Avery Chi on Building AI Systems That Actually Work

As artificial intelligence becomes embedded across enterprise platforms and business operations, organizations are under growing pressure to move quickly. But successful adoption requires more than access to powerful tools. It requires experienced professionals who understand how technology, people, processes, data, and business objectives fit together.

Avery Chi has built her career at that intersection. With more than six years of experience implementing enterprise SaaS solutions across governance, risk, third-party risk, operational resilience, and incident management, she has helped organizations translate complex operational challenges into scalable technology solutions. In her work as an Implementation Consultant and Solutions Architect at Riskonnect, Chi has developed deep expertise in implementation strategy, systems thinking, stakeholder alignment, business requirements, and organizational change.

Her experience gives her a distinctive perspective on one of the most important challenges facing companies today: how to use AI to create meaningful business value rather than simply adding new technology to existing problems. She is also an entrepreneur and business builder, bringing a broader perspective that spans technology, advisory services, business development, and professional services.

In this interview, Chi discusses what separates successful AI adoption from expensive experimentation, why organizational readiness matters as much as technical capability, and how leaders can use AI to strengthen decision-making without losing the human judgment required to make technology truly effective.

1. As AI becomes embedded across enterprise platforms, why do you think organizations need to define the business problem before deciding where or how to deploy AI?

AI is powerful, but adopting it simply because it is available is backwards. Start by identifying the real problem: where the friction is, what is causing it, and what outcome needs to change.

Then determine whether the root cause is technology, data, workflow, ownership, or people. Before asking what AI can do, get clear on what actually needs fixing.

AI can absolutely help diagnose problems as well as solve them. But even the most powerful tool needs someone who knows where to look and what questions to ask. Before asking what AI can do, get crystal clear on what needs to change.

2. You’ve spent more than six years implementing enterprise SaaS solutions across governance, risk, third-party risk, operational resilience, and incident management. What has that experience taught you about the organizational conditions that need to exist before AI can create meaningful value?

The technology can be ready while the organization may not be. 

Across dozens of enterprise implementations, I have seen sophisticated technology underperform because the surrounding conditions weren’t there: unclear objectives, fragmented ownership, weak data, poor adoption, or stakeholders who were brought in too late.

Meaningful AI adoption requires three things: clarity, ownership, and participation.

People need to understand what problem matters and why it matters. Someone needs to own the outcome. And the people whose work will actually change need a voice in designing that change.

That’s where change management begins, not after implementation, but during design.

When people help build the change, they are far more likely to own the change. AI readiness isn’t just technical readiness. It’s organizational readiness. 

3. AI and automation can make inefficient processes move faster without necessarily making them better. How can leaders determine whether they have a technology problem or an underlying workflow, ownership, or operating-model problem?

Before automating the process, interrogate the process.

Where does the work actually start? Who owns the decision? Where does it stall? Where are people creating workarounds? What has become so normal that nobody questions it anymore?

Organizations learn to tolerate dysfunction. Sometimes it takes fresh eyes and the willingness to play devil’s advocate, to notice what everyone inside the system has simply learned to live with.

But don’t assume an awkward process is automatically a bad one either. Something that looks unnecessary may exist for a reason. Take it apart. Pressure-test it. Ask what breaks if you remove it. 

Then the diagnosis becomes much clearer: Do we actually need better technology, or are we asking technology to cover up a problem somewhere else? Don’t automate the mess before you actually find the mess.  

4. Why can an AI implementation be technically successful and still fail to produce the business outcome an organization expected?

Because “it works” and “it worked” are two completely different statements.

A system can be configured correctly, integrated successfully, and perform exactly as designed, and still solve the wrong problem.

There is another failure mode I have seen repeatedly: we design the technology and underestimate the human system around it. Successful adoption depends on understanding how roles, decision-making authority, accountability, and day-to-day behavior will change once the technology is introduced. If those shifts are not addressed deliberately, even a technically strong solution can fail to produce meaningful results. 

That is where change management stops being the soft side of implementation and  becomes part of the architecture. 

One of my favorite questions is deceptively simple: If this works exactly as intended, what will actually be different? If we cannot answer that in both business and human terms, technical success may simply give us a very sophisticated way to stay where we are. 

5. Enterprise AI often touches multiple teams, systems, and decision-making processes at once. How should organizations think about people, processes, technology, data, and organizational dependencies when designing an AI-enabled workflow?

Start with the future state, then work backward through the system.

What should be measurably different when this new workflow exists? Then map the moving parts: whose work changes, which decisions change, how responsibilities shift, what data moves, which systems connect, and what downstream dependencies are affected.

I think of it like opening up a wall during a renovation. It may look like one wall, but before knocking it down, you need to know whether it is carrying the weight of something else.

Enterprise workflows behave the same way. A seemingly small technology change can alter ownership, data flows, controls, responsibilities, and decisions several steps downstream.

That is why systems thinking matters. Systems thinking means seeing the connections before they become consequences.

6. Governance and risk functions increasingly have access to AI tools that can analyze large volumes of information and automate parts of decision-making. Where do you see the greatest opportunities for AI in these functions, and where is human judgment still especially important?

AI can metabolize information at a scale humans simply cannot.  In governance and risk, that computing power is enormously valuable. AI can synthesize massive volumes of information, surface patterns and anomalies, monitor changes, reduce repetitive analysis, and help organizations move from reactive reporting toward more informed planning and decision-making.

But information is not transformation. Seeing the signal and deciding what the signal means are different jobs. Someone still needs enough business context and subject-matter expertise to ask the right questions, interpret what comes back, understand the consequences, and make a decision.

AI can give us something close to superhuman visibility across information. The opportunity available is to give good thinkers dramatically more computing power.

7. How does the introduction of AI change the role of implementation professionals and solutions architects? Are skills such as requirements gathering, systems thinking, and stakeholder alignment becoming more important rather than less?

 AI can replace a tremendous amount of the work. It cannot replace the responsibility for thinking. For an implementation professional or solutions architect, AI can feel like gaining a brilliant assistant who never sleeps. It can process documents, organize discovery notes, summarize workshops, analyze data, compare tools, draft requirements, and compress hours of manual work into minutes. 

That’s extraordinary leverage. But it also creates a risk: the more capable the assistant becomes, the easier it is for the architect to become intellectually lazy. 

AI should free experienced professionals from mechanical work so they can spend more time on the work that requires judgment: asking better questions, seeing connections, challenging assumptions, aligning people, and making sense of complexity 

8. As companies race to adopt AI, what separates organizations that will successfully integrate it into their operations from those that will end up with expensive tools layered on top of unresolved business problems?

The organizations that succeed with AI will understand what they are trying to improve before they introduce the technology.

Think about putting premium fuel into a car. If the engine is healthy and everything is working together, better fuel may improve performance. But if the transmission is failing, the wheels are out of alignment, and warning lights are flashing, premium fuel isn’t the answer. AI works much the same way.

The strongest organizations will address those fundamentals first, define the outcome they want, and then use AI to help them get there faster.