Before you approve the AI SDR pitch, run these four checks

A consultant sent you a deck. The deck shows 14.2% reply rates on AI-personalized outreach versus 3% for humans. It quotes 29.5x ROI. It notes that 83% of human SDRs miss quota. Pricing starts at $900 a month, which is less than a loaded SDR for a week. You run a business doing $2M in revenue with a team of 4. The math looks like it closes itself. You are 48 hours from signing.

Stop.

The pitch is not wrong. The pitch is also not exactly right. It is built from averages that do not describe your business, and it omits the operating costs that land in month 3. Before you approve a line item that will touch your domain reputation, your CRM, and your buyer list, there are four checks worth running. None of them take longer than an afternoon.

The math is a composite, not a forecast

The 14.2% reply rate is real. It is also peak — drawn from well-matched lists, warmed sending infrastructure, and copy a human reviewed before send. Two of those three conditions are not included in your subscription.

When operators rebuild the math from their own pipelines, the distribution collapses. Some deployments hit the advertised numbers. Many deployments hit zero. The failure rate concentrates in businesses that bought the tool to replace a sales motion they had not yet built. The 29.5x ROI number carries the same asymmetry — it is the best of a bimodal distribution, not the center.

A $2M business buying an AI SDR to build the outbound motion is not running the same experiment as a $50M business buying one to amplify an existing motion. The deck treats both buyers as interchangeable. They are not. When you ask the vendor for benchmark data, ask for the distribution, not the average. If the answer is the average, the average is hiding the outcome that matters to you.

The 83% quota-miss statistic for human SDRs

is doing a specific job in the pitch. It is framing AI as the cure for a broken human function. The framing skips a step. The reason most SDRs miss quota is not that humans are inefficient. It is that most outbound programs are pointed at the wrong list with the wrong offer. Replacing the human does not fix either.

The $900 subscription is not what you will pay

Sending at AI-SDR volume requires sending infrastructure. That means at least 3 secondary domains, each with mailbox warmup, running 50 to 150 dollars per domain per month. It means enrichment credits, because the built-in data quality will not survive contact with your ICP. It means a CRM sync layer, because the native integration will miss your custom fields or overwrite them in ways that take weeks to untangle. It means human review time, because signing your name to AI outreach without a review layer is one bad sentence from a public screenshot.

Operators who tracked the all-in cost across 12 months report the $900 plan arriving at $2,500 to $3,000 monthly once warmup, enrichment, deliverability tooling, and API overage land. That number still beats a human SDR's loaded cost. It does not beat the gap between what the founder thought they were buying and what the P&L actually shows when the quarter closes.

Here is the cost nobody prices in. When an AI SDR sprays your warmed secondary domains with 50,000 contacts, 15% bounce rates are common on cold lists. ISPs read that pattern as spam. Your secondary domains get flagged. If you were sloppy enough to let the tool sample your primary domain during setup, the flag migrates. Now your founder-from address — the one you use to close real deals — is landing in spam at your top 40 accounts. You find out when a prospect says "did you ever get back to me on that?" and you did, three times, and they never saw a single email.

This is the most common failure mode in the category. It is also the one failure that cannot be undone inside a quarter. Domain reputation takes months to repair, and during those months your actual sales motion is operating at a fraction of its normal reach. The subscription cost pales next to the pipeline cost.

Who the tool is actually built for

AI SDR platforms are built for companies with three structural conditions. A clean, large, actively maintained ICP list. A dedicated ops function that can babysit deliverability, data quality, and CRM hygiene. And a sales motion that already converts, where the constraint is volume and not targeting.

A $2M founder with a 4-person team, an ICP still being clarified, and no dedicated ops role meets zero of those conditions. The tool is not built for that buyer. The deck is targeted at that buyer.

This is the pattern across the AI application layer right now. The tools are real. The marketing around the tools is calibrated to close buyers the tools cannot serve. The vendor is not lying. The vendor is describing a best-case customer, and most buyers are not best-case customers, and the distance between the average case and the best case is where the churn lives.

Roughly 70% of teams that deploy an AI SDR tool are no longer using it 90 days later. That figure is reconstructed from public post-mortems and sales-ops reports, not from vendor decks. The pattern across failed deployments is not that the tool did not work. The tool worked exactly as described. The deployment still failed, because the tool's output is capped by input quality, and input quality is not a software problem. It is a list-hygiene, positioning, and offer-clarity problem. Fix those and the tool amplifies clean work. Skip those and the tool amplifies mess, at 10x the speed of a human.

The right sequence is always the same. Define the ICP. Refine the offer. Prove the motion manually, with a human sending 50 emails a week and tracking which messages to which titles produce replies that convert. Then automate. Automation compounds whatever is already working. It also compounds whatever is not.

Three questions before the next call

First, what does implementation look like for a company with fewer than 5 ICPs and no dedicated sales ops resource? If the answer leans on "managed services" or "we assign an implementation specialist," ask what happens in month 4 when the specialist is reassigned. The right answer is concrete and short. The wrong answer is a second sales call.

Second, what is the all-in monthly cost — subscription, domains, warmup, enrichment, API overage — to send 5,000 outbound emails per week at 99% deliverability? If the vendor cannot produce the number, the vendor does not know the number, and you will discover it on your credit card in the fourth month.

Third, what is the honest 90-day churn rate in the customer cohort that most resembles your business? Ask for the raw number, not the rollup. If it is not offered, the answer is in the silence.

Run those three before the contract. If the vendor survives, you have a real decision in front of you. If the vendor does not, you have saved yourself the cost, the domain repair, and the quarter.

— Stacey Tallitsch, Stronghold CMO