An AI SDR is software that does the work of a sales development representative: it researches companies that could buy from you, finds the person to write to, writes the first email, follows up and sorts the replies. For a small B2B team or an exporter without a sales department abroad, that is most of the work between "we want buyers in Germany" and a first call. The hard part is not sending emails; it is sending the right email to a company that really buys what you sell, without burning your domain on the way.
Contents
- What an AI SDR does
- Where AI SDRs fail
- What to check before you buy
- Why exporters need something different
- How Outfound does the SDR job
- Running it from Claude or ChatGPT
- What Outfound does not do
What an AI SDR does
A human SDR's week is mostly five jobs. An AI SDR is worth buying only if it does each of them well enough that you would sign the result.
| Job | What a human SDR does | What the software should do |
|---|---|---|
| 1. Research | Reads the company's website, news and product range to decide if it fits | Read each company's current site and show why it fits, with the source |
| 2. List building | Finds companies that match the ideal customer, removes the ones that don't | Find candidates in the target market and keep only those that pass the research |
| 3. First email | Writes a short email that shows they looked at the company | Write each email from that company's research, in the buyer's language |
| 4. Follow-up | Writes again after a few days, with something new | Send follow-ups on a schedule and stop when the person replies |
| 5. Reply handling | Reads replies, answers questions, books the call, removes people who say no | Collect replies in one place, sort them, and leave the answer to you |
Most tools sold as an AI SDR do jobs 3 and 4 well and jobs 1, 2 and 5 loosely. That is where results come from, so look there first.
Where AI SDRs fail
The failures are predictable. Each one is easy to spot in a trial if you know where to look.
- Made-up personalization. A language model asked to "personalize" will write something specific-sounding even when it knows nothing about the company: a project it never did, a market it never entered. A buyer who reads one false claim about their own business stops reading. Ask where each sentence came from. If the tool cannot show a page, the sentence is a guess. We wrote about the line between useful context and guesswork in buyer context: when it impresses and when it creeps.
- US-centric data. Many tools filter a contact database built mainly from US and UK companies. For a manufacturer selling to distributors in Poland, the UAE or Mexico, the coverage drops and the "verified" emails are often old. Run your own target country in the trial, not the demo country. Our guide to B2B lead generation tools shows how to test coverage in your market.
- Burned domains. Sending thousands of cold emails from your main domain, without warm-up and without checking addresses, is how a company ends up with its invoices and quotes in spam. Google asks bulk senders to keep the spam rate reported in Postmaster Tools below 0.3% (Google email sender guidelines) and to publish a DMARC record; our DMARC adoption by country study shows how many B2B domains have one. An AI SDR that sends at volume has to protect your domain, not just use it. More on this in your buyer email landed in spam.
- No human approval. "Fully autonomous" sounds efficient until the software writes to your biggest competitor's distributor, or tells a buyer you offer a certificate you do not have. For a small team, every first email going out under your name should be something you have read.
- Wrong-language emails. An English email to a mid-sized German or Turkish buyer can work. An email that was written in English and machine-translated after the fact often reads like a template. Language should be a choice made per market, and the email should be written in that language from the research.
What to check before you buy
Use the trial to answer these questions with your own product and your own target country. A demo with the vendor's example will not tell you much.
| Check | Good sign | Warning sign |
|---|---|---|
| Why this company? | Each company comes with a reason and a link to the page it came from | A score with no explanation, or filters only (industry, size, country) |
| Your target country | Real, active companies in the country you actually sell to | Good results only in the US or UK; thin or old results elsewhere |
| The person | A decision-maker with a role that buys what you sell, email checked before sending | Generic info@ addresses, or no address status shown |
| The first email | Every specific claim traces back to the research; written in the buyer's language | Flattering lines you cannot check; English only; translated templates |
| Approval | You see and approve each first email; you can edit or regenerate it | Sending starts as soon as the list is ready |
| Sending setup | Separate sending domain, SPF, DKIM and DMARC set up, warm-up, automatic pause if problems rise | Sends from your main domain at full volume from day one |
| Replies | Replies stop the follow-ups and land in one inbox you can answer from | Follow-ups keep going after a reply; replies scattered across mailboxes |
| Price unit | You pay for companies that match, with a clear monthly allowance | You pay for every record pulled, matching or not; seat fees on top |
Why exporters need something different
The usual AI SDR is built for a US software company selling to other US software companies: the buyers are in one database, everyone reads English and the ideal customer can be described with job titles and headcount. Selling physical products abroad is a different job.
