AI in Action: Sales Org Autonomy

00:00:27 - Bayram:
Okay, let me set this up.00:00:34 - Bayram:
So I can see the chat. Okay, just one more minute and we'll start.

00:01:02 - Bayram:
Just to get a sense of the audience. Let me know if you've been to the previous seminar, which we had, I think about a month ago, about AI and B2B sales. Just send a message to the chat if you are part of that webinar. Hold on.

00:01:30 - Bayram:
Okay. My laptop. Okay. Okay. I think we can start just in time. So, hi, everyone, and thanks for attending this webinar. This is the second one in the series of AI in action, and the last time I've briefly touched the idea of sales Org autonomy and my vision.

00:02:01 - Bayram:
How companies or sales organizations in those companies will evolve, starting from doing everything manually up to basically having autonomous revenue engine that just sells for you. And of course, it's going to take us time to get to that autonomous level. But I think it's great to do two things. First of all, to understand what's the path.

00:02:32 - Bayram:
Towards that autonomy. Second is to basically understand where we are right now and what's the next best action that we could take on this path. I'm not encouraging you to use any of the specific products I'm going to mention. It's just as an illustration to let you know how and what kind of products address what kind of levels of autonomy. Today we'll dive into this levels of.

00:03:02 - Bayram:
Autonomy. And my. The metaphor I'm using or the inspiration for this is the levels of driving automation. I'm sure you know the FSD's and terms like that, and probably some of you own an autonomous driving car, or at least a car on some level of autonomy. But as we can see, There are basically six levels. Five of them are levels of autonomy, but the 01 is just basically no automation. We do everything.

00:03:33 - Bayram:
Manually. And most of the cars out there are somewhere on the third or fourth level, which is not the case with Salesforce, as we will review today. But as you can see here, on each and every stage of this autonomy, there's a changing ratio of human versus AI role in this activity, in this case, the driving activity. So if we.

00:04:04 - Bayram:
Start doing everything manually, then maybe at some point we can just sleep. As you can see, on the fourth level of automation, we will be pinged by autonomy or AI when needed. Or at some point we can even just turn back and just have a casual conversation with your friends, which makes perfect sense. And I think this is a great illustration to how to think about sales org autonomy or any kind of.

00:04:34 - Bayram:
Autonomy. We start with L0, level zero. Everything is done manually, and we gradually through five levels, automate some parts of this process and achieve incremental gains in terms of the margins or time to close the deal, and similar metrics that are important for any chief Revenue officer or a founder. So let's review each and every level in detail.

00:05:06 - Bayram:
The first one is pretty straightforward actually. It's the 01. We have no automation at all. 100% of tasks are done by the human. Human is in control. Human is actually doing everything. I believe that most of you and most of the salesforce out there are actually higher than the zero level autonomy. Although I'm sure there are still some organizations on this level. But I won't spend too much time on this one. This is pre.

00:05:35 - Bayram:
Pretty straightforward. Humans are doing everything. And there are two issues with this level. First of all, we are limited in terms of our scale. Like one of our customers, for instance, she owns a travel agency, and every time they want to offer their services in the new market, she has to spin off an office in that country, hire people in those countries, people who speak local languages, manage those.

00:06:07 - Bayram:
People and things like that. Obviously, she has a physical and mental limits to the scale that she can achieve with her business. And this is the key blocker on this stage. That's why we want to automate some of that stuff and we are moving to the next level. So on this level, the AI acts as an assistant and basically we can say that the role is recommender and rec.

00:06:38 - Bayram:
Researcher because what they do is that they recommend the next best action. They can create content, they can research some stuff, say deep research in ChatGPT. But human is the decision maker. Is the decision maker. Human controls all decisions, cherry picks and tweaks AI outputs and provides feedback to the AI. And as time goes, as tasks are delivered, as time.

00:07:08 - Bayram:
Tasks are given feedback to the AI can learn and let us get to the next level. But let's review what kind of tasks specifically we could be doing and you're probably doing right now. The first one is we can let AI prep us for an upcoming meeting. This is a screenshot from our product where basically before an upcoming meeting, you can.

00:07:39 - Bayram:
Trigger AI to go and research this person and provide information like what this person cares about, what are the conversation starters that we could use, what are the notable accomplishments we could praise, and things like that. Basically, think of this as an intern that helps you prep for the upcoming sales meeting. Of course, maybe in B2C sales. This is not in the business to consumer sales.

00:08:10 - Bayram:
This is not that important, while for B2B business this is very important. You want to know what are the pain points? What are the recent highlights of the company you're going to try and approach? You want to know about the person. You want to know how to tailor your language, your content, even your presentation to appeal to the needs, pains and wants of those people. Of course, this is purely assistance.

