What is an SDR? (sales development representative, explained)

An SDR, or sales development representative, is a salesperson who owns the top of the funnel: building and checking the target list, running outreach across email, phone, LinkedIn and messaging, qualifying whoever engages, and booking meetings for the account executives who close. The output is an accepted meeting, not revenue. That single split — prospecting separated from closing — is the whole idea, and everything else about the role is a consequence of it: how the handoff is defined, how capacity is calculated, how the person is paid, and when hiring one is premature.

What the role actually is, and why the job got split

Before the split, one salesperson did everything: found the account, opened the conversation, ran discovery, negotiated, closed, and often handled the renewal. The specialisation argument is that these tasks reward incompatible temperaments and incompatible calendars. Prospecting is high-volume, high-rejection, interruptible work. Closing is low-volume, high-preparation, protected-calendar work. Someone doing both will do the second, because the second is where the commission is, and the pipeline dries up two quarters later.

So the job was cut in two. The SDR owns everything up to and including the booked meeting. The account executive owns everything after it. The boundary is the only genuinely contested part of the design, and it is where most implementations of the role go wrong.

Terminology, briefly, because the questions are constant. SDR and BDR are used interchangeably in most companies. Where a distinction is drawn, SDR usually means qualifying inbound demand and BDR usually means outbound prospecting into cold accounts, but that convention is not universal and is often reversed. The words do not tell you what a specific job is; the compensation plan and the handoff definition do. The one-screen definitions live at SDR and BDR if you want them side by side.

No statistical agency describes this job

Worth knowing before you read any salary article. The US Standard Occupational Classification, which underpins every official employment and wage series, has no code for a sales development representative. The nearest is 41-3091, Sales Representatives of Services, Except Advertising, Insurance, Financial Services, and Travel, which is a broad category covering people whose work looks nothing like an SDR's.

The nearest thing to an official adjacent number: the US Bureau of Labor Statistics' Occupational Outlook Handbook reports that wholesale and manufacturing sales representatives held 1,571,400 jobs in 2025, with a median wage of $72,080 in May 2025 for the non-technical group and $104,920 for the technical and scientific group, and projects little or no change in employment from 2025 to 2035. That is a different occupation and should not be read as an SDR salary. It is included here to make the point that when a role has no official classification, every number attached to it comes from a private survey whose sampling frame you cannot inspect.

The practical consequence is that you should build your own compensation model from your own economics rather than benchmarking against a figure with no provenance.

The handoff contract is the whole job

An SDR's output is a meeting. If "a meeting" is not defined precisely, the role produces arguments instead of pipeline. Write the definition down before the first hire, and write it as three separate facts.

What counts as booked. A slot exists on a named account executive's calendar, with a named person from a named company, at a specific time, accepted by both. Not "they said they were interested". Not "I sent a link".

What counts as accepted. The account executive can reject a meeting, and the criteria for rejection are fixed in advance and are about the account rather than about the outcome. Wrong company size, wrong function, no relevant problem: rejectable. Attended and went badly: not rejectable, because that is the closer's job. Without this rule the SDR games the number; with a vague version of it, the AE games the number in the other direction.

What counts as held. Booked and held are different quantities and both need tracking, because the gap between them measures the quality of the booking rather than the quantity. A team booking 40 and holding 22 has a qualification problem that a booking target will never surface.

Pay on held, count booked, and review accepted weekly. The three-number discipline is what stops the role becoming a calendar-filling exercise. The related question of what makes a lead sales-ready in the first place is a different boundary, worked through in the difference between an MQL and an SQL.

Capacity: what one person can physically do in a week

Most SDR targets are set by wishing. Set yours by arithmetic, and start from the ceilings rather than from the goal, because the ceilings are not negotiable.

Some of them are platform limits rather than human ones. Autocloz's autopilot ships with a per-workspace cap of 250 actions a day, a 09:00 to 17:00 send window, and per-channel daily caps that are deliberately conservative on the platforms that punish volume: 20 LinkedIn invitations a day and a trailing seven-day cap of 80, against LinkedIn's own roughly 100-a-week ceiling, leaving headroom for the same human's organic activity on the same account. The free plan's own per-channel daily limits are 1,000 emails, 250 SMS, 250 WhatsApp, 100 calls and 50 LinkedIn actions across the whole workspace.

Now the human ceilings. These vary enormously by market, so the numbers below are illustrative — replace every one of them with your own measurement.

Take a five-day week with four hours a day of genuine outbound time, which is realistic once you subtract meetings, research and reply handling. Suppose a dial with dispositioning and note-taking averages 3.5 minutes. Two hours of calling a day gives roughly 34 dials, or 170 dials a week. Suppose 6% of dials reach the person you wanted: 10 conversations. Suppose a third of those conversations produce a meeting: 3 meetings a week from the phone.

Email and LinkedIn run alongside rather than instead. At 40 new email prospects a day that is 200 first touches a week, and with a five-touch sequence the steady-state daily send load is around 200 messages — comfortably inside the plan cap and comfortably outside what a single mailbox should send, which is the constraint that actually binds. LinkedIn adds 20 invitations a day against the trailing-week cap of 80, so 80 invitations a week is the hard number regardless of ambition.

