The Best AI Strategy May Begin With a Stopwatch

Artificial intelligence is often sold through possibility: a model can write, summarize, analyze, design, code, search and converse. For a business owner, that list is impressive and almost useless.

The operating question is narrower: Which part of the company should change first?

James Galang’s answer is to look for time. His Aldie-based firm, Galang AI, embeds with organizations to study how work moves, where people repeat themselves and which manual steps prevent leaders from focusing on decisions that matter. He calls himself a forward-deployed operator rather than a forward-deployed engineer because the work begins with people and process.

That orientation is timely. Census Bureau data from late 2025 and early 2026 put overall business AI use between 17 and 20 percent, with 37 percent of firms employing at least 250 people reporting use. (U.S. Census Bureau) Adoption is no longer rare, but it is far from universal. Many companies are caught between casual employee experimentation and a governed system that creates measurable value.

Galang’s own route into the field was practical. He studied biology, entered technology through an unexpected personal connection and built a career across operations, sales and enterprise systems. AI became a way to optimize his work and reclaim time with his family. A community communications experiment with his wife demonstrated that the tools could improve engagement. Work for a biotechnology company showed that they could automate more consequential workflows. Galang AI emerged from demand rather than a prebuilt product searching for customers.

Find labor, not novelty

A good discovery process begins by identifying repeated labor. For one week, ask employees to record tasks that are:

• performed at least weekly;
• governed by recognizable rules;
• built from digital inputs;
• easy for a qualified person to review; and
• frustrating because of volume rather than complexity.

Meeting follow-up is a common example. The raw material already exists in the conversation. A system can create a transcript, identify decisions, assign action items and draft a summary. Human participants still confirm the commitments. The machine removes clerical delay.

Galang described a more connected workflow for a medical-media client: source material could be transformed into a citation-supported article, published to a media center and repurposed for social channels. The business value was not “AI content.” It was a reduction in the number of disconnected handoffs required to complete a known publishing process.

Automate the stable process first

Owners often try to automate work that is still poorly defined. That produces faster inconsistency. Before involving AI, write the current process in five parts: trigger, inputs, steps, decision points and completed output. If experienced employees disagree about those elements, the immediate need is process design.

Once the process is stable, classify each step:

• Automate: predictable transformation with clear rules.
• Assist: judgment remains human, but AI can research, summarize or draft.
• Protect: sensitive or high-consequence work requires strict access and review.
• Leave alone: the human interaction is itself the value.

Galang notes that clients have different appetites. One team may want to keep creative direction fully human while automating formatting and posting. Another may be comfortable generating a draft as long as a named expert approves it. The system should reflect the organization’s risk tolerance rather than the vendor’s enthusiasm.

Build governance into the workflow

The Government Accountability Office has noted that generative AI may improve productivity while also spreading misinformation and introducing security and environmental risks. (GAO) NIST’s voluntary framework offers a compact way to think about those risks through governance, mapping, measurement and management. (NIST)

For a small business, governance does not need to become a binder nobody reads. It can begin with a one-page automation record:

1. Business purpose and owner.
2. Systems and data the automation can access.
3. Model or vendor used.
4. Output that requires human approval.
5. Accuracy, privacy and security risks.
6. Failure procedure and manual fallback.
7. Metric and review date.

That record makes it possible to answer basic questions later, especially when an employee leaves, a vendor changes terms or a model begins producing different output.

Measure capacity returned, not content produced

AI projects are easily celebrated through volume: more posts, more drafts, more reports. Galang’s emphasis on time suggests a better scorecard.

Measure hours saved, turnaround time, error rate, rework, adoption and the higher-value work that replaced the manual task. If an automation saves ten hours but creates eight hours of review and correction, it is not leverage. If it returns four hours to a founder who then uses them on sales calls that generate qualified opportunities, its impact is much larger than the labor saving alone.

This is how a company can “grow without growing overhead,” in Galang’s words. The claim should not be interpreted as a promise that AI eliminates the need for people. It means capacity can increase before headcount does, especially in administrative processes that expand with every new customer.

Design for zero friction, then earn deeper adoption

Many technology projects fail because the organization must change too much before receiving value. Galang AI tries to integrate with familiar tools or replace a complete process so the user does not have to learn an entirely new platform.

That is a sound adoption strategy. Start with one visible win that requires minimal behavior change. Let the team experience the result. Then use the trust and evidence from that pilot to examine a more important workflow.

The founder’s first question should therefore not be “What can AI do?” It should be “Where are capable people spending time without applying much judgment?” Put a stopwatch on that work. Map it. Protect the risky parts. Automate one stable segment and measure what comes back.

The technology is broad. A valuable implementation is specific.

That specificity also makes procurement easier. Before hiring an AI vendor, ask for one named workflow, a baseline cost, the data that will be touched, an accountable reviewer and a 30-day test. A proposal that cannot define those basics is still a technology demonstration, not an operating plan.

Interview source: James_Galang_1(2).txt