AI Sales Agents for B2B: What They Do, What They Cost, and Buy or Build

    What an AI sales agent does for a B2B company and where it stops, buying a tool versus building on your own stack, the AI Act and GDPR, running costs, and our price.
    Sander VergouwenUpdated: 16 November 2026

    In short

    An AI sales agent is software that handles one part of the sales process on its own, such as qualifying inbound leads or following up on no-shows. It combines a language model for the conversation with a workflow layer that writes to your CRM, and hands over to a person when it is unsure. A Sales Surge AI Sprint builds one agent in 10 working days, from €4,997.

    Source: Sales Surge, AI Sprint pricing and methodAs of: 16 November 2026

    In this article
    1. What is a B2B AI sales agent?
    2. AI sales agents, AI sales agent software, or an AI sales agent platform?
    3. What can an AI sales agent actually do for a B2B company?
    4. What an AI agent cannot do (yet)
    5. Buy an AI sales agent or have one built?
    6. Is an AI sales agent allowed? AI Act, GDPR and call recording
    7. How Sales Surge builds AI agents
    8. How long does it take to build and launch?
    9. What does an AI sales agent cost?
    10. AI sales agent running costs: what you pay per month
    11. What we measure on an agent
    12. Where to start with AI in sales
    13. Frequently asked questions
    An AI sales agent picks up a new lead without an owner, enriches it, scores it on fit and routes it to the right rep within seconds.
    An AI sales agent picks up a new lead without an owner, enriches it, scores it on fit and routes it to the right rep within seconds.

    What is a B2B AI sales agent?

    A B2B AI sales agent is software that does one specific job in the sales process on its own. It qualifies inbound leads, follows up on no-shows, or chases deals that went quiet. The agents that work in production do one job. The ones that fail are sold as a complete AI salesperson.

    Underneath there are two layers. The language model reads the prospect's message, decides what to say and writes a reply. The workflow layer connects that model to your CRM, calendar, telephony and data sources. Without the workflow layer the model is a clever chat that goes nowhere. Without the model the workflow is the rules-based automation companies have run for years.

    Berend de Visser, our virtual assistant, is an AI agent and not a person. We recommend that every client is just as open about their own agent. A buyer who finds out later that they were talking to software trusts you less, and since 2 August 2026 the AI Act requires you to tell them.

    AI sales agents, AI sales agent software, or an AI sales agent platform?

    AI sales agent software is a product you license, with a builder and connectors. An AI sales agent platform goes one step wider: a place to run many agents, with monitoring around them. Both are containers. An agent is the thing inside that does the work: one job, permission to act in your CRM, and a defined point where it hands over to a person.

    That is why the market for AI agents for sales looks confusing. Most of what is advertised sits at the software or platform layer, so a demo shows a builder rather than a working agent. For a first agent, a platform is usually more than you need.

    What can an AI sales agent actually do for a B2B company?

    An AI sales agent handles work that is high in volume, repeats and has clear rules. These are the jobs we scope agents for.

    • Qualify inbound leads by chat or phone, around the clock. The agent asks the questions your team would ask. Anything inside the ICP goes to a booking flow.
    • Follow up on no-shows and stalled deals. The agent watches Pipedrive for missed meetings and deals without activity, sends a message that knows the context, and books a new slot or moves the deal to nurture.
    • Book meetings in the rep's real calendar. Once a lead qualifies, the agent offers real slots and logs the meeting in the CRM with the conversation attached.
    • Update the CRM after every conversation. Stage, next step and qualification go back into structured fields, not into a note nobody reads.
    • Hand hot leads to a person in seconds. When a lead names a budget, a deadline or a competitor, the agent alerts the rep in Slack or by phone.

    What an AI agent cannot do (yet)

    One version of the pitch says you can replace your whole SDR and AE team. That version is wrong. This is what agents do not do well, and why a person still owns the high-stakes parts of a deal.

