Editorial policy

Every tool in this directory is reviewed against the same process. We do not review from vendor demos, feature pages, or other comparison sites. We run the tools the way a sales team would, from finding leads to measuring what a booked meeting actually cost.

Our review process follows six stages:

1. Setup and infrastructure. We test every tool from identical, purpose-built infrastructure: fresh sending domains with SPF, DKIM and DMARC configured, warmed for two to three weeks before any test begins. Deliverability differences between tools are therefore attributable to the tool, not to our setup.
2. Lead finding and data accuracy.We run the same benchmark list of 50 companies we know well through each tool’s data and research features. We verify emails, direct dials and firmographic fields against ground truth, and we count hallucinations — wrong funding stage, wrong product, invented pain points — rather than accepting vendor accuracy claims.
3. Personalisation and writing quality.We grade AI-written outreach on a fixed 1–5 rubric across the full benchmark list: is the personalisation specific to the company, or a template with the name inserted? Scores are comparable across every tool we review.
4. Sending and deliverability.Each tool’s sequences are sent to a seed panel of 30+ inboxes we own across Google Workspace, Microsoft 365 and independent mail hosts. We record inbox versus spam placement per provider, and whether the tool’s tracking behaviour triggers filtering. No test emails are sent to people who have not agreed to receive them.
5. Live pipeline testing.Where a tool claims to book meetings or manage replies autonomously, we run it on real, consented campaigns — our own outreach or a partner company’s pipeline — with a fixed offer, list size and duration. We read every AI-handled reply thread and record positive reply rate, meetings booked and meetings held. Reply-rate figures from these samples are reported as directional, and we say so.
6. Operations, billing and exit. We test what buyers only discover after paying: whether suppression lists sync from the CRM, who owns the sending domains, what the tool logs about its own activity, how support responds on the tier we paid for, and what data leaves with you if you cancel.

What we do not claim

This is not an exhaustive benchmark. We do not publish a performance league table, a resolution-rate ranking, or a word-error-rate comparison, because those need sample sizes and controlled conditions we do not have across 264 products. Where a figure comes from one of our own samples rather than from a vendor, it is labelled directional.

What this site is good at is narrower and more useful: telling you what a tool actually costs, on a stated date, against what it advertises. Every rate here was read off the vendor’s own page and carries the day we read it — 111 of 150 priced tools currently carry that date. The all-in figures are arithmetic we show our working for, computed from published component rates rather than measured from an invoice. Those two things are never blurred, and a computed figure is never presented as a measured one.

Where a tool is gated behind a sales call, that is a finding, not an obstacle — and we say so in the review, in those words. 108 of the 264 tools here cannot be bought without talking to somebody first. A directory that quietly omitted them would be reporting on an easier market than the one you are buying in.

Two numbers appear on this site and they come from different places. Keeping them apart is the point of this section.

The review rating, from testing

Tools we have taken through the six stages above are rated on a 1–5 scale using weighted criteria:

• Data accuracy and deliverability — 30%
• Personalisation and reply handling quality — 20%
• True cost versus advertised price — 20%
• Operational transparency (logs, suppression, domain ownership) — 15%
• Setup, integrations and documentation — 10%
• Support quality and exit terms — 5%

The transparency score, computed

Separately, every one of the 264 tool pages carries five bars. Those are not a performance rating and they are not derived from testing. They are computed at build time from published material, so that a tool nobody has run yet still gets a comparable, checkable score rather than a blank. The weights:

• Price transparency — 26%. Published rate, and whether you can buy without a sales call.
• Cost honesty — 22%. How close the advertised rate lands to the computed all-in figure.
• Capability — 20%. Whether it completes work, hands off, or only reports.
• Independence — 16%. Whether it still sets its own roadmap.
• Cost predictability — 16%. Whether the bill stays proportional as usage grows.

