AI SDR vendor RFP template: questions, scoring and a 30 day test

A useful AI SDR RFP has six sections: scope, channels and limits, data, control and safety, price and terms, and proof. Score each answer from 0 to 3, weight the sections by what matters to you, and then run a 30 day trial with pass marks you set before it starts. The trial decides more than the RFP does, because claims are cheap and held meetings are not.

Copy this into your own document and send it to each vendor. Our buyer's checklist explains why each question matters, and the security checklist goes deeper on that section. This version is meant to be sent.

The cover note

We are evaluating AI SDR vendors for [use case, for example outbound to mid market logistics firms in Germany]. We plan to contact about [number] people a month on [channels]. Please answer the questions below in writing, attach evidence where asked, and tell us if a question does not apply. We will run a 30 day trial with two finalists. Responses due [date].

Section 1: scope

1. What does your product do, in two sentences, and what does it not do?
2. Which job does it replace: finding leads, writing, replying, booking, reporting?
3. Who in our team must be involved weekly, and for how long?
4. What is a typical customer like us, and what results did they get? Please attach one case study with the base and the definitions.
5. What do you not recommend it for?

Section 2: channels and limits

1. Which channels does the agent run, and which are inbound or reply only? Please mark each.
2. How does it connect to our accounts? Does it store passwords?
3. What daily and weekly limits apply per account, and who sets them?
4. Is there a quiet window or sending schedule?
5. How does it handle replies on one channel when we contact the same person on another?
6. What happens at the end of a sequence?
7. What does it do about machine mail such as auto replies and bounces?

Section 3: data

1. Where do contacts come from? Do you provide data, or do we import ours?
2. If you provide data, how large is the database, how is it verified and how fresh is it?
3. What does a lookup cost, and what happens to unused credits?
4. Can we export our leads and conversations at any time, in which format?
5. How do you handle a request to erase a person?
6. Do you use our data to train models?

Section 4: control and safety

1. Can every message wait for our approval? Can we switch it off per campaign?
2. How do we take over a conversation, and hand it back?
3. When does the agent stop on its own?
4. How do you prevent the agent from stating prices, dates or claims it was not given?
5. What happens when someone says stop? Is the suppression list checked before every send?
6. Which languages does it write in, and how do you check quality?
7. What do you document about security: sub-processors, data location, DPA, breach notice? Do you claim any certification? If so, attach the report and its date.
8. What do you say about the risk of platform restrictions, and what controls reduce it?

Section 5: price and terms

1. What is the full price for our scope, per month, with every fee listed?
2. What is the pricing unit, and what makes the bill grow?
3. What is not included: setup, seats, credits, overage, onboarding?
4. What is the contract length, and how do we cancel?
5. Is there a free trial, how long, and does it need a card?
6. How do you handle price changes?

Section 6: proof

1. How do you define a meeting? Booked, held, qualified?
2. How do you report results? Please share a sample report.
3. What is your measured reply rate on comparable campaigns, with the base and the definition?
4. Can we speak to two customers with a similar use case?
5. What went wrong for a customer last quarter and what did you change?
6. What is your support response time?
7. What is on your roadmap for the next six months that we should rely on, and what should we not?
8. What would make you advise us not to buy?

The scoring sheet

Score each answer 0 to 3: 0 no answer or evasive, 1 vague, 2 clear, 3 clear with evidence. Weight the sections to your risk. An example for a small team:

Set the weights before you read the answers. Mark questions 7, 19, 23 and 33 as must pass: an evasive answer to any is a stop.

What do good answers look like?

Scoring is easier when you know what a 3 sounds like. Here is how Sellinger answers four of the must pass style questions, so you can hold every vendor, including us, to the same standard.

• Question 7, connecting accounts. You connect your own LinkedIn account through a hosted login and Sellinger does not store your LinkedIn password.
• Question 8, limits. Each LinkedIn account has a daily invitation limit (default 20, varied by a few each day) and a weekly cap (default 150). Email defaults to 50 sends a day per mailbox, with an optional warm up that starts at 5 a day and doubles every 3 days up to the limit.
• Question 19, approval. A campaign can run in approval mode, where every message waits for a person, or send automatically. A person can take over any conversation.
• Question 23, stop requests. Before every send on email, LinkedIn and WhatsApp the platform checks the team's suppression list and refuses the send when the contact is on it or when the check cannot be completed.

Notice what makes these a 3: each is specific, each names a default you can verify, and none claims that nothing can go wrong. We also say plainly that no tool can guarantee LinkedIn will never restrict an account. An answer that promises zero risk should score lower, not higher.

What are the warning signs in a response?

• The answer repeats the question back as a slogan.
• A number has no base, for example a reply rate with no count of people messaged.
• Meetings are reported as "booked" with no definition of held.
• The price is "from" a figure and the real scope needs a call to price.
• The vendor will not offer a trial on your own list.
• A certification is claimed without a report and a date.

Score these as a 0 or 1 and ask one follow up question. If the second answer is still evasive, move on. Vendors who are good at this expect to be asked.

The 30 day trial plan

An RFP shows what a vendor says. A trial shows what the product does. Run the same test for each finalist.

1. Same list. 200 people who fit your buyer, split into equal groups if you test two vendors.
2. Same offer. One offer and one channel pair.
3. Approval on. Read every message for the first week.
4. Set pass marks first. For example: acceptance at or above [your norm], positive replies at or above [number], at least [number] meetings held, no invented claims in a sample of 30 messages, flagged conversations answered by a person within [hours].
5. Count held meetings. On our own account 351 meetings were booked and 33 confirmed held, so a vendor that reports booked meetings alone tells you little.
6. Review the hours. How long did you spend per week?
7. Decide at day 30. Pick the one that met the pass marks with the least of your time.

Sellinger offers a 30 day free trial on messaging modules and bundles with no card, so a trial costs nothing but your hours. Prices and terms for other vendors change often. Our AI SDR pricing comparison dates and sources each figure, and best AI SDR tools lists the options.