    • Negotiate with several stakeholders. Once procurement, legal and a buying committee join, an agent cannot read the room.
    • Build the relationship on a large deal. Large B2B deals close on trust, and trust takes months of judgement calls in tense moments.
    • Calm an upset customer. The answer has to come from someone who can fix the problem. An agent makes that moment worse by sounding rehearsed.
    • Commit to anything the company must answer for. Prices, dates and deliverables stay outside what the agent may promise. That is why it works inside tight guardrails.

    Agents win on volume and speed. They take the top of the funnel and the long tail of follow-up, so senior reps keep their time for the conversations that need judgement.

    Buy an AI sales agent or have one built?

    Buy a ready-made AI SDR when your process fits the tool and you sell mainly in English. Have one built on your own stack when the agent has to write to your CRM in your fields, speak Dutch, or follow a process that is specific to you. Several vendors sell an AI SDR as a product. This is how they compare with a sprint on your own stack, as of November 2026. Their products change quickly, so ask each vendor for a demo on your own data.

    OptionWhat it isChannelInbound or outboundWhere the agent runs
    11xAI SDR products for outbound prospectingEmail and phoneOutboundIn the vendor's platform, with a sync to your CRM
    ArtisanAn AI BDR for outbound prospectingEmailOutboundIn the vendor's platform, with a sync to your CRM
    QualifiedAI SDR for conversations with website visitorsWebsite chatInboundIn the vendor's platform, strongest alongside Salesforce
    Regie.aiAI agents and content for outbound sequencesEmail sequencesOutboundNext to your CRM and sales engagement tool
    AI Sprint on your own stackOne agent built on your CRM, telephony and Make.comChat, phone or email, one per agentEither, one job per agentIn your own tools, writing to your own Pipedrive fields

    The question behind 'buy an AI sales agent for pipeline growth' is usually speed. A product can be live quickly if your process matches its defaults. A build takes 10 working days but follows your process, your fields and your tone.

    In any demo, ask to see a real CRM record before and after the agent acted, and a case where the agent decided not to act. A demo that only shows a builder is not an agent yet. If you sell in Dutch, ask for that demo in Dutch, and ask for a quote on your own volume rather than a list price. The AI Sprint is from €4,997 fixed, plus your own licences.

    Is an AI sales agent allowed? AI Act, GDPR and call recording

    Yes, as long as people know they are dealing with AI, you have a legal basis for the data, and calls are recorded only with notice. This is general information and not legal advice. For your own situation, have a lawyer check the set-up.

    • Tell people they are talking to AI. Article 50 of the EU AI Act requires that people are informed when they interact with an AI system, unless that is obvious. These transparency rules apply since 2 August 2026. The EU has proposed shifting some AI Act deadlines, so check the current status. We build the notice into the first message or the start of the call.
    • Have a legal basis for the lead data. Under the GDPR you need a basis to process a lead's data. In B2B that is often legitimate interest, which needs a written assessment. Enrichment from external sources counts as processing too.
    • Record calls only with notice. Tell the caller at the start that the call is recorded and why, and offer a way to continue without recording. Keep recordings no longer than you need them.
    • Sign a data processing agreement with every vendor. The voice platform, the telephony provider, the language model and the automation layer each process personal data. Each needs a data processing agreement.
    • Prefer EU hosting. Where a tool offers EU data storage, use it. Where data leaves the EU, check that the vendor has valid transfer safeguards in place.

    Outbound calling has its own rules too. In the Netherlands, calling consumers requires prior consent. B2B calls are allowed under conditions. An agent that calls out should follow the same rules as a person.

    How Sales Surge builds AI agents

    We build agents in a fixed programme of 10 working days called the AI Sprint. Fixed price from €4,997, fixed scope, one agent live in production at the end of the second week, with 30 days of support afterwards. We keep it short by limiting the scope to one agent and bringing a stack that is already connected.

    LayerToolRole
    VoiceElevenLabsThe voice and the conversation layer of a voice agent, with low enough latency for a normal back-and-forth
    TelephonyCloudTalkNumbers, recording and a direct Pipedrive integration, so calls log against the deal
    WorkflowMake.comMoves data between model, CRM, calendar and Slack, and handles the escalation paths
    CRMPipedriveThe agent reads deal context here and writes the outcome back as field changes

    More on the telephony side on the CloudTalk page.

