What an AI SDR Actually Automates

An AI SDR is software sold to do the repetitive part of a sales development rep's work: choose the accounts, write the first message, send it, and run the follow-ups. The category is young enough that no two products mean quite the same thing by the term, which is why the label tells you less than the list of jobs underneath it.

The disclosure first. We build Keaz Signals, which sells buying signals for ecommerce outreach and includes an agent that writes per-store copy. We are inside this category and we have an interest in how you read it. What follows is a breakdown of the four jobs an AI SDR is asked to do, and an account of which ones the technology handles well today. Two of them it does. One it does partly. One it usually does not touch at all, and that one decides whether the other three were worth anything.

Strip the marketing and an AI SDR is asked to do four things, in sequence.

• Sourcing, deciding which companies belong on the list at all. Automates partly. A tool can filter a pool by platform, country, size or installed software. It cannot tell you the segment is wrong, and a precise filter over a stale pool returns a precise list of the wrong companies.
• Timing, deciding that this company should hear from you this week rather than next quarter. Usually not automated. Most products treat the list as static and let the sequence schedule stand in for timing.
• Writing, producing a first message specific enough that the recipient can tell it was not sent to ten thousand people. Automates well, and this is the part that genuinely changed in the last two years.
• Sending and following up, which means mailbox rotation, warmup, throttling, and stopping a sequence the moment someone replies. Automates well, and has done for longer than the AI label has existed.

Note the shape of that list. The two jobs that automate cleanly are the two that were always mechanical. The two that resist automation are the two that require a judgement about a specific business, which is also where the money is.

A firmographic filter answers a question about a state. This company runs Shopify, sells apparel, sits in Germany, employs eleven people. That state was true last month and will probably be true next month, which is exactly why it cannot tell you when to write. Everyone with the same filter is looking at the same list on the same day.

An event answers a different question. The store started running Meta ads three days ago. It launched a product. It installed a retention tool. It rebuilt its storefront. Those facts carry a date, and the date is the reason the message is welcome rather than annoying. Our own pool sits at 4.1 million Shopify and WooCommerce stores across Europe and North America, re-scraped weekly, as of August 2026. The size of the pool is not the interesting number. Which of them did something this week is.

So here is the one test worth applying to any AI SDR you evaluate. Ask what makes it send today rather than yesterday. If the only answer is that today is the next slot in a schedule, then the product automates writing and sending, and you are supplying the timing yourself whether you planned to or not. That is a legitimate purchase, but it is a different purchase from the one the marketing describes.

The marketing in this category runs on reply rates. Three times more meetings. An eight per cent reply rate. A pipeline figure in the first month. We are not going to give you one, and it is worth saying why in the text rather than in a footnote.

We have not measured what our agent does to your reply rate, with your offer, to your audience, from your domains. Neither has anyone else who is quoting you a number. A reply rate is a property of a specific offer sent to a specific audience from a specific sending setup, and the vendor controls exactly one of those three. A figure produced under someone else's conditions is not evidence about yours. It is decoration.

What we will tell you is what the product does. The agent writes from the signal plus a knowledge base you supply, and sending runs through your own Instantly workspace, so replies arrive in your inbox and campaign status syncs back. Whether that beats what you send today is a question your first month answers, not our homepage.

Several of these rule us out specifically. They are worth stating before you spend a week on a trial.

If the people you sell to are not ecommerce stores, we are the wrong tool. Our pool is Shopify and WooCommerce, in Europe and North America, and nothing else. If your buyers live on LinkedIn, we have no LinkedIn signals and no LinkedIn sending, and no plans to add either. Buy the tool that watches LinkedIn.

If you need the leads outside the system, we do not export. Leads move into campaigns and stay there. For some teams that is a dealbreaker, and it should be.

And one that is true of every product in the category, ours included. Automating the top of the funnel does not create capacity to work the bottom of it. If nobody has time to answer the replies, more replies is not an improvement, it is a backlog with better branding. The clearest sign this kind of tool fits is paid sales capacity that is not fully booked. If your calendar is filled by referrals, you probably do not need one yet.

Concretely, for an agency selling a service to ecommerce brands, the whole thing is five decisions and one piece of writing.

1. Pick one signal that means somebody needs what you sell. If you sell creative, a store that just increased its ad count has a reason to talk this week. If you sell retention, a store that just installed an email tool is mid-project and has a budget open.
2. Define the segment narrowly. Platform, country, size, and whatever else separates a plausible client from a name.
3. Write the knowledge base once. This is the piece nobody can do for you: what you sell, what it costs, who it is not for, and the two or three sentences you would actually say on a first call.
4. Connect your own sending infrastructure and let it warm up before you point volume at it. This is the step people underestimate. Domains and inboxes take weeks, not an afternoon, and no agent can shorten that.
5. Let the agent draft, and read the first fifty messages before you trust the next five hundred.

What the agent writes from, and how it is set up, is described on the sales agent page. Every event we currently detect is listed in the buying signal catalogue, which is the honest place to check whether step one is possible for your offer before you commit to anything.

• Keaz Signals product pages, /signals and /sales-agent and /ecommerce-leads-database, retrieved 31 August 2026
• Own pool figures: 4.1 million Shopify and WooCommerce stores under watch across Europe and North America, re-scraped weekly, as of August 2026
• No third-party performance figures are cited in this article, by design. The reason is in the section on reply rates.