I am Anne and lead Go-To-Market at Formant. We are building a new AI product to help optimize physical operations. We are now testing it with partners. If you are interested in an AI pilot with us, please send an email to support@formant.io and we will be happy to discuss!

In part 1, we argued that reality is the ultimate evaluator of any AI system. Whether it’s a robot, a sensor network, or a connected machine, performance isn’t judged in a lab — it’s judged on-site with operators who demand results.

The fastest, most reliable way to ground AI in reality is through focused AI deployments. These aren’t demos. They’re proving grounds with clear success metrics. They tell us—fast—whether AI actually improves outcomes for people working with machines in physical space.

Why We Focus on Deployments

There’s an old saying: the devil is in the details. In real world deployments, those details don’t hide for long..

• Did the AI produce an outcome that mattered?

• Did people on-site trust and adopt it willingly?

• Did it save time, cut costs, or reduce risk or move some other KPI?

Deployments answer these questions early, so teams don’t spend years (and millions) scaling an idea that never quite worked outside a slide deck.

What We’ve Learned: Context Is Everything

Physical environments are messy. Knowledge isn’t always written down. Tasks evolve faster than process manuals. Different stakeholders care about different things — uptime, compliance, safety, cost, usability — all at once.

Our biggest learning? Context wins.

AI systems that succeed are the ones that:

• capture and organize human know-how,

• adapt to shifting workflows,

• enhance decision-making at every level — from front-line technicians to operations managers.

We have learned that successful pilots all require that AI understands and captures this context.

This enables:

• smarter task assignment,

• higher efficiency,

• faster onboarding for new employees,

• better troubleshooting when something breaks.

Companies that master this layer succeed. Those that ignore it struggle, no matter how “advanced” their models are.

Why a Limited Scope Deployment Format Works Best

This format lets us iterate and learn quickly, focusing on one clear metric at a time. Instead of chasing ten problems at once, we pick one, validate it, and scale from there. This:

• shortens time-to-value,

• reveals real integration challenges early,

• builds confidence for bigger rollouts later.

We’d rather solve one problem completely than “kind of help” with several.

How We Choose Our Deployments

We use strict criteria to make deployments meaningful:

• Mature business case – clear ROI and strategic importance.

• Low technical hurdles – quick to set up, no endless integration effort.

• Scalable use case – broad market potential once proven.

• Strong customer need – urgency drives engagement.

• Clear success metrics – alignment on what “good” looks like.

This discipline keeps our pilots focused, lean, and impactful.

Our Current Focus: Incident Management in Physical Operations

Today, we’re investing deeply in a use case where machines, sensors, and staff already work together naturally — and where AI can quickly unlock measurable business value:

1. Incident identification – detecting defects, hazards, or anomalies in real time.

2. Incident diagnostics – understand the root cause to help solve the problem in real time.

3. Incident resolution – propose a resolution path and resolve the problem working under clear guidelines

We’ve developed extensive expertise in these workflows and have proven that physical systems, AI, and teams can align to produce real operational impact.