Two conclusions fall out, and they hold across most sets of numbers you might substitute. The binding constraint is almost never the plan tier. It is mailbox count on email, the weekly invitation ceiling on LinkedIn, and hours on the phone. And reply handling scales with sends, so an SDR who doubles output halves the time available to answer what comes back — which is the single fastest way to destroy the value of the extra volume. The cadence design that respects these ceilings is covered in sales cadence best practices, and the sequencing mechanics in what a sales sequence is.

Autocloz's free plan covers 5 users and 10 mailboxes, which is enough for a first SDR pod running all five channels behind one login — start free and measure your own conversion rates before you commit to a target.

Compensation, and the two things it always distorts

There are only a few structures and each one bends behaviour in a predictable direction. Pick the distortion you can live with.

Base plus per-meeting variable is the common design. It is simple and it distorts toward volume: the rep books whatever will accept a slot. The counterweight is the acceptance rule above, applied consistently, including in the last week of a quarter.

Base plus variable tied to pipeline created moves the incentive downstream and distorts toward large, unrealistic deals, because a rep paid on opportunity value will find opportunity value. It also introduces a lag that is demotivating for someone whose work is daily.

A clawback on meetings that do not hold looks like the obvious fix for no-shows and is usually a mistake at this level. The SDR does not control whether a prospect attends a meeting three weeks out, and paying people on outcomes they cannot influence produces cynicism faster than it produces attendance. Reduce no-shows with reminders and shorter booking lead times instead.

Whatever you choose, model the cost of the whole seat rather than the salary. Tooling, the manager's time, and the ramp period during which output is low are all real, and the commission calculator will do the variable half of the arithmetic if you want to compare two plan shapes before you commit to one.

One classification point that surprises people. In the United States, the outside sales exemption from overtime does not cover a desk-based SDR. The Department of Labor's Fact Sheet 17F sets a two-part test — primary duty must be making sales, and the employee must be customarily and regularly engaged away from the employer's place of business — and states plainly that outside sales does not include sales made by mail, telephone or the Internet unless that contact is merely an adjunct to personal calls. Whether some other exemption applies is a question for employment counsel, not for a sales blog, but the outside-sales one is off the table.

The permission model an SDR seat actually needs

A CRM seat for this role is not "give them access". Over-granting is how a departing rep leaves with the customer list, and under-granting is how a rep spends their week asking an admin for things.

Autocloz's role system carries six permissions per module — view, create, edit, delete, export, send — and each granted permission also carries a row scope of off, own, team or all. That grid is what lets an SDR seat be specific rather than approximate.

A defensible starting configuration:

• Leads: view, create and edit at team scope. No delete. Deletion is an admin action, because an accidental bulk delete by someone learning the tool is unrecoverable.
• Export: off. This is the single most consequential toggle in the grid. An SDR does not need to export the contact database to do the job, and export is how a list walks out of the building.
• Campaigns and templates: view and create, edit at own. A rep should be able to build their own sequence without being able to edit the sequence three colleagues are running.
• Send: on, for the channels they work. This is separate from edit for a reason: authoring a message and dispatching it to five hundred people are different privileges.
• Analytics: view at own or team, depending on whether you want reps comparing numbers. Both answers are defensible; pick deliberately.
• Suppressions and deliverability: view only. A rep should be able to see that an address is suppressed and must not be able to unsuppress it.

The general rule is that scope should match accountability. If a rep is accountable for their own accounts, own is right and all invites both accidents and quiet list-building. Shared workspace controls are what make this practical across a pod rather than a per-user negotiation, and the multi-client version of the same problem is worked through in the Autocloz and Salesloft comparison.

Ramp is a real cost, and it is longer than the plan says

A new SDR is not productive on day one and the shape of the curve matters more than its length. Three things have to land before output is real: the product knowledge to answer a basic objection, the territory knowledge to know which accounts are worth a touch, and enough sequence history that their own numbers mean something.

The last one is the constraint nobody plans for. A rep needs a few hundred sends before their reply rate has any statistical content, so the first several weeks produce numbers that cannot be coached against. Coaching a reply rate computed on 60 sends is coaching noise, and it teaches the rep to distrust measurement.

Three practices that shorten this honestly. Give a new rep an existing, working sequence rather than asking them to write one, so their early variance is about execution rather than copy. Have them handle replies to somebody else's sends in week one, because reading real objections is the fastest product education available. And set the first target on activity rather than outcomes, switching to outcomes only once the sample size supports it.

What an SDR cannot fix, and what the tooling will not do for you

Four limits worth being explicit about.

An SDR cannot find your market. A rep given a vague profile will produce a vague list and a quarter of well-documented rejection. Somebody who understands the product has to define who it is for before the role can be executed, and no amount of activity substitutes for that definition.

An SDR cannot repair a message that does not work. If the founder cannot get replies, a new hire with less context will get fewer. The honest sequence is that outbound is proven by someone who knows the product, then handed to someone who can do it at volume.

The platform will not decide the handoff for you. Autocloz records meetings, enrolments and stage transitions, but "accepted meeting" is a definition your team writes and enforces. No CRM field arbitrates whether a meeting was worth an account executive's hour.

Automation relocates the work rather than removing it. Every draft an assistant writes is a draft somebody reviews, and review scales with volume the same way composition did. Where that boundary should sit — which parts of the role are safe to automate and which are not — is a separate argument, made in full in AI SDR versus human SDR.