# Can AI make sales calls to upsell customers? > Justas Butkus is a fractional AI officer based in Vilnius, Lithuania, founder of AINORA, MB – the company behind the Ainora and Impetora brands – and a graduate of Vilnius University. Justas Butkus is a fractional AI officer based in Vilnius, Lithuania, working with mid-market companies and scale-ups across the UK, EU and US at two to four days a month. He builds and operates the production AI systems he advises on, and takes no commission, referral fee or revenue share from any vendor. Video: Can AI Make Sales Calls to Upsell Customers? What It Takes (https://youtu.be/Uv3xiJJmcIg) **Yes, AI can make sales calls to upsell customers: it can call the people in your CRM, check in with them and pass the interested ones to your sales team. It is not plug and play. It works as well as the script, the test and the review behind it, so treat it like training a new hire from zero. Start with 200 or 300 calls to old leads, and go in knowing that scaling it is a proper project.** Canonical: https://justasbutkus.com/answers/can-ai-make-sales-calls-to-upsell-customers/ Last updated: 2026-10-22 --- ## Can AI really call your CRM and upsell your customers? Yes. The answer is obviously yes, with conditions. The idea is not to replace your sales reps but to give them longer hands: the AI makes the calls your team does not have time for, and the people who show interest go to a human. The rest of this page is the conditions, and most of them I learned the hard way. ## What happened with my first customer? My first customer ever sold software to construction companies. They wanted to offer their existing customers higher-level packages, extra training or other materials. The decision on their side was that AI would call, ask whether everything was fine with the software, and offer a human rep for anything more. It was closer to a market survey than a hard sell, and I still think it was the perfect case. The technology was not there yet, and they canceled. Their customers picked up on a building site, on speaker, with humming noises behind them and someone talking over them, maybe in another language. The AI struggled to work out who was speaking. We have since overcome those problems: filtering the sound and answering questions reliably. The lesson is the first thing to test: in what environments do your customers pick up, and does the AI stay on track when there are several voices? ## What does a call script need to cover? A proper script is a decision tree: every way a conversation can go, every objection, and where each path leads. Take "I cannot speak right now". In the first calls with that customer, the AI marked those people as not interested. Better: it asks when it can call back. If the person says "in 15 minutes" or "I am driving", it schedules the call-back for then. If they hang up at once, it tries again in a couple of days. It also follows every promise to the letter. One person said to call tomorrow, and tomorrow was an international holiday. The AI called, and the person was annoyed, until we pointed to the recording of them asking for it. That is the point: it follows the rules exactly, so the rules have to be right. ## Why is a box off the shelf not enough? A one-shot solution you buy out of a box can work in simple cases, with a couple of qualifying questions. A more sophisticated buyer needs more. Someone who wants one product needs different questions and different criteria from someone who wants another. What we build is a team of callers. One general caller handles the call, then transfers to a specialist trained on that product: its conversation flow and its knowledge base. The transfer is immediate, unlike a human transfer where you wait. For very high-end products or quantities, you may not want the AI selling at all: it notes down a message for the manager and tells the customer the manager will call back. If you understand your company and its products really well, the AI can do the work of your best people. It does not have their emotional intelligence yet, but it follows the instructions you script. Think of hiring a new person and training them from zero. Then review. Check which logic mistakes it made and why, whether an instruction was unclear, whether a product needs its own specialist. The first version is only a version: you adjust it as you see how people actually respond. ## What should the first test look like? A test drive of about 200 or 300 calls, on one segment but mixed leads. Give it the leads that are a bit older, the ones sitting in the CRM because nobody has the manpower to reach them. Test it on your worst leads. If it gets them interested, or even makes sales where you only need to send the invoice, you can expand. It also protects your team. Sales reps get tired of bad leads, so they work with better energy when the leads reaching them are qualified. You are protecting your scarcest resource: human time, understanding and focus. Do not expect a dumb robot that gives a discount when someone says it is too expensive. That is the logical move for an untrained human or AI, and it may be wrong, because maybe the value was not explained. If a discount is right, the AI can say it will call back after speaking to its manager, and really ask the manager, or ask what to do in a strange case. ## Can it make 10,000 calls? Yes, from 10 calls to a million, but each step up changes the requirements. Compliance comes first. In Europe, from what I know, you need consent from people to call them. Talk to your legal team; I am not a lawyer. **What changes as the volume grows** | What | Why it matters | | --- | --- | | Consent | You need people's consent to call them; settle it with your legal team | | Phone numbers | Phone providers flag spam, so volume needs more numbers and human-like intervals | | Server capacity | It sets how many calls can run at once | | Provider limits | Check your AI providers for concurrency limits | Building this in-house means understanding all of that. We generally recommend working with professionals to build it. It can be a really big project, just so you know. ## Who should not do this? If you do not have enough leads, if few people come to you and you already follow all of them up, it is not the case to invest. This is for more traditional businesses with volume, where the calls outnumber the people who can make them. ## Frequently asked questions ### Does AI replace my sales reps? No. It gives them longer hands: the AI makes the calls there is no time for and passes the interested people to a human. ### Can I just buy a ready-made solution? For simple cases with a couple of qualifying questions, it might work. Once buyers differ in what they want, you need different questions per product, and that takes a script built for your business. ### How big should the first test be? About 200 or 300 calls, on older leads your team never had time to reach. ### Will it give a discount when a customer says it is too expensive? Not automatically. It can say it will call back after speaking to its manager, and really ask the manager. ### Do I need consent to call my customers? In Europe, from what I know, yes. Settle it with your legal team; I am not a lawyer. ## Related - [The 3 levels of AI adoption](/answers/3-levels-of-ai-adoption/) – Where calling leads and following up sits among the levels. - [Implementing without getting burned](/answers/how-to-implement-ai-without-getting-burned/) – What to check before you commit budget or headcount. - [Is this AI vendor legitimate?](/answers/is-this-ai-vendor-legitimate/) – Questions to ask before you hand a vendor your customer calls. ## About the author **Justas Butkus** – a fractional AI officer based in Vilnius, Lithuania, founder of AINORA, MB – the company behind the Ainora and Impetora brands – and a graduate of Vilnius University. ## Book a call to see if it fits your CRM Send a short email describing who you would call and why, and we can book a call to see whether a test of 200 or 300 calls makes sense. Contact: justas@ainora.lt · [LinkedIn](https://www.linkedin.com/in/justas-butkus/)