“We want to identify risks earlier.”
For a mining company managing multiple tailings facilities, this is a natural business objective: regularly acquire satellite imagery, detect changes in the facilities and their surrounding environments, assess potential risk signals, and issue alerts when anomalies may require further attention.
From a technical perspective, this may not seem particularly difficult.
Optical satellites can provide detailed visual and multispectral information about the Earth’s surface. Synthetic aperture radar (SAR) can acquire observations at night and through cloud cover. Historical Earth observation data can form time series, while artificial intelligence can help identify change across large volumes of imagery.
But once the technology enters the actual operating environment, the problem becomes far less straightforward.
Because detecting change is not the same as identifying risk.
A change on the ground does not necessarily require the engineering team to intervene immediately. An area flagged as anomalous by a model may not be the place most deserving of a field investigation today.
Nor is an investigation decision usually based on satellite imagery alone. Engineers may also need to examine ground-monitoring data, recent weather, historical inspection records, and facility operating conditions. They then combine this information with professional judgement to decide whether a signal warrants further attention.
That information may be distributed across different systems and stored in different formats. Some decision criteria may never have been formally documented. Other results may already be generated by a system but never reach the workflow engineers actually use each day.
The real problem the organization needs to solve may therefore be:
With assets spread across many locations and limited engineering resources, how can the organization identify the places that genuinely require attention earlier—and deliver the right information to the right decision-makers at the right time?
This is the “last mile” Earth observation so often encounters. It is also where STARPATH GLOBAL’s Forward Deployed Engineers (FDEs) work: entering that gap and making the entire path function in practice.
Enter the Real Workflow, Not Just a Requirements List
FDE stands for Forward Deployed Engineer. An FDE works directly with customer teams to understand an operational problem, build and deploy a working solution, and turn field learning into reusable product capabilities.
At STARPATH GLOBAL, FDEs work at the intersection of satellite-based geospatial intelligence, AI, product technology, and the customer’s business. They understand both what Earth observation can provide and where its limits lie. They also understand the business outcome the customer is trying to improve. Drawing on capabilities across Earth observation, AI, product development, and software engineering, STARPATH GLOBAL’s FDE team connects data and models to operational workflows—and takes responsibility for turning technical capability into measurable business value.
An FDE does not simply relay requirements between the customer and the product team, nor do they develop features from a predefined checklist.
Their first task is to uncover how the work is actually done.
Consider tailings-facility monitoring. A customer may say:
“We want to use satellites to detect risk automatically.”
But two STARPATH GLOBAL FDEs, Du and An, would not immediately begin comparing satellite resolutions. Nor would they break the request straight away into a feature list such as “imagery, change detection, risk scoring, and dashboard.” Instead, they would follow the entire operating chain alongside the engineers responsible for day-to-day monitoring: where a signal originates, which systems it passes through, who reviews it first, what additional information the engineers need, and what ultimately determines whether an investigation is initiated.
Only then do the differences between the formal process and actual practice become visible. In a meeting room, the official workflow may be described as:
Monitoring → Anomaly → Engineering review → Risk assessment → Field investigation
In practice, however, engineers may have to match some information manually to inpidual facilities. Historical records may not be reliably linked to current observations. Different data sources may use different identifiers. Some crucial decisions may rely on experience that has never been incorporated into a system. An alert may even be generated correctly yet still be overlooked because it appears in an interface people rarely use.
These are not peripheral details.
They are where the most important information lies.
FDEs focus on precisely these seams between people, data, systems, and decision-making. They determine which part of the process is most worth redesigning first, and what minimum change could improve the performance of the entire operating chain.
As they follow the real workflow, the original request becomes more specific:
The customer may not actually need to detect every change. What they need is to identify—earlier and with limited inspection resources—which locations should be investigated first.
Do Not Wait for Every Question to Be Fully Defined—Prove the Shortest Path First
Much of the knowledge embedded in complex operations is difficult to capture completely through interviews or process documents.
Engineers can explain their general decision principles, but it is often only when they see actual outputs that they can articulate why a particular type of change requires no action, what information would raise its priority, and what further evidence they would need before taking the next step.
That is why FDEs do not wait until every dataset, interface, and rule is in an ideal state before they begin building.
Working around the question “Which facilities should be investigated first?”, Du and An would begin by proving the smallest viable path to an answer.
They might first link the facility information, historical records, and field data the customer can already provide with Earth observation data suited to the use case. Satellite observations provide signals of change. Historical records and environmental conditions supply context. AI and business rules support the initial screening, while engineers review uncertain or potentially consequential results.
The first version does not have to be a complete platform.
It may consist of a data-processing script, a temporary adapter connecting existing records, an initial prioritization logic, and a simple interface through which engineers can inspect the evidence and provide feedback directly.
If standard interfaces are not yet available, an FDE does not allow the entire value-validation process to stall. Subject to data-security and access requirements, the team can begin with data the customer is already able to provide and use lightweight adapters to validate the critical path. Once the value and the approach have been confirmed, the temporary engineering can be progressively hardened for production.
Its purpose is to answer one essential question as quickly as possible:
Can EO data and AI analysis help engineers decide more effectively where to look first?
One important difference between the FDE model and traditional consulting is what happens after the business problem has been understood: the FDE remains directly involved in building and validating the first working solution.
Nor would an FDE blindly search for “the best satellite.” More data is not automatically better, and clearer imagery is not automatically more valuable.
If the task is to screen a large number of facilities on a recurring basis, coverage, revisit frequency, historical continuity, and data reliability may matter more than achieving the highest possible resolution in a single image. Once a facility moves into a deeper investigation, higher