    How long does it take to build and launch?

    One agent takes 10 working days from kick-off to live in production.

    Days 1 to 3

    Design

    The single job, the trigger, the tone of voice, the AI notice, the success metric and the escalation rules. You sign off a one-page agent spec.

    Days 4 to 6

    Build

    The prompt, the Make.com workflow, the CRM connection and the dashboard. The agent runs end to end in a test environment.

    Days 7 to 9

    Test and train

    Real scenarios from your pipeline go through the agent, edge cases go back into the prompt, and the team practises the hand-over.

    Day 10

    Go-live

    The agent goes into production with documentation, followed by 30 days of support.

    Agents with more than one job take longer and are better split into several agents, each with its own sprint.

    What does an AI sales agent cost?

    The sprint has a fixed price because its scope is fixed: one job, one channel and the standard stack. Voice, several languages, integrations outside the standard stack and high volume push the price up. A single-language chat agent with one integration keeps it down.

    AI Sprint

    from €4,997

    One agent, scoped, built, tested and live in 10 working days, with 30 days of support. Fixed scope and fixed price.

    Custom agent

    quoted after scoping

    Several jobs, integrations beyond the standard stack or multi-step logic. Always above the sprint, with a fixed price in the quote.

    Retainer

    Priced in the quote

    Prompt tuning, new escalation rules, edge cases and a quarterly review after the support period ends.

    Prices exclude VAT. Licences for your tools and the running costs below come on top. Is a quote much cheaper, ask to see the agent change a field in your CRM. If it can only answer questions, you are buying a chatbot. More on the AI sales agents service page.

    AI sales agent running costs: what you pay per month

    AI sales agent running costs are the sum of four meters, and each one grows with volume: voice minutes, language model usage, telephony and Make.com operations. The build price above is fixed. The running cost depends on how many conversations the agent has and how long they are. Vendor prices change often, so below is the volume calculation without prices. Fill in your own tariffs.

    A worked example. You receive 300 inbound leads a month. The agent handles 60% of them, so 180 conversations. A voice conversation takes 4 minutes on average, with about 10 turns in which the model reads and replies. Each conversation takes 8 Make.com operations: read the deal, look up the person, write three fields, check the calendar, book the slot and post to Slack.

    MeterFormulaIn the exampleYour tariff
    VoiceConversations × average minutes180 × 4 = 720 minutesPrice per voice minute × 720. Chat-only agents skip this meter
    Language modelConversations × turns180 × 10 = 1,800 model callsPrice per call, which depends on how much text and CRM context the agent reads per turn
    TelephonyNumbers + call minutes1 number + 720 minutesNumber subscription + price per minute × 720
    Make.comConversations × operations per conversation180 × 8 = 1,440 operationsFits in the plan you already have, or the step to the next plan

    Set the total next to the hours the agent takes over. In the example that is 180 conversations of 4 minutes, so 12 hours of calls, plus the logging and follow-up a rep would do after each one. If the four meters together cost more than those hours, the job is not a good fit for an agent yet, or chat is the better channel. Voice is usually the meter that grows fastest, so run this sum before you choose voice over chat.

    What we measure on an agent

    We judge an agent on what happens in the CRM, not on how natural it sounds. These are the numbers we set up for every agent we build, from the first week. Agree the threshold for each one before go-live, so a bad week leads to a change instead of a discussion.

    • Conversations per week, per channel. Rule: compare with the number of leads that came in. If the agent talks to clearly fewer leads than arrive, the trigger misses some of them.
    • Share that ends as qualified, and share that ends with a hand-over. Rule: set both against what your team reached by hand on the same lead source. A qualified share far above that usually means the questions are too soft.
    • Time from first contact to hand-over or booked meeting. Rule: measure the median, not the average. One lead that waits a weekend pulls an average out of shape.
    • CRM fields filled in correctly after a conversation. Rule: check a fixed sample every week, for example 20 records. Every wrong field goes back into the prompt or the field mapping.
    • Conversations where the agent should have stopped and did not. Rule: the target is zero. Every case gets reviewed and leads to a new escalation rule.

