Why Your AI Initiative Is Stalling (And What to Do About It)

Full Transcript

Jill: Hello, everybody. Welcome back to another installment of Machine Logic. I’m Jill Golden, and this is my partner, Charlie Nadler, and we’ve got a special guest today. Very excited about a big trend that we are seeing right now with a lot of AI and go-to-market is folks who have really neat ideas and who are tinkering and piloting with tools, but then they get stalled when they actually go to operationalize these initiatives. And so we couldn’t be more excited to have Nick Bhavsar joining us. Nick, we’ve partnered on several projects with him, and he is an outstanding AI go-to-market trainer, consultant, and helps essentially tech and SaaS revenue teams use AI to accelerate their pipeline. And he has a tremendous finger on the pulse of this. And so we wanted to bring Nick into the fold today to join us on Machine Logic. Nick, welcome.

Nick: Yeah, thanks, Jill. It’s good to chat with you again. Thanks, Charlie. It’s always good to catch up with y’all. So it’s fun trying to keep up with all this crazy stuff that’s going on in AI right now. So excited to chat with y’all. I’ll give you just a quick sort of background on me as well. I’m based out of Austin, Texas, go-to-market consultant. I’ve been doing sales and marketing B2B for, gosh, now 25 years or so. I cut my teeth at a company called SolarWinds and then Spiceworks and a company called Levelset and then launched and co-founded a company called Velocity Engine that was in the go-to-market space. And then more recently, just been all in on AI and really trying to help teams leverage and gain productivity through the various tools. Excited to chat with y’all.

Jill: Terrific. Yeah, I know your experience is deep in this area and very exciting. So talk to us then, Nick, about what you’re seeing. I know we’ve had some cross-pollination on several clients and projects, but what else are you hearing when it comes to what companies are up against when they’re trying to get AI initiatives up and running and to stick?

Nick: Yeah, it’s a great question, Jill. So I think part of it — we’re on this evolution and different people are on the evolution at different stages. Just in the last six months, the amount of AI acceleration that’s happened has been tremendous. I’d say really since the birth of ChatGPT three years ago, the first two and a half years were essentially kind of like Google replacements — people were using chat to go fetch information, asking different questions. And then probably in November/December we started to see this shift where you started to see things like Claude, Cowork, Code, and all kinds of different applications start to come up. And so around Q1 of this year we started to see skills and agents and connectors — people and companies start to use AI beyond just fetching information, but actually getting insight from Salesforce or HubSpot or different tools like that. And that dramatically changes what’s possible.

And then in Q2, we started to see people really start to tap into actually building workflows and building agents that can perform different tasks. Some of the tools have helped a tremendous amount there. Even Claude has functionality like scheduled tasks where you can actually do something like prepare a weekly report for me. So seeing that evolution where people who are non-engineers can now create these different workflows themselves.

And then just coming up to today, what we’re starting to see is people have those individual tasks and they want to deploy them enterprise-wide. They want to do these workflows where they have processes they’ve been doing and they want everybody on their team to be able to do this. So it no longer works on your laptop. You have to put this in some cloud infrastructure. Starting to see real adoption from tools like N8N. That’s sort of the progression — going from a Google replacement to workflows, and now trying to get this deployed throughout different organizations.

Jill: Yeah, so with that influence — you said people have implemented something for themselves and now they want to take it enterprise-wide. It sounds like there are some other players that need to be looped in regarding how to actually bring this to life organization-wide, and some processes where it’s like, hey, this works for me but how do we deploy this all around? That’s a lot of moving parts. How do you help folks reconcile those three things to actually roll it out?

Nick: Yeah, it’s a good question. You know, when we think about software and SaaS, what are people doing all day? They’re basically interfacing with some database — whether it’s HubSpot, Salesforce, whatever. They’re looking at some data, manipulating some data, using their judgment, creating some other things, and then putting it in another database. That’s really all we’re doing. And it could be creating a marketing campaign, sending an email, running ads, things like that.

The old workflow tools — N8N, Make, and tools like that — did really well when there was really good structured data and they could move data in and out of different tools. But as we know, and y’all are really good experts at this, data is not always clean. It’s just a mess oftentimes. And so this is where AI can help — it can tap into less structured data. It still needs some structure and reliability, but if it’s not exactly the same every time, it can still adjust and modify.

So what teams are now trying to do is, at the macro level, look at what the labor is essentially doing — taking information from one database, pulling it down, adjusting it or making judgment calls, and then putting it somewhere else so it can perform some action. That’s essentially a workflow or an agent. And the really savvy companies are now starting to map that out and say, okay, what is our business, what do we do very often, let’s map all that out, and then let’s have AI do those functions.

