volume, and the tools are cheap. A full stack of data, sending and automation tools runs €400 to €800 a month.
If the offer is simple, the market is large, deals are small, and someone in-house reviews what it sends every day. It fits badly when one clumsy email to the wrong account costs you a deal you'd otherwise have won.
If you've closed enough deals to know who buys, your own time is the bottleneck, and you want one party accountable for the list, the sending infrastructure and the replies, not three tools and a hope.
Side by side
The rows are the questions that decide whether outbound books calls. The features list on a vendor's pricing page mostly isn't.
| Question | Do it yourself | AI SDR platform | Done-for-you partner |
|---|---|---|---|
| Who builds the list | You, out of a data tool like Apollo. | The platform, from its own database and the filters you set. | The partner, starting from the deals you actually closed, then verified. |
| Who decides who gets mailed this week | You. | The model, inside the rules you gave it. | The partner, with your sign-off on every new segment. |
| Which domains send | Whatever you set up. Often the main domain, which is the expensive mistake. | It varies by vendor, so ask before you sign. It's the most important question on this page. | Separate domains in your name, warmed before the first real send. |
| Who watches deliverability | You, when you remember to. | Often nobody by name. Ask who sees the bounce rate and what they do about it. | The partner, every week, before it becomes a problem. |
| Who reads the replies | You, in between everything else. | The platform classifies them, and some answer on your behalf. | Classified and routed to you, with a draft answer waiting for approval. |
| Your time each week | The most of the three, and the first thing to go in a busy month. | Low once it runs, as long as someone reviews what it sends. | Roughly an hour to approve replies and read the review, plus the calls. |
| What it costs | Tool subscriptions, a few hundred euros a month, plus your hours. | A subscription, priced per seat, per contact or per meeting. Read the contract term. | A monthly retainer. Mine is €5k to €15k, build included. |
| What you keep if you stop | Everything. It was always yours. | Check it. Lists, domains and conversation history can live inside the platform. | Should be everything: accounts, domains, lists and documentation in your name. |
| How it usually fails | You get busy, outbound stops, and the pipeline dries up a month later. | Volume without judgement: a burned domain, or the right account hearing from you the wrong way. | A partner paid on meetings booked, optimising for meetings rather than for buyers. |
What an AI SDR actually automates
Researching an account, drafting a first line, sending on a schedule, sorting replies into interested, not now and unsubscribe. That's pattern work, and models are genuinely good at it. I use them for most of it in the engine I run, so this isn't an argument against AI.
What a platform can't do from the inside is know which of your past deals were the good ones, so its idea of your ICP is your filters, not your history. It can't decide that this account should wait three weeks for a better reason to hear from you, and waiting for a real signal is often the strongest move in outbound. It won't spot your existing customer on the list unless someone loaded the suppression first. And it can't take the call.
So the useful question isn't AI or no AI. It's whether a person who knows your market approves what goes out before it goes. If the answer is nobody, you're buying speed with your reputation as the collateral.
The arithmetic every option has to survive
None of the three can sell their way around this. Across 12,973 delivered cold emails we logged a 2.07% reply rate, about one reply per 48 emails, and a reply isn't a meeting. That means a pipeline target turns into a send volume, and a send volume turns into sending domains.
Each of those domains needs warming up before its first real send, and every address needs verifying, because a high bounce rate is what kills a domain fastest. An offer that promises far more than this from less infrastructure is either sending more per domain than inbox providers tolerate, or counting replies differently. Both are worth asking about before you sign. The full numbers and the method are on the cold outreach benchmarks page.
LinkedIn is the other half, and it behaves differently: 21.1% of our connection requests were accepted, and 43.3% of the people who accepted went on to reply. Far stronger per touch than email, and capped by LinkedIn, so nobody can buy more of it for you. Any option that treats LinkedIn as a volume channel will get the account restricted.
Eight questions to ask whichever one you pick
They apply to your own setup too. If you can't answer one about your own outbound, that's the first thing to fix.
- Which domains does it send from, and who owns them if we stop?
- What's the daily cap per sending domain, and is the tool enforcing it, or just good intentions?
- Is every address verified before the first send, and what bounce rate makes you pause a domain?
- Are my customers, my open deals and my first-degree LinkedIn connections suppressed before anything goes out?
- Can the same person get pitched twice, from two campaigns or two channels?
- Who reads a reply, how fast, and does the sequence stop on both channels the moment someone answers?
- When you quote me a reply rate, what's it divided by?
- What do I take with me if I cancel: lists, domains, sequences, conversation history?
For the first two, you don't need to take anyone's word for it. Put the sending domain into the cold email domain checker and it tells you in thirty seconds whether SPF, DKIM and DMARC are in order and whether it's your main domain at risk.