    The last one matters most. An agent that escalates too often costs time: if a person still has to take over in most conversations after the first month, the job is scoped too wide. An agent that does not escalate when it should costs deals.

    Where to start with AI in sales

    Start with one job that is high in volume, has clear rules and already lives in your CRM. Three first agents fit most B2B teams. Inbound qualification, because response time on inbound leads is where an agent helps first. Follow-up on no-shows and stalled deals, because nobody likes doing it and it is pure rules. And the CRM update after every call, because it saves every rep time without the buyer ever seeing the agent.

    If you cannot write the job down as an instruction for a new colleague, it is not ready for an agent. Fix the process first. The sales automation guide covers the workflows that should run before an agent goes on top.

    Frequently asked questions

    What does it cost to build an AI sales agent for a B2B company?

    An AI Sprint builds one focused agent on our standard stack in 10 working days, from €4,997 at a fixed price, with 30 days of support. Agents with several jobs or integrations beyond the standard stack are quoted after scoping and always cost more than the sprint. A retainer for tuning and new rules is priced in the quote. Licences and running costs come on top.

    Should I buy an AI sales agent or have one built?

    Buy a ready-made AI SDR when your process matches the tool's defaults and you sell mainly in English. Have one built when the agent must write to your own CRM fields, speak Dutch or follow a process that is specific to you. In both cases, ask for a demo on your own data that shows a real CRM record before and after the agent acted.

    Is an AI sales agent allowed under the AI Act and the GDPR?

    Yes, with conditions. Article 50 of the EU AI Act requires that people are told they are interacting with AI, with these rules applying since 2 August 2026. Under the GDPR you need a legal basis for the lead data, a data processing agreement with each vendor and notice before you record a call. This is general information, so have a lawyer check your own set-up.

    What does an AI sales agent cost per month to run?

    Running costs come from four meters: voice minutes, language model usage, telephony and Make.com operations. Each grows with the number and length of conversations. A chat-only agent skips the voice meter, which is usually the largest at high volume. Because vendor prices change often, we calculate the running cost per project with your expected volume before the build starts.

    Can an AI agent replace my SDR team?

    No, and any vendor saying yes is overselling. Agents replace specific tasks, not the SDR role. They take inbound qualification around the clock, follow-up on no-shows and the CRM update after each conversation. The SDR keeps the conversations that need judgement and the relationships with key accounts. How much time that frees up depends on how much of the day is repeat work today.

    What is the difference between an AI chatbot and an AI sales agent?

    A chatbot answers questions in a window, and when the visitor closes the tab nothing else happens. An AI sales agent is that conversation plus a layer that acts. It writes back to the CRM, books a real meeting in a real calendar, alerts a rep when a lead is hot and starts a follow-up when the conversation stalls. You can measure what an agent did afterwards.

    Do I need one AI agent or several?

    Several, and smaller than you think. The pattern that works is one agent per job: one that qualifies inbound, one that follows up on no-shows, one that keeps the CRM clean. Each has a narrow instruction you can test. Projects fail when one agent has to do sales and marketing at once, because when it misbehaves nobody can tell which of its jobs went wrong.

    How do I know if my sales process is ready for an AI agent?

    Ask whether you could hand the task to a new colleague with a written instruction. If yes, an agent can probably do it too. If you cannot write it down because it depends on judgement nobody has put into words, an agent makes that inconsistency faster instead of smaller. Fix the process first, then add the agent on top of a CRM that is in order.

    Sander Vergouwen

    About the author

    Sander Vergouwen

    Co-founder of Sales Surge

    LinkedIn

    Get an AI sales agent live in 10 working days

    Start with the growth scan, or book a 30-minute call. We scope the agent and confirm the fixed price. If you need to sort out your CRM first, we will say so.