00:08:40 - Bayram:
Of. And we can. What I, for instance, noticed in my experience is that sometimes we, as salespeople, account executives, we don't have time to do that. We don't have time to prep. And that ends up with basically lost opportunities because it takes from 30 to 60 minutes to prepare for a meeting, to do a deep dive into their. Into the company, into the person themselves, into a bunch of people.

00:09:10 - Bayram:
People if it's a group sales meeting. That's why this is very important. And cuts about 50% of times for sales prep. In fact, according to Gartner, by 2026, 50% of time we spend on meeting preps will be eliminated through AI and definitely we're on the path there. In addition to getting information from the public sources, of course.

00:09:40 - Bayram:
Course, information should be fetched from internal systems because that provides more context to the person. And one of the challenges in this space is to actually connect your AI to internal systems, let them learn and get the context from the CRMs, from the past meetings, from the conversations over WhatsApp, for instance, for this customer, this is a huge challenge that we will.

00:10:11 - Bayram:
Address at subsequent stages. The other very simple example, I'm sure you're probably using ChatGPT or Claude for these purposes, or you probably notice that Google starts incorporating those into many products that we use, but basically helps us to write good stuff. Me being not a native English speaker, of course I'm very interested in this, and I'm using this kind of functionality because.00:10:41 - Bayram:
It helps me structure my thoughts, it helps me deliver the message, and it helps me to basically appeal and talk the language of my customer, even if that customer speaks the language that I have no idea about. Like with my previous venture at Apnea, we were serving people all over the world, and sometimes we had an incoming support request in, say, Chinese or say, some sp.00:11:12 - Bayram:
Specific dialect in India, and we had to use Google Translate for those purposes. But now, and that would take us time, we need to switch between the screens and things like that. Of course, incorporating those suggestions into our workflow products, especially if they have access to the context from the CRMs and other data sources, this helps a lot in terms of preparation and delivering the message.00:11:42 - Bayram:
And last but not least, you probably noticed that one of the participants in our call right now is onsa sales associate. Basically, it's a meeting bot. I'm sure you're familiar with these kind of products that dials into the meeting and basically transcribes in real time everything you are saying or your meeting participants are saying. And in fact, it connects to your backend system, to the CRM.00:12:12 - Bayram:
And gives you some ideas and advices, some quick tips, how to address the questions or reminds you about the important questions that you forgot about. I think this is very important, again, to provide the context from those systems, because as we know, the more context companies, or sorry, AIs have, and actually people as well, and companies, the more context they have, the more personalized the tips are going to be. That's why.00:12:43 - Bayram:
The sales associate that we have not only transcribes and will actually send you a transcript of this meeting, but also provides me some in context tips and suggestions or reminds me of the stuff that I forgot. So again, the level one is more about assisting us, basically saving us time on some small task in a bigger workflows and.00:13:13 - Bayram:
Tapping into the knowledge that there is out there in the Internet. Moving up the partial autonomy. Now, the AI does more than just suggest. In fact, they execute some huge portion of the workflow. But they pause for your for humans guidance. So essentially the role is the task executor. But we need a sign up.00:13:45 - Bayram:
So think of this as a manager that has employees, but those employees do not have enough autonomy to act on their behalf. Say, for instance, an account executive approach you as a sales leader and asks to approve a discount for a corporate customer. That would be a good example of this level of autonomy. But instead of an account executive, this is an AI doing that. So.00:14:15 - Bayram:
They execute the routine steps, but they pause for your sign off. That's why human acts more as a supervisor or editor that basically approves or reviews at defined handoff points. Let me show you a couple of examples of this level of autonomy in action. For instance, one of our agents scans the CRM and your website and.00:14:46 - Bayram:
Creates an icp, an ideal customer profile, and then searches through open and proprietary databases to find relevant companies and contacts. It does that job, and it presents the list of what we call reference prospects and asks human to actually rate those prospects. For instance, let's say you're targeting the founders of tech companies that recently raised.00:15:16 - Bayram:
Their Series A. And unfortunately, sometimes open and proprietary databases can lack some information or context. Or for instance, the filters that AI used were not that accurate. That's why when AI presents this list of reference prospects, human has to provide the feedback on the relevance of those prospects. Basically, they score whether this this is a good ICP or bad icp.00:15:48 - Bayram:
And thereby provides the information or approves the reach out process to these kind of prospects. The same goes for the text with the messages, the outreach messages and the follow up sequences that are sent for this. And what's great about this is that we can make messages more personalized. Like for instance, one of our customer approached luxury retail shops and brands and we, our agent.00:16:19 - Bayram:
Collects information about that shop to the point of actually reading the terms of the terms and the terms of use of the website and who is processing the payments for that specific website to make sure that this luxury brand is actually processing the payment. Because what they offer is the payment engine that helps them streamline and support multiple currencies.00:16:49 - Bayram:
So in this case, agent does all of that job, provides that information. But sometimes, especially in the initial phases of learning, human has much richer context about the stuff you want to check about the stuff you want to double check. For instance, another customer that we source tech founders for, they have a limit in terms of what kind of founders they want to.00:17:19 - Bayram:
Approach in terms of what was the for instance, the amount they erased or the time they spent in the usa. So for these cases, sometimes in the initial ICP creation phase, when we learn from the one's website, the agent can miss those important criteria. So on this review and approve phase, human provides that feedback and then agent incorporates that feedback in those criteria.00:17:49 - Bayram:
To improve their algorithm, the search algorithm, and to make sure that the prospects that we go after on behalf of this customer are more relevant to this company then, for instance, further in the workflow. And right now, since we're still on the partial level, that another agent can actually reach out to that prospect. So you can see a screenshot from a Slack channel of one of our.00:18:20 - Bayram:
Customers where basically our AI agent provides information about the prospect and the history of communication with that person and then suggests response to one of the questions. So in this case, we can see that this person is on H1B visa. This customer provides immigration services, so they're on H1B visa, and they are open to other work visa options, like, for instance, O1 or EB1 and similar.00:18:50 - Bayram:
So what happens here is that this customer is not yet comfortable letting AI. And actually, I think this is great. And at this stage of the AI, I would strongly encourage you to review the messages that they write because there are some really bad cases that happened in the last three, six months with some very heavily funded AI sales startups where basically.00:19:21 - Bayram:
Without human in the loop, without human oversight and supervisory. The messages that were sent were out of context and actually hurt the brand of the company. The way we structured this workflow, every message that goes out on behalf of this customer actually is sent to the responsible account executive. And actually it's one account executive that manages multiple, basically accounts that we reach out on behalf of.00:19:52 - Bayram:
And suggest a response. But here you can see that the human acts as a reviewer or supervisor because they either approve the suggested response or they can edit and actually change the message. The outgoing message which will feed back to the algorithm of this agent. And further, with the more data coming in, it will improve the initial message and hopefully get to the point where most account exists are just.00:20:23 - Bayram:
Just click approve approve approve 99% of times. And when the cost of benefit, basically saving time, outgrows the cost of error, basically losing one in hundred relevant prospects, then we can do that in a very autonomous manner, which will basically help us get to the next level. And last but not least, again I mentioned the Sales Association.00:20:53 - Bayram:
Associate meeting participant that you can see in the meeting participants of the Zoom. This is basically our sales associate. And in addition to transcribing the message, it helps us and helps one to basically get the context and remind the important things that they should ask as we have the conversation, not afterwards, but as we're speaking to our customer, which basically helps us save.00:21:24 - Bayram:
Increase the outcome and save on the errors that we can have and we can just know about those post meaning moving on to the next level. Remember I mentioned the point where the trust to the AI agent that writes for instance the outgoing messages is so high and the stats are so good and the cost of benefits saving time and money.00:21:55 - Bayram:
Basically the time to close the deal outpaces and outgrows the cost of error losing prospect. And for some cases, this actually can be achieved after about three or four weeks of training the outgoing agent, message writing agent. Then we can move on to the next level where basically AI acts as a workflow orchestrator.00:22:25 - Bayram:
The idea is that they run the entire workflow end to end under most conditions, but sometimes it calls in humans for some ambiguous cases or some high impact exceptions. Again, say if it's early in the funnel, maybe the cost of losing this prospect is not that high. So we can tolerate 1% of inaccuracy, but when it's the final stages of the funnel and we are.00:22:55 - Bayram:
On the brisk of winning a million dollar contract. Well, maybe for these really high impact exceptions, we don't want to leave this autonomously to an AI. We want to control and be in control on every decision that is of the relevant and high enough high impact situation. So that's why on this case, AI actually does most of the job.00:23:26 - Bayram:
Entire workflow and just calls in human for ambiguous cases. What human does, as you can see from the illustration here, just basically monitors the outcome. Dashboards get alerts if the AI encounters something unusual or decision threshold is crossed, say, amount of deal and things like that. Let me show you a couple of examples with our customers. Unfortunately, I can't show you the internal systems, but I'll show you the diagrams.00:23:57 - Bayram:
That basically illustrate how it's done. So, for instance, with one of our customer, the entire outreach process is done autonomously. So AI actually finds prospects, launches the tailored outreach LinkedIn and email campaigns and books meetings. What human does is just account executive arrives to the meeting 5, 10 minutes before the meeting, reads that memo.00:24:27 - Bayram:
That I mentioned on the first level of autonomy and then hopefully closes the deal. The entire workflow is done here. Yes, absolutely. NAN is a great workflow. On our subsequent seminars, I will actually do some workshops where we will use specific instruments to automate entire workflows like this. And you're absolutely right that NAN is a.00:24:57 - Bayram:
Great way to do that. Workflows. But the basic idea is we can automate the entire workflow, but if the exception comes up, then we ask for humans decision. Here's another agent that we implemented recently, just about a week or two ago. Basically, it's a sales analyst agent. What it does is it observes all of the metrics that the previous agent, the outreach.00:25:28 - Bayram:
Agent generates. Like, for instance, how many prospects were contacted, how many prospects responded, how many prospects were interested, how many showed up to the meeting and things like that. And of course, we're experimenting with different Agent is experimenting with different aspects of that process. Essentially, every outreach process is four elements. It's who we're targeting, what we're telling them.00:25:58 - Bayram:
When we're telling them and through what means we're telling them. For instance, we could approach Byram through LinkedIn with the EB1 visa proposal. That would be a combination of those four elements. But what sales analyst agent can do is actually observe all of those metrics and find some insights and actually propose actions and recommendations to the.00:26:29 - Bayram:
Person. So in this case, this is actually a real diagram of one of our customers where we're testing different types of messages, as you can see here. And we're comparing the response and interest rates. Response is any kind of response to the outreach message, and the interest rate is actually prospect interested in having a conversation about this value prop. We can see that there are different types of messages.00:26:59 - Bayram:
Being tested short, long discovery and video. And we see that the lowest performance here is the video message. So what sales analyst agent does is actually spots this exception, delivers, communicates to the human responsible that, hey, the video message is performing poorly. So I will turn it off until you redo it and reassign the traffic.00:27:31 - Bayram:
To the other message types, in this case to the short and discovery types of message. It proposes this action but never takes it because the cost of error is huge and actually human want to be in control. The sales analyst agent can basically observe, analyze data autonomously, provide insights and actually some specific actions.00:28:01 - Bayram:
But never take those actions because these are exceptions and things that they don't want to miss. Another excerpt from Sales Analyst Agent Outcome and Analysis for our other customer that is in travel industry, what you see here is that after analyzing all of the requests or all of the leads for the last I think in this case was.00:28:32 - Bayram:
30 days or something. It encounters some interesting insight. Basically that there are some special requests. And what's great about LLMs is that the special requests section is always plain text. LLMs do a good job of understanding what was the special request. The special request could be, we need halal food, or we need vegetarian food, or we need a private driver the whole day and things like that.00:29:02 - Bayram:
That. And what it does is it analyzes all of the requests for the last 30 days, reads those special requests, and then comes up with ideas how to expand the market or the product offering. So you see here that the agents suggest to director, to the sales director here that, hey, you actually should come up with some specific packages because a lot of special requests.00:29:32 - Bayram:
Were for J. Lo's Jennifer Lopez upcoming concert. That's why it suggests bundling the tickets to those concerts with the offering that they have right now. So it spotted some new market preference, some new type of demand, and now it basically responds to that in terms of suggesting having a new product type and basically dynamically packaging.00:30:03 - Bayram:
That product type. Or you can see that it advises to actually build some partnerships with the local restaurants, especially Indian and Halal food providers, because that could be a great way to capture more revenue in that wallet of their travelers. This makes perfect sense. As you can see here. This is not only in terms of.00:30:32 - Bayram:
Expanding the product line, but also expanding the average customer value and things like that. Moving on further, you see in the previous case agent spots, that video message performs poorly but never takes action. It would be great if at some point we're so confident about the needs and skills and capabilities.00:31:03 - Bayram:
Of our agents that we can actually tell them that in these specific boundaries you can actually self remediate most issues, so you can self fix them. So in the case of video message, just turn it off and redirect. And again, that happens when there's enough trust in those recommendations. And assuming that, for instance, something could actually be done by this type of AI agent, let me give you a very good example. And it.00:31:33 - Bayram:
Addition to the previous one. For instance, agent spotted for one of our customers that the email bounce rate is too high. What it proposes is to basically remove the links from the message and use some other API provider to find the emails for those prospects to reach out to them. You see the way it spots the issue but fixes it autonomous.00:32:05 - Bayram:
So it self corrects in most situations while human actually just provides high level constraints and goals. Like for instance, close as many prospects as possible or get as many interested outbound leads as possible, reduce the bounce rates, avoid being included into anti spam lists and things like that.00:32:35 - Bayram:
And AI autonomously operates and does the self correcting measures. What we see like elements most of organizations are somewhere on the Level 1 or between Level 1 and Level 3, but we see some early indicators with some of our customers that level four is possible and the most the easiest. Actually the EAS.00:33:05 - Bayram:
Easiest sales process or sales workflow that could be automated with a high level of autonomy is leads qualification and prioritization. So what our agent does for two of our customers basically autonomously by having some initial boundaries and goals. And those goals not only book as many meetings as possible or qualify as many customers leads.00:33:36 - Bayram:
As high priority as possible. Some of the constraints could be hey, prioritize meetings with the higher score prospects higher in the calendar in the upcoming calendar of the account executive. Meaning that to avoid the situations where low priority, low score leads actually capture all of the slots of your account executives.00:34:06 - Bayram:
And you're losing on the interested leads that sometimes are actually inbound. So they're really interested, but you're losing because some lower priority, lower score leads captured all of the free time of your account executives. So all of this workflow can actually be done on the level four of high autonomy, processing the incoming leads, collecting information online and from internal systems.00:34:37 - Bayram:
About this prospect, prioritizing those, and then actually assigning specific account executives that are the best fit for this kind of customer. For instance, for instance, if this is a prospect coming from a specific industry, then we want to assign a meeting to account executive that has a lot of recently closed deals with the customers from that industry because his or her examples.00:35:09 - Bayram:
And cases will be more relevant and probably they have more context and confidence as well to close that kind of customer. So all of this process can be done with a high level of autonomy. And last but not least, eventual goal is where we achieve the Level 5 full autonomy, where basically we have an end to end revenue engine. And in addition to performing most of.00:35:39 - Bayram:
The workflows, all of the workflows, sorry and self correcting those actions. It actually learns, optimizes and executes all of the stuff that they offer and all of the stuff that they learn and they change the strategy, they change the icp, they change the target audience, they test new target audience based on early signals from the previous prospect and helps you exp.00:36:09 - Bayram:
Expand the market. Think of this as a chief revenue officer on autopilot that learns, optimizes and executes. And what's great about this kind of chief revenue officer is that it actually does the job of the chief revenue officer. Sometimes in the companies head of sales or chief revenue officers are too busy with some high impact accounts and they never have enough time to actually learn what's going on, learn the early signal else.00:36:42 - Bayram:
From the market, learn and teach and coach people. They just don't have time for this. That's how I see eventual level of autonomy for sales orgs where AI is essentially an end to end revenue engine while the human acts as a board of directors, as a visionary and governor that sets strategic vision and governance. Acts as board of directors but never actually drills down into specifics.00:37:12 - Bayram:
Think of this if you're into games, in RPG games, think of this as a player in the RPG game. Player has a lot of different agents. Agents have different skills capabilities, confidence scores, and things like performance indexes and things like that. Some agents focus on this, some agents focus on that. There is a player and there's a digital twin of that player that actually collects all of that information.00:37:42 - Bayram:
Makes the decisions, but basically just ask for vision and governance, similar to how board of directors act, of course, the issues and questions and challenges of ethics, challenges of things we want to do and won't do. Imagine a scenario where, for instance, AI agent this full autonomy.00:38:13 - Bayram:
Chief revenue officer agent learns that there's a very lucrative segment, say, in a country that is not, you know, very democratic, for instance, and autonomously it makes decision to actually enter that market. But you might get a backlash from your existing customers because they are not happy with you serving that market, that international market, because of.00:38:43 - Bayram:
The values that your customers have. So you don't want to get into that situation. So human board of directors provides this governance provides these values and constitution, if you think to basically avoid these kind of situations. And we know that this happened even with humans. We know biggest companies in the world entering new markets, say defense tech market and things like that. But then.00:39:13 - Bayram:
Actually retracting for it after a backlash from their employees, customers or shareholders. This is the kind of things that I think humans will have to do and act as governors and visionaries for these kind of agents. Just to summarize all of those levels of autonomy in details. You can review this deck later in details, but this has some specific example tasks for each and every level.00:39:44 - Bayram:
Of autonomy, in addition to what I covered and illustrated today, and I mentioned that the previous seminar is that it's always the case that you don't, you know, you learn about some great vision of going up the ladder of autonomy, but how do you actually do something specific? How do you apply to your specific organization? And the framework that I suggest is the 3, 2, 1 rollout framework, essentially.00:40:14 - Bayram:
It builds on the. I'm sorry, my. My keynote actually crashed. So this three to one rollout framework basically gives you some specific path, how to apply this and how to think about going up that ladder of autonomy. And let's review this rollout framework. We start with the simple things, things like.00:40:44 - Bayram:
Meeting prep report. And this is L1 autonomy level. But even at this stage, you have some early wins that let you gain trust from your peers, from your executive and supervisors that, hey, this AI thing is actually delivering some tangible value. So let's cut the current executives prep times, or let's at least encourage them to actually prep for.00:41:15 - Bayram:
For this. They don't have time. We know, but now they just have to read the report in five minutes before the meeting. This is much easier to do and hopefully they will be doing and hopefully this will increase our conversion rates. Then we could to win the trust and support of those account executives, we could automate some of their job, like for instance, logging call notes. We as founders, executives and chief.00:41:46 - Bayram:
Chief sales leaders. We need objective information and up to date information. This CRM. Unfortunately this is not the case with many CRMs because it takes time and if you have a bunch of meetings one by one, you never have time for that. And actually this is a great situation when you have meetings end to end meetings, back to back meetings. But the problem is that this learning and optimization flow of your sales org.00:42:17 - Bayram:
Network ports and you can spot the issues too late. So why don't we use AI to. We know that they can transcribe the message, but then actually log the call notes to the CRM, extract relevant information and deliver it to specific interested parties. Like, for instance, one of our customers. We log the call notes, we push them to Salesforce, and then they have an int.00:42:47 - Bayram:
Integration with Slack and some parts of those notes are put are pushed through to the responsible teams, for instance marketing market related information or competitors related information to the marketing team or for instance some feature request to the product team and of course all of that to the customer success or customer onboarding teams. Elite Qualification is a really quick win if you can.00:43:17 - Bayram:
Basically formulate what are your criteria for the best leads. But what we realized actually if you have at least 100 customers or deals closed, actually this information is already present in your CRM. So what you could do is actually learn your Use what we call an ICP agent to learn about the deals you closed in the last year or so. And then.00:43:48 - Bayram:
Formulate those criteria, qualification criteria, to automatically qualify incoming leads and again, save time for to save time for account executives to spend more on the most promising leads and thereby increase the results of your sales process. Inbound sales process. You do this quick wins. They're relatively easy. And I strongly encourage you.00:44:18 - Bayram:
You to do right now because the quality of these process is so high that it already delivers some tangible results. Then I have this framework from Stanford that's called Explore, Exploit, basically that you want to 80% of your time exploit the stuff, you know, like the three quick wins I mentioned. But then you want to spend 20% of your time experimenting so that.00:44:48 - Bayram:
How I suggest you try dealing with these second bucket of experiments. Basically, think of this as explore. Think of this as something that you want to try. You're not sure if you're going to get to success, but at least I guarantee you that as you experiment with this, it could work and deliver some really good results in terms of saving time and actually increasing the results, because Those agents can work. 24.00:45:20 - Bayram:
And things like that. Of course, similar kind of the tasks that we could experiment with could be an automated outreach and then account executive life coaching and debrief those in meeting assistants that actually learn about the meeting. And then I had a conversation with one sales leader and he said, byron, I don't need to scale all of my account executives. I want to scale.00:45:50 - Bayram:
I want to have a million copies of my best account executive. So every CRO has this job of making sure that the knowledge and skills of the best performing account executives are propagated to the entire organization. Because it's the case that the best account executives leave your company at some point. We know the case of Salesforce founder who actually left the company because.00:46:21 - Bayram:
He was one of the best performing account executives there. And I think this similar risk can happen to any organization. That's why the task of learning from meetings and from people and propagating that knowledge, transferring that knowledge across the company is very important. And that's where the executive life coaching and debrief agents can actually deliver a huge value. And of course, sometimes we need some.00:46:52 - Bayram:
Something huge, something that sounds like, you know, some new reality, sounds like a moonshot idea, but actually can deliver some interesting results. We see this with only one customer at the moment that the after the sales analyst agent realized that there's a very interesting new sub segment, actually a niche segment of.00:47:23 - Bayram:
Prospects that convert much, much higher than others, but that were underrepresented in our initial lead scoring and lead, you know, lead sourcing algorithm. It suggested to expand the play based on those external signals and expand the total addressable market for this person. And we did that, and we see huge wins from that. So a great success.00:47:53 - Bayram:
Criteria could be what percentage of your closed deals are actually from new segments. It's like this. Remember this matrix of I know what I know, I don't know what I know, I don't know what I don't know. So this is exactly the quadrant where I don't know what I know, I don't know that there are prospects in the specific niche segment that we actually reach out to, but they are so small that we.00:48:24 - Bayram:
We can't see that niche in the overall, you know, overall picture. But agents can spot those. They can source additional information. They can look for similar for signals that may be overlooked by people and deliver that insight to you and expand the play. That's basically the kind of rollout framework I strongly encourage you for. So to summarize.00:48:54 - Bayram:
There are six levels of autonomy. The metaphor of self driving cars applied to Salesforce. What we want to do first is to basically understand, where are we now? You could use the description that I provided today or specific examples of tasks that you can do. And you can just ask yourself or your sales leader, hey, are we doing this right now? Why yes or why not?00:49:25 - Bayram:
So when we realize where are we now, Then we can move on to the next stage, which is the 3 to 1 rollout plan. But maybe the buckets and the specific tasks you want to automate will be different based on your level. So for instance, if you're on level one, so probably maybe you could start qualifying lead qualification process or some, for instance, small aspects of it.00:49:55 - Bayram:
Or you could automate part of your prospect discovery or prospect sourcing play. And then you can pick those experiments, pick those early quick wins and pick those moonshot ideas and work them in a small group of combination of a salesperson and a person with the technical skills to be able to automate those tasks and workflows. My key takeaway today.00:50:26 - Bayram:
Is that? Basically, I think yes, maybe at some point in future we'll get to the point where AI acts as an end to end revenue engine. But right now, we're early in the stage. The trust is not that high actually. It's slow. And sometimes because of the stuff and actions of some players on the market, it can actually fall even lower. That's why right now, as we are just learning and as we as humans are evolving with the new.00:50:56 - Bayram:
Tools, because in the human, in the humanities, history, every new tool that we had actually not only increased our productivity, actually it helped us to evolve, to rethink what we're doing, to rethink why we're here on this planet and things like that. So similarly, AI is a new tool. It will help us, and actually it will encourage or maybe nudge or force us to.00:51:26 - Bayram:
To evolve together with AI because we will realize that we need to think more about the ethics and values rather than how to prep for the for the meeting. That's why it's more of a team play, AI and humans, rather than completely one or other dominating here. So that's all I wanted to share today. Hopefully this gave you some specific examples in picture how to approach this.00:51:56 - Bayram:
Sales autonomy levels and how to actually realize where you are now and how to move. Yes, this is a great question, Andrew. I think the best slide that addresses this question is here. So you see that depending on what kind of process or what kind of workflow I'm targeting or I'm automating, we have a difference.00:52:27 - Bayram:
Success KPIs. As you can see here, with a meeting prep or logging call notes to CRM, we're actually freeing up time of AES. This may not directly impact the revenue numbers, while for instance the lead qualification could increase the conversion rates the SQL per prospect metric and thereby increase the revenue. So what I'm trying to say is that.00:52:57 - Bayram:
I think in terms of, for instance, metrics and the impact on metrics, because productivity for each type of role, for each role in the sales process mean different things. That's why I think we should think further, you know, think more broadly about this question. And but to be specific, I think in the on the first two L1 to L3s were getting more on the cost.00:53:28 - Bayram:
Side. So we're saving money, and we're. We can basically either process higher volumes or optimize our headcount. While on the upper, you know, level three, level four, level five levels, we can actually increase the revenue by expanding the market or doing more for the same calendar time. So my, my best answer would be,00:53:58 - Bayram:
Like this. So level five, probably in terms of the revenue will be on the scale of a couple of, you know, two, three times. But the path there is very, you know, is very challenging. Do you address the issue of human natural impulse to not speak? Yeah, that's a good one. Actually, we have another problem. We have a problem that, in fact, sometimes.00:54:28 - Bayram:
A human actually sends a message, but the prospect thinks it's an AI. And this is very funny, but this happens, unfortunately. So, yes, there is a natural impulse to avoid AI agents, and that's why we try to improve the prompts. Like, for instance, I've once shared a prompt that was used in one research to fake to basically convince people that.00:54:58 - Bayram:
It's a person to basically pass the Turing test. And if you look into the specific prompt, you will see instructions like hey, make occasional typos because people make typos, while AI agents never without explicit instruction. Or for instance, we have to configure and train and put the examples of the style of this person into the prompts or into the.00:55:29 - Bayram:
Fine tuned version of the models LLMs that power those message writing capabilities to actually act like this person. It takes time. There is a challenge, but there are very low hanging fruits. Frankly, if you review all of the quick wins experiments and moonshots I mentioned here, most of them are not about writing text. I think this is very important because.00:55:59 - Bayram:
We should focus on the stuff more on in the back office backend things rather than front end because we still can and should care about those things. Sorry, I think I missed another one. Hold on, John. What tools are used for build? So in our case, we're using code because we're developing a product, but when I'm prototyping.00:56:30 - Bayram:
My typical scenario when I'm prototyping is like this. I first use ChatGPT to actually let me give you an example of sales analyst agent that we recently built. So I started with basically taking all of the outbound campaign data, feeding IT to the ChatGPT deep research and asking it to prepare the report. Then I reviewed the report, I made some.00:57:00 - Bayram:
Suggestions. I got to the point I like it and I showed three reports to three different customers and I got their feedback as soon as I was confident that we can do. First of all, customers are interested in these kind of insights and I knew where the tweaks could be applied to improve the whole outcome of this process. I moved on to prototyping phase.00:57:30 - Bayram:
And then now after prototyping, we'll incorporate it as a standalone agent. What I'm saying is that tools, even ChatGPT tools, are very important and useful in the early proof of concept prototyping phases. And N8N is a great example. You probably would need some information from public data sources. That's why I strongly encourage you to look into mcps. This is a new product called.00:58:01 - Bayram:
Call to connect to third party APIs to avoid reading and learning their APIs through invoking them through Postman or things like that. Basically, you can connect your LLMs to your agents, LLM powered agents to those APIs through MCPs without actually learning how they work, how to write the code to invoke them, and use those to prototype. So essentially what I do is actually.00:58:31 - Bayram:
In corsor. I can connect different MCPs. I can prototype right there in the chat and the Agent Space code space and then deliver the results, test those results, and get back to incorporating them. There are two great services that help you actually collect information about a given person and the company that we use. First of all, it's.00:59:01 - Bayram:
I'll just share some examples. So this is a great service that provides. Actually provides. Hold on. This is a great service that provides MCP server and APIs to get information about people from LinkedIn. So it helps you not only.00:59:31 - Bayram:
Only LinkedIn we're using for LinkedIn purposes, but actually it covers much more than that and gives you a great way to basically automate that part of sales outreach agent that actually sources prospects through different means. X LinkedIn and some other places. The other the other service that we frequently.01:00:01 - Bayram:
Use and they in a way actually cross in terms of functionality offered. They similar to Horizon data wave, it's unipile.com that helps you automate the outreach through different means WhatsApp, LinkedIn, email and things like that. So that part of the out Sorry, I'm not sharing that part of your journey could be actually automated of the same.01:00:31 - Bayram:
Sales outreach process. And of course John mentioned N8N this is a great exact this is a great way to automate some stuff and actually connect those APIs that I mentioned to process some steps of the workflow. Hopefully I addressed your question, John Boris, this is a great.01:01:02 - Bayram:
Question, Boris. And first of all, remember when I mentioned that sales associate assistant that we have on our call right now, and I mentioned that we need objective and up to date information in the CRM? You're absolutely right. That unfortunately, garbage in, garbage out. And if we, like for instance, with one of our customers, account executives, actually close the deal, create.01:01:32 - Bayram:
And close opportunities in Salesforce when they actually closed the deal. So not before, not when they had the initial contact with the customer and things like that, but right when they actually closed the deal. So if we use sales analyst agent to reason about this person's, this account executive's performance, there is an exception because they close opportunities in like couple of second.01:02:02 - Bayram:
Seconds. Wow, this is cool. Of course, this kind of exception will be highlighted by AI agents. And what I know is that garbage in, garbage out. You're absolutely right. But we can have agents that are responsible for increasing the quality of data that comes into the system or to the context of the agents that rely on that data. So, like,01:02:33 - Bayram:
In software development. When we develop products, we have quality assurance engineers. So we should have data accuracy agents that are responsible for ensuring a high quality of data to prevent exceptions or things like I mentioned with the 1 second deal close performance and the sales associate agent that.01:03:03 - Bayram:
Dials into your meetings and then uses that information to log into the CRM with human supervision. If we are on L2, for instance, or data Accuracy Agents, those agents will and should be responsible to increase the quality of data in those systems. So yes, you're absolutely right. In addition to that, it's not only about that salespeople are worse when it comes to.01:03:33 - Bayram:
Data filling. And that's why we want to automate that. And that is the sales associate we have, agent we have for. But the other thing is that sometimes human. And I think this is great, actually, this is a very important character of any account executive or a salesperson. We're very optimistic. Like, we think this deal will close. We assign higher probabilities to these. While in reality this could not be the case, we say,01:04:04 - Bayram:
They have budget. But this was never mentioned on the call. Actually, either some aspect of that information was somewhere in a different channel, or this is an assumption that there is no evidence for. That's why those agents are very useful in terms of increasing the data quality and actually reducing or neutralizing this excessive.01:04:33 - Bayram:
Optimism of our account executives. Hope it makes sense. Do we have any tool for which helps better analyze work of exist salesforce? Yes, of course. For instance, you can use Basically, we we have an agent that learns actually dials into all of the meetings, processes the meeting notes and then provides summaries to the head of sales about.01:05:05 - Bayram:
What's going on, who performs better, who performs worse, and suggests some training and things like that. So, yes, we offer that agent, but actually, there are some other companies out there. I strongly encourage you just to, you know, for instance, gong.com is, sorry, IO is very, you know, established player in this, in this market. There are some very focused sales coach products out there.01:05:35 - Bayram:
So yes, there are a bunch of tools probably established player wise, this is gone. But there are new tools that pop up, I think, every day. And what's great about this is that I see specialization happening in this space, meaning that, for instance, a sales coach for a medical organization or for a hospital or for a tech company differs from a salesperson in for.01:06:05 - Bayram:
For instance, consumer product business. So what's happening is that different sales coach have different playbooks. Like for instance, there are some heavy frameworks like Medic or Spin or step pay and things like that that are used for B2B sales, but they don't make sense for B2C sales. That's why I see some specialization happening out there, okay?01:06:37 - Bayram:
Okay, I think I addressed all of the questions. I hope this was useful. You will get the deck and recording link to this recording in the next day or so, as well as a detailed transcript because we have this sales associate that actually transcribes and will summarize this meeting. Thank you. And please let me know if you have any suggestions or preferences for.01:07:06 - Bayram:
For some subsequent seminars because. Or topics. Because that's what I'm going to cover. We will have these webinars happening once a month or so, and I'm very happy to address the needs and wants that you have in terms of moving up this ladder of sales or autonomy. Have a good one. And if you're celebrating, Happy Memorial Day, goodbye.