- The fit is in what the company sells or uses, not in its industry code. A distributor of commercial kitchen equipment that carries Italian cooking lines but no stainless shelving is a buyer for a shelving manufacturer. No filter says that; reading the company's product pages does.
- The buyer is an importer, distributor or manufacturer, often mid-sized. These companies are underrepresented in contact databases and change less visibly online, so their pages have to be read directly. Our guide on how to find buyers for export compares the sources.
- Several markets at once. You may be testing Germany, Poland and the UAE in the same quarter, each in its own language and with its own way of doing first contact (see first contact across cultures).
- Small team. Often one export manager, sometimes the owner. Software that needs a full-time operator does not fit; software that asks for approval and then gets on with it does.
How Outfound does the SDR job
Outfound was built first for exporters. It covers the SDR job by email, from research to the replies inbox, and asks for your approval before anything leaves.
- You describe the buyer. A sentence about what you sell and who buys it, plus the target countries or regions. Outfound reads your website to build an outreach strategy: segments, angles and tone.
- Live web research per company. Outfound does not filter a static database. Each candidate company in the target market is researched on its current website and compared with your buyer description. You see why it fits, with quotes and links to the pages the evidence came from. Only companies that match count against your allowance.
- Decision-maker and verified email. For each company that fits, Outfound finds a person in a role that buys what you sell and checks the email address before anything is sent. Each address shows its status, for example verified, accept-all (catch-all) server, or invalid.
- First email in the buyer's language. Each email is written from that company's research, so its specific claims come from the evidence you can see. Emails can be written in 46 languages; you choose the language per market.
- Your approval before sending. Nothing is sent that you have not seen. You approve, edit, regenerate or reject each draft; only approved people go into a campaign.
- Separate sending domain with warm-up. Emails go out from a separate sending domain set up with SPF, DKIM, DMARC and link branding, so your main domain is not used for cold outreach. A new sending domain starts at 20 emails a day and, with regular use, reaches 1,500 a day in about 15 sending days. If bounces or spam complaints rise, warm-up slows down or the campaign pauses.
- Follow-ups that stop on a reply. You set the follow-up steps and the waits between them in the campaign sequence. When someone replies, their sequence stops at once. People who bounce, unsubscribe or report spam get nothing further.
- Replies inbox. Replies come back to one inbox, linked to the email they answer and sorted, so you can see who is interested and answer from there. The answer, the call and the quotation stay with you.
Across all email sent from the platform between 25 March and 25 September 2026, the delivery rate was 98.8%. We do not publish platform-wide open or reply rates, because they depend on the product, the market and the list more than on the software.
Running it from Claude or ChatGPT
If you already work in a chat assistant, you can run Outfound from there. In Claude, Outfound is a connector on the web, desktop and mobile apps; in ChatGPT, it is added as a custom MCP server on Business, Enterprise and Edu plans (read-only on Pro). You ask in your own words, for example "show me the emails waiting for approval" or "what came in overnight?", and the assistant uses your Outfound account with the same role, plan and limits as the web app.
Drafts appear as cards you can open, approve or reject. Anything that spends credits, sends an email or answers a reply is shown to you first and only goes ahead after your yes. The setup steps are in the help centre: How do I use Outfound in Claude? and How do I use Outfound in ChatGPT?
What Outfound does not do
It is better to know the limits before the trial than after it.
- No phone calls. Outfound works by email. It does not dial, and it does not find phone numbers during discovery; you can connect your own phone lookup provider if you need numbers.
- No social media messaging. First contact is by email only.
- Not a CRM. Replies are collected and sorted so you can answer them, but Outfound does not replace a CRM you already use. Outbound webhooks can send events to Zapier, Make or your own server.
- Not for mass mailing to purchased lists, newsletters or consumer email. It is built for writing to companies one by one, with a reason.
- It does not close the deal. Samples, quotations, payment terms and the first order are still your work. Our guide to the first sample and the first quote covers that stage.
The quickest way to judge any AI SDR, including this one, is to run it on your own product and one real target country. Outfound's trial gives you 20 company discoveries and 40 email drafts without a card, so you can read the research and the drafts before deciding. If you are still building the list by hand, start with our guide to building an export buyers list.
Last updated: · By: Tolga Tatar