These percentages are read directly out of src/lib/rating.ts when this page is built, so the number you see here is the number the bars were drawn with. It is shown as five bars rather than one figure because a single number is unfalsifiable and five are auditable: you can disagree with one and keep the rest.

Scores are derived, never assigned. No editor can nudge one, and no commercial relationship can reach them — the scoring code cannot import the commission data, and the build fails if it ever does.

• We pay for what we test. Vendors do not provide free accounts, do not see reviews before publication, and cannot pay to appear or to rank.
• Prices are re-verified quarterly.Every price on this site carries the date we last confirmed it against the vendor’s own pricing page or contract data.
• Claims are checked against primary sources.When a vendor claims an accuracy rate, a compliance status or a capability, we check it against their own published documents and our own tests, and we publish the verdict — including “not addressed in published material.”
• Corrections are made and dated. If we got a price, a feature or a claim wrong, tell us and the correction is published with the date it was made.
• Ethics of testing. Our deliverability panels are inboxes we own. Live campaign tests run only on legitimate, consented outreach. We do not send unsolicited email for the purpose of writing reviews.

Every review shows a last verifieddate. That date is the day somebody opened the vendor’s own pricing page and read it, not the day the file was last edited — a date that moves on every deploy teaches you to ignore it.

Prices are re-checked quarterly. The audit that runs on every build flags any rate older than 90 days, so a stale figure surfaces as a warning to us before it misleads you. Of 150 priced tools, 111currently carry a verification date; the rest say “not re-checked” on the page rather than implying a freshness we cannot support.

A re-review is triggered by any of: a pricing page change, a tier being renamed or withdrawn, an acquisition or shutdown, a reader-reported correction, or the 90-day clock running out. When a figure changes materially, the page is re-dated and the change is described — the old number is not quietly swapped out.

If something here is wrong, tell us at corrections@ai-sdr.cc. A price, a tier, a claim, a dead link — all of it. Corrections are made on the page and dated, and where the error was material the correction is described rather than quietly absorbed.

A worked example. Synthflow was recorded here as $29 a month, self-serve, with published tiers. It is not. The published entry is an annual enterprise contract with a $30,000 a year floor, five seats, and no per-minute rate published anywhere. The entry price and the per-minute field were emptied rather than estimated, the contract terms were written up as their own record, and the page carries the date of the re-check. The error mattered because a $29 sticker put Synthflow in the same mental bracket as tools costing a thousandth as much, which is exactly the mistake this site exists to stop.

The blog is where work that spans more than one tool goes: a count across all 264products, a cost model with its arithmetic shown, a careful read of somebody else’s pricing page. It answers questions a single review cannot, such as how many tools publish a price at all, or what a voice minute costs once every component is added up.

It is held to the same rules as the review surface. Every figure carries the date it was read. Where a source has an interest in the answer, the source list says so — which matters more here than anywhere, because almost everything published about this subject is written by somebody selling into it. Posts are dated on publication and re-dated when corrected.

Some links on this site can be affiliate links, and we may earn a commission if you buy through one. It never affects ratings, rankings or ordering — nobody can pay to appear here or to rank higher. It costs you nothing extra.

These relationships never influence ratings, rankings or editorial content. Tools with and without affiliate programmes are evaluated identically. Nobody can pay to appear here, to move up, or to soften a review. There are no sponsored posts, no vendor-supplied copy, and no drafts sent to vendors before publishing.

That claim is enforced in the code rather than promised in prose. The commission data lives in a module the scoring logic is forbidden to import, and the build fails if that boundary is ever crossed — so if the ordering had been for sale, this site would not compile. No commission programme is live at the time of writing, and the boundary exists so that this stays checkable when one is.

AI SDR Stack is written and maintained by Anusha Rathi, Editor. 55 of the 264 tools here carry a written review; the rest are listed with their published figures and are marked as not yet reviewed rather than padded out.

Questions, submissions and partnership enquiries go to hello@ai-sdr.cc. Corrections go to corrections@ai-sdr.cc, which is read first.