A quick example — I just helped a company last week that had an HR wiki where it’s like, do we have July 4th off, what’s our policy for this. Usually they have those wikis and people can go find them and search, but employees don’t always do that. It’s some huge document, hard to navigate. So what we did was build a little HR Slack bot. You go to Slack, type in your message, it goes into the Slack bot, queries the relevant documents, synthesizes that information, and gives a concise response in near real time. Your employees don’t have to figure it out. It helps the HR team because now they’re not taking triage from different questions all day — they can just update the document when new policies come out. Very simple workflow, but right away that probably saves five to ten hours a day of somebody going and fetching that information. And employees feel more comfortable asking sensitive questions because they’re asking a bot instead of a person.

The key is mapping out those workflows and understanding the process before you try to create a system to do it.

Jill: Right on. And you touch on some things that I know fall squarely in Charlie’s world a lot these days — the load-bearing but very unglamorous work of data hygiene, data trust, processes, enforcement. Which isn’t the most sexy part of AI, but is very foundational. Charlie, can you speak a little to that?

Charlie: Yeah. And Nick, you know this well too. A lot of clients we’re helping are under tremendous pressure to deploy AI — for marketing, for sales, for service ops. They’re being pitched constantly: “what if we could turn all of your operations into AI today?” It’s very tempting. And then where we tend to come in is: okay, yes, absolutely, we should be activating. But what is the process? Is it documented? How is data being entered? Can you trust it? We have our methodology for snapshotting that and then diagnosing and fixing it.

Are you seeing a similar thing when you’re first engaging with clients? Are they feeling that pressure or is it more self-driven? And I’m also wondering, from a people readiness standpoint — culturally, how far along are they? Are they ready to go, or do people need coaxing?

Nick: Yeah, I definitely see it at the executive and board level. We’ve seen companies where the board discussion is, how do you increase the productivity of the teams? And that can be very extreme — you need to increase productivity by 50%, capital is very restricted right now. And AI is that ultimate productivity hack, at least that’s the promise.

So there’s tremendous demand — never seen demand like this for an initiative, both from the top-down executive teams wanting more productivity, but then also from the employee side. Most employees want to embrace these tools. They want to take the more execution-oriented aspects of their jobs and be freed up for the strategy side. But with that comes a tremendous fear as well. Everybody’s worried — is this going to take their jobs, is this going to replace them?

What I find is that the savvier companies and employees quickly figure out: look, this is just the way work is going to happen in the future. If you can get on board and get ahead of this, you can take your expertise and do the human aspects that are much more difficult for AI to do, and create a lot more execution power.

Charlie: Because your process — you’re doing listening tours to figure out where people are at. As you’ve been getting this front-lines feedback, where are the biggest gaps that you’re seeing? The things that need to be figured out before companies can really start activating?

Nick: Yeah, so my process typically has three different flows. First, just sitting down and doing the assessment — sitting with the executive teams and departments, trying to understand how things are working, what tools they’re using, what challenges they have. And right away, we typically see one of three buckets.

The most advanced folks — usually engineers and product people — are doing agentic coding, leveraging the latest tools, and frankly don’t need help. They’re off to the races.

Then you have a tier in the middle — the excited and productive — who are starting to do things with these tools. We’re essentially moving towards engineering-like motions for people who are not engineers. Workflows, agents, skills — all pretty technical terms, but the capability to do these things is now possible for non-engineers. That group wants to know what the latest skills are, what’s approved by their organization, how to get access to different databases. They’re pushing aggressively for more access. The big challenge for them is that because they’re not engineers, they don’t think about things like security or data hygiene or the back-end infrastructure required to do things at scale. Sure, you can whip up a little prototype to do something kind of cute, but when you’re trying to do it across your organization for tens or hundreds of people, it needs to be well thought out and done in a more robust way.

And then you have that final group that’s sort of deer in the headlights — still using ChatGPT as a Google replacement. They’re very interested, but getting them on board typically requires some training: get access to your email, your calendar, your project management solutions.

Too often people lump all three cohorts together and that’s a mistake. If you put the engineer and the newbie in the same training, the engineer’s frustrated because they already know all this stuff, and the tier one person is freaking out because they’re looking at code they don’t understand.

The traditional way is to group by department — marketing, sales, customer success — and do trainings that way. But you have people at different extremes within each department. A better way is to group people by their AI maturity and train them accordingly.

So the three aspects of my process: one, get a good assessment of where your maturity is. Two, figure out training — and this is not a one-time thing, these tools are advancing incredibly fast, you have to continue. Three, go department by department and figure out the workflows. What are the repeatable tasks? And there are two phases: can we automate and agentify existing workflows? And then, what can we now do that we could never do before?

Some of the use cases we’ve talked about — the Slack bot, meeting prep for sales organizations — having a bot go off and research who you’re meeting with, what’s happened with the company recently, thinking through implications, identifying other stakeholders. These all have immediate revenue impact. AI doesn’t do it perfectly, but it can assist quite a bit.

Jill: It’s interesting how much this puts a spotlight on competent management. Because everything you just talked about — prepping for meetings, analyzing deals, using recordings — that only works if everyone’s doing it in equal measure. If you’re recording your meetings and I’m not and Jill does it half the time, AI can so easily spit out something that sounds smart but is completely missing the point because it’s only getting a slice of the picture. So in one way, AI is helping these companies do so much more, much quicker, but it’s creating new work. Essential work — like, what is this process, and how do we make sure everyone’s actually doing it? Otherwise we’re scaling inaccuracies.

Charlie: Right. What operational weaknesses is this exposing now, as we layer AI in? That’s real work people have to do, and it’s not something another AI tool can solve.

Nick: Absolutely. And it starts to highlight the holy grail here, in my opinion. You have some centralized database — could be Salesforce, could be HubSpot — and you have these agents doing work. It’s one thing to automate: at the end of a Gong call, take the summary and send it out or stick it in Salesforce and fill in the fields. No big deal, and you even get better adoption that way because the AI always does it whereas a sales rep sometimes does it.

But the real unlock is how do these different workflows and systems talk to each other? How do we have product marketing see in real time that a new competitor is starting to come up, and then that information makes it to engineering — this competitor is touting a feature we don’t have — and then that new product feature makes it into customer success stories and release notes? That’s what companies are trying to strive for.

We talk a lot about what is the first trillion-dollar company with three people. The systems are all talking to each other in real time, passing information back and forth, making those decisions. We’re a ways away from that. But the first step is taking the labor and agentifying it, putting these processes in place, and then starting to have an overarching management layer saying — are we doing all this? Do we need to be doing all this? Can we cut some of this out? What’s the most effective way? Really exciting and terrifying at the same time.

Jill: As we kind of wrap up here — if your company is at that inflection point where it’s like, all right, we’ve been piloting, it’s time to take this seriously and operationalize it across the enterprise — what advice do you have for those folks as far as what to expect and how to measure progress?

Nick: Yeah, it’s a great question. The old rules of business don’t fall apart here. You just have to start by saying: what’s the real challenge for the teams, what’s preventing them from hitting their goals? And then it almost always comes back to some amount of training and helping teams be more productive, and then identifying a few of those processes or workflows, prioritizing them, and having some business impact tied to them.

I just talked to a client the other day who mentioned their sales team spends eight to ten hours a week on non-selling time — updating Salesforce, summarizing calls, getting insights, internal meetings. If you can cut that in half and give yourself four hours of selling time, that’s a meaningful impact. It immediately hits the bottom line.

We’re not just in the business of agentifying and creating workflows. We’re in the business of hitting revenue goals and milestones. Really getting a good understanding of what the bottlenecks are — the same way we think about OKRs. And then not trying to solve all of it at once, but chipping away: let’s try this area, let’s build out a workflow, let’s see if we can get it automated.

The other thing is that automation can be done in stages. It’s not bad to prototype something, get it working on your computer for an individual person, then get it working in the cloud for more people, then get engineering involved where it goes through a more robust security audit and is built for enterprise scale. This is not wave-your-wand AI that’s just going to fix a bunch of things. There’s real work that has to happen. But it can also dramatically improve the productivity for those teams.

Jill: Yeah, AI is being marketed as instant transformation. But for most businesses, that is just not the reality. Early progress looks like the use cases you shared — reduction of manual effort, incremental efficiency gains. Our respective organizations are a testament to that — people need help getting there. We really appreciate your perspective on all of this.

Nick: Yeah, absolutely. It’s always fun to catch up with y’all. I appreciate the invite.

Jill: Hey, if you’re up for this again — you touched on some things that are worth a deeper dive. Additional use cases as they bubble up, governance, prioritization, the change management component. The awareness about AI isn’t the gap anymore. It’s the operational readiness, and that’s the part no one should skip. We’d love to do a deeper dive on some of those things with you in the future, Nick.

Nick: Let’s do it. That sounds great. Thanks again.

Jill: Awesome. Thank you. Thanks, Nick. Thanks, everybody.