Guides

How to choose an SDR agent for an enterprise sales team

How an enterprise team can choose an AI SDR: security review, SOC 2 and DPA requests, CRM governance, approvals, brand risk and how to run a pilot.

Short answer: Run the security review first, then check how the tool writes to your CRM, who approves each message and how you stop it. Test it in a time-boxed pilot before you sign, because some vendors sell annual or multi-year terms that can outlast a pilot [[1]](#sources).

Who this is for

This guide is for people who sign off on software at a larger company: sales leaders, revenue operations, security reviewers, legal and procurement. You are not picking a tool for yourself. You are picking one that a security team will question, a CRM admin will have to live with, and a brand owner will worry about.

If you are one founder choosing a first tool, read How to choose an SDR agent for an early-stage startup instead. If you run outbound for clients, read How to choose an SDR tool for a lead generation agency. If you are still deciding between software, a sales engagement platform and an outside agency, start with SDR vs sales engagement platform vs outbound agency. For a definition of the term, see AI SDR.

We make Outlevo, so read our examples as one option among several. We do not offer SOC 2 or SSO today. If either is a hard requirement for your security team, we are not a fit for you, and the rest of this guide still applies to the tools you do shortlist.

One more note on method. We do not list other vendors' SOC 2 status, DPA terms or data residency options here. Those change, and a blog post is the wrong place to read them. Ask each vendor for the current document and its date.

Steps

1. Write your gates before you see a demo

List what a tool must have to stay in the process. Typical gates are:

  • A current SOC 2 report, or the equivalent your policy accepts.
  • A signed data processing agreement.
  • Single sign-on for your identity provider.
  • A statement of where data is stored.
  • A way to limit what it writes to your CRM, such as writing only to fields the tool owns [[2]](#sources).
  • A person-approved step before anything goes out [[3]](#sources).

Mark each as a hard gate or a preference. A demo is much easier to judge when half the list is already settled. A tool that fails a hard gate leaves the list on day one.

2. Send the security questions in writing

Ask all shortlisted vendors the same questions and keep the written answers. Do not accept "yes, we are secure" from a call. Ask for the report itself, the period it covers, what is in scope and what is excluded. Ask for the DPA, the list of subprocessors and the storage locations. Ask whether SSO is included in the plan you would buy or sold as a separate tier.

Vendors differ here, and the answers are not interchangeable. A data platform, an AI SDR agent, an all-in-one sales platform and an open-source project will each hold different documents. Compare what each one gives you in writing, not what its category usually has.

3. Map what the tool reads and writes in your CRM

Ask for a field-by-field list. Which objects does it read? Which fields does it write? Does it touch your standard fields, or only fields it owns? Who approves the mapping before the first write? Can contact import be switched off?

4. Check approvals and stop controls

Find out what stops a bad message before it leaves. Look for three things: a policy a person approved, a record of each action, and a switch that pauses sending at the level you need (one campaign, one channel, everything).

5. Check the sending setup

Ask who owns the mailboxes and domains, how new mailboxes are warmed up, and what is checked before a campaign sends. Ask how unsubscribes are handled. Your deliverability and your brand sit on this layer.

6. Read the contract terms before the pilot starts

Check the term length, the renewal terms and what happens to your data when you leave. Do this before the pilot starts, not after it succeeds.

7. Design the pilot before you buy

Decide what counts as success before the first send. The section on pilot design below covers this.

What to check

Security review: what each vendor can show

Ask what scope a SOC 2 report covers, and ask for a plain description of the controls as well as the report.

As an example of the second kind, here is what we publish on our security page. Every customer table is protected by row-level security in the database. Personal data such as email addresses is encrypted at the field level. Staff support access is time-limited and shown in the customer's activity feed, with the reason and the number of changes made [[4]](#sources). That describes controls. It is not a certification, and we do not claim it replaces one.

Questions to put to every vendor, including us:

  • Which parts of your product does the SOC 2 report cover?
  • Is the DPA standard, or negotiable for our size?
  • Where is our data stored, and can we choose?
  • Which subprocessors touch our contact data?
  • Can your staff see our data? How is that access limited and logged?
  • What do you do with our data when we leave?

Write down any answer given only by phone, and ask for it in writing.

Single sign-on and access control

If SSO is a hard gate, check it in the plan you would actually buy. For tools that run on API keys, also ask how keys are scoped and who can create them. In our API, an owner, admin or developer creates keys, and the keys carry scopes [[5]](#sources). Whatever the tool, the question is the same: who can create credentials, and can you see what each one did?

