William Lopez has 10 years in AI and 17 in go-to-market, and works at ABL Technology Group, a holding company behind six startups that build full go-to-market orchestration for enterprises with $80 billion or more in global revenue. Instead of off-the-shelf language models, his teams build mathematical optimization, digital twins and AI agents that read the microeconomic pressures hitting a client's business.
Connect on LinkedIn ↗One of ABL Technology Group's clients, a major bank, spends between $10 and $12 million a month on GPU tokens. Not on strategy, not on licenses. On the raw compute behind models that several internal teams had been building in parallel without talking to each other. The bank hired ABL to find out where the money was going.
That number is a good place to start a conversation with William Lopez, founder and CEO of ABL Technology Group. ABL is a holding company for about six startups that build go-to-market orchestration for enterprises doing $80 billion or more in global revenue. Lopez has spent ten years working in AI and seventeen in go-to-market, and his read on the current wave is unusually blunt.
Lopez draws a hard line between what ABL builds and what most of the market sells. "We don't do any off-the-shelf stuff [...] that's garbage," he says. The work underneath is mathematical optimization, operational research, and digital twins used for simulation testing.
For him, a large language model is a thin layer on top of that: "It's not good because it just scrapes public data and says, 'Oh, this is a good email for you to send. This is a good person for you to call.' [...] Any kid can build that."
What ABL builds instead reads micro signals and microeconomic stressors around a client's customers. His example: a drought means fewer tomatoes, which pushes prices up for a food service company operating at $80 billion in revenue. That is a long-term economic stressor, and it changes how a sales team should open a conversation. His instruction to reps is to lead with curiosity rather than a diagnosis: "just say hey, you know, we just heard this, I want to know how this is affecting you."
ABL runs on premise, which means the infrastructure bill lands on them directly. That is why Lopez keeps coming back to hardware selection as an engineering skill rather than a procurement detail.
"You could be spending a million dollars to develop an AI using very high-end GPU, but at the end of the day, the function of that AI [...] doesn't necessarily require you to use the H level GPU," he says. "You can use the A level that will cut the cost down from a million dollars to $200,000."
In the bank's case, the inefficiency was organizational. Ten groups were building two different classes of AI, the agents themselves and the deployed systems working with brokers and traders, with no coordination between them.
Asked where the opportunity is, he named three areas: sophisticated machine learning, operational monitoring, and optimization modeling. His reasoning is that everything else is crowded. By his estimate, 70% of what LLMs are doing today can already be done with open generative tools, and 90% of companies are developing things that will not work.
He also expects the interface layer to flatten out. "Prompt engineering is going to go away, everything is going to be natural language processing," he says.
What survives, in his view, is the ability to work alongside the system rather than around it: "The only differentiator for talent is going to be what do I bring to the table that works cohesively with AI?"
Lopez puts a number on it. About 30% of staff currently doing ICP development, market penetration analysis, market opportunity development, and forecasting can be automated now.
He is equally direct about the discipline itself. "Go-to-market orchestration has been overhyped in the last year," he says, and it does not fit every company. "If you have a small company that's doing $10 million a year, you have to be an idiot to use GTM." His argument is that the machinery is overkill at that size and the real fix is internal: remove the silos between sales and marketing so the message stays consistent. "Everything that a company does, marketing, LinkedIn post, everything is sales oriented," he says. "It should be the same cohesive message [...] aimed at solving the problem that's relevant, not just generalized problem."
The projects ABL takes on run between $700,000 and $5 million, and Lopez uses price as a rough filter. "Everything is about saving money or making more money," he says. "If you don't solve any of those two problems for a company, then you're not going to get anywhere."
He walked through the math on a customer support use case. A company spending $35 million a year on service calls saves $3.5 million by cutting 10% of that volume. If the solution costs around $900,000, the payback lands near eighteen months. The mechanism is unglamorous: the AI handles the level-one script (did you unplug it, did you restart it) and hands off with a diagnosis attached once it runs out of road. He estimated that around eight minutes of a thirty-minute call fall into that category, and named Carrier and Shark Ninja as companies that understood the problem before buying the tool.
What does not work: "A $99 off-the-shelf AI agent is not going to solve anything."
The failure patterns he sees are mostly organizational. Codependency on the tool. Shadow AI creating security holes. No AI governance, no AI ops, no management layer. And the one he calls the biggest problem: companies adopting AI before they can measure it. "They are thinking that it solves problems that they don't even know how to measure its effectiveness," he says. His fix is a scheduled review at three, six, and nine months with actual metrics attached.
Underneath that sits something softer. Lopez expects AI companions to become normal at work, closer to a coworker than a tool. He already uses a few. "I get my priority list from my companion. I get my schedule for the day," he says, and expects the same systems to eventually give feedback on how someone is performing.
Whether that works depends on two things he frames as a relationship question: "Do you trust their skill set? [...] The veracity of what they say, do you believe it? Can you check them?" And then the harder version: "Does your inexperience allow you to continue trusting something without checking it?" He calls this the overlooked piece, and says he is surprised how few people think about it.
Lopez's closing argument was aimed at startups in the region, and it was the least comfortable part of the conversation. The democratization of technical knowledge has made the talent pool competitive, he says, because "you can acquire it from the comfort of your house in Medellín." But he sees a trap on the other side of that: "If we don't nurture that to create uniqueness, then we become a commodity. Then we're just bidding on who's the cheapest."
His alternative is to go narrow rather than wide. "Solve local problems because that's where the market is," he says. "No one in the US is going to solve your local problem. They don't care about it." The advantage is not cost, it is proximity: "No matter how much market research I do, you know how people interact in the store, how people interact with technology."
He offered a concrete opening, prompted by the interviewer's background in law. Digitizing notarization, or building a certified document vault for the paper trail people leave behind when they die, with a lawyer and an accountant attached. Red tape, heavy process, clear local rules. He pointed to how the pandemic made digital banking normal across Latin America as proof the trust barrier is already down, and framed the strategy as finding the path of least resistance.
Three things carried through the whole conversation. Cost discipline is an engineering skill, and the GPU decision is where a lot of budgets are lost before anyone reviews the model. Measurement has to exist before adoption, because a system nobody can evaluate cannot be corrected. And the human stays in the loop for judgment, context, and verification, which means the trust question is a design problem rather than a training slide.
Lopez's own summary of good implementation work is worth sitting with: "We implement a lot of AI. Nobody sees it, but we're good with that because nobody needs to see it. If you don't see it and you can't tell it, then we did our job."