CRM governance

A tool that writes into your CRM can damage the data your forecast depends on. The safest pattern is a tool that writes only to fields it owns, and only after your admin approves the mapping. Two examples from our own integrations show the pattern.

For HubSpot, Outlevo reads companies, deals and owners and writes its status, fit score and next step to fields it owns. Standard HubSpot fields are never edited, and nothing is written until an owner or admin approves the field mapping [[2]](#sources). Contact import is optional and happens only after an admin turns it on with a stated purpose [[2]](#sources).

For Salesforce, Outlevo reads accounts, opportunities and users, and writes only its own fields, Readiness Card records and tasks, through the Outlevo package [[6]](#sources). Two details matter to a security reviewer. The Salesforce connection is a pilot, and it needs the Outlevo package in your org, which Outlevo installs with you. There is no AppExchange listing [[6]](#sources). A rep's Accept or Reject on the Account is read by a 15-minute sync, not pushed to Outlevo by a webhook [[6]](#sources). Details are on the Salesforce integration page.

Use the same questions on any vendor. Does it need a managed package? Is it listed in a marketplace your security team already reviews? How quickly do rep actions reach the tool, and does that delay matter to your process?

Approval workflows and kill switches

The enterprise question is not "can the AI write a good email". It is "what can reach a customer without a person having approved it".

At Outlevo, nothing is sent outside a policy that a person approved. Policies set who can be contacted, on which channels, how often and with what budget. Every external action is checked against the approved policy before it runs, and recorded in a ledger. Kill switches pause a campaign, a channel or the whole workspace at once [[3]](#sources). See how it works.

Other tools handle this differently. One open-source project, AgentSDR, describes an AI CRM that classifies replies and drafts answers, with nothing sent without approval [[7]](#sources). In your evaluation, ask each vendor to show you the approval screen and the stop switch, not describe them.

Brand risk: what the tool is allowed to claim

A common brand risk is a tool stating something false about a prospect. Ask how each fact in a message is sourced, and what happens when the tool is unsure.

In Outlevo, each qualified account gets a Readiness Card with fit, intent and data quality scores. A card needs at least three observations, each with a source and a date. Facts read from a website are kept only when their quote is found on the page, and guesses are labelled as hypotheses, never presented as findings [[8]](#sources). Our platform page describes this in full.

The same test applies to everyone else: ask for a sample record and check whether you can trace every claim to a page and a date.

Sending setup and the rules of the inbox

Google's sender guidelines, checked 2026-10-22, say: "Marketing messages and subscribed messages must support one-click unsubscribe, and include a clearly visible unsubscribe link in the message body." [[9]](#sources) The standard behind one-click unsubscribe, RFC 8058 (checked 2026-10-22), "describes a method for signaling a one-click function for the List-Unsubscribe email header field." [[10]](#sources) Ask each vendor how it meets both.

Also ask what must be true before a campaign sends. In Outlevo's Smartlead integration, every mailbox must be connected and warmed up, and its domain must have SPF, DKIM and DMARC checked in the last 7 days [[11]](#sources). With managed mailboxes, a new mailbox sends nothing until it is warm: at least 14 days, usually about 3 weeks [[12]](#sources).

Check mailbox support too. AgentSDR sends from Google Workspace mailboxes only, and Microsoft 365 is not supported yet (checked 2026-11-09) [[13]](#sources). If your company runs on Microsoft 365, a limit like that ends the evaluation early.

Your counsel decides this, not a vendor. For context, GDPR Article 6(1)(f) allows processing that "is necessary for the purposes of the legitimate interests pursued by the controller or by a third party" (checked 2026-10-22) [[14]](#sources). Recital 47 adds: "The processing of personal data for direct marketing purposes may be regarded as carried out for a legitimate interest." (checked 2026-10-22) [[15]](#sources). Whether that covers your sending, in the markets you have chosen, is a question for your legal team. Ask each vendor how it records the basis and how it handles opt-outs and suppression.

What each kind of vendor publishes

Vendors in this space describe themselves in very different terms, and the description tells you what you are buying.

  • ZoomInfo calls itself "The AI platform for go‑to‑market teams" (read 2026-11-16) [[16]](#sources). It says its pricing is based on features and functionality, number of licenses and credit usage (read 2026-11-16) [[17]](#sources).
  • Apollo says it is "built to work for companies of all sizes" (read 2026-11-16) [[18]](#sources).
  • Clay's Enterprise plan has custom pricing, described as "For organizations building GTM systems at scale." (read 2026-11-09) [[19]](#sources)
  • Amplemarket describes an all-in-one sales platform with AI agents, B2B data, multichannel sequences and parallel dialing (checked 2026-11-06) [[20]](#sources).
  • Bitscale describes a data waterfall across 18+ contact-data providers, with each result verified before it reaches the CRM (checked 2026-11-09) [[21]](#sources).

None of those lines tells you a vendor's certifications, so none replaces the questions above. For price comparisons with dates, see our pages on Apollo, Clay and ZoomInfo, or the category pages for data enrichment tools and CRM tools.

Pilot design

A pilot is where an enterprise buyer finds out whether the answers in the security review hold up. Set it up like a test, not a trial of the interface.

Start in a sandbox. Our test API keys run against a sandbox with fake companies and contacts, and are not charged [[22]](#sources). The sandbox has 100 fake companies, so you can build without sending anything or spending credits [[5]](#sources). Ask every vendor whether it has an equivalent. If not, ask what a pilot means for your real contacts.

Test the failure paths. Every Outlevo API write takes an Idempotency-Key, so a retried request never runs twice [[5]](#sources). Webhook deliveries are signed with an HMAC-SHA256 header and retried for up to 12 hours. Each delivered event costs 1 credit [[23]](#sources). Whatever you evaluate, ask what happens when a request is repeated and when your endpoint is down.

Plan around warm-up. If the tool sends from new mailboxes, a two-week pilot may end before anything is sent. Managed mailboxes in our setup need at least 14 days before sending [[12]](#sources). Set the pilot length after you know the warm-up time.

Fix the success measure up front. Decide the metric, the comparison group and the owner of the number before the first send. Be careful with small samples. Our win/loss analysis counts companies, not deals, and treats fewer than 10 won companies as very low confidence [[24]](#sources). If your pilot segment is small, expect any reading to be weak, and say so before the results arrive.

Limit the blast radius. Pick one segment, one channel and one region you already sell into, and use the vendor's own caps. Keep the stop switch tested. Run it once on purpose during the pilot, so the first time you use it is not an incident.

Mind the contract. Some vendors sell annual plans by default. 11x, for one, offers two- and three-year commitments on its Pro and Enterprise plans (checked 2026-10-22) [[1]](#sources). A pilot priced as a short test is a different thing from a contract that outlasts it. Ask for a pilot term in writing, with the exit conditions.

Price and contract shape

Enterprise buying is rarely a price per month, but a reference point helps. Outlevo's plans are Free at USD 0 per month, Builder at USD 29, Growth at USD 149, Self-Serve at USD 199 and Managed from USD 999 per month on a contract [[25]](#sources). The pricing page lists the credit costs for each action. For comparison, one published guide from March 13, 2026 puts monthly retainers for an outsourced SDR or appointment-setting agency at $4,000 to $18,000 (checked 2026-11-11) [[26]](#sources).

Ask each vendor to price your actual pilot, including any per-seat charge, credit usage and the plan tier that includes SSO.

Next step

Take your list of gates from step 1 and send the same written questions to every vendor on your shortlist. Drop the ones that fail a hard gate. Run a pilot with the rest, in a sandbox first.

If you want a low-risk first look at what we would work from, our free ICP assessment reads up to 5 public pages of your website. It returns facts labelled as verified observations or AI inferences, and up to 5 experimental segment hypotheses [[27]](#sources). It will not settle your security review, but it shows how we label what we know and what we guess.

If you use Salesforce, read the Salesforce integration page before you involve your admin. If you are weighing us against a sales engagement tool, see the Reply.io alternative comparison.

Sources

  1. 11x pricing (FAQ)
  2. HubSpot integration
  3. How Outlevo works
  4. Security
  5. REST API and SDK integration
  6. Salesforce integration
  7. AgentSDR on GitHub
  8. Readiness Cards
  9. Google: Email sender guidelines
  10. RFC 8058: Signaling One-Click Functionality for List Email Headers
  11. Smartlead integration
  12. Managed mailboxes integration
  13. AgentSDR on GitHub
  14. GDPR Article 6(1)(f): lawfulness of processing
  15. GDPR Recital 47: legitimate interests of the controller
  16. ZoomInfo: positioning
  17. ZoomInfo: pricing
  18. Apollo.io: company size
  19. Clay: pricing
  20. amplemarket.com
  21. bitscale.ai
  22. Developer hub
  23. Webhooks integration
  24. ICP from evidence
  25. Outlevo pricing
  26. RevNew, SDR outsourcing cost (2026)
  27. Free ICP assessment

Related

See what Outlevo finds on your own website

Free, no sign-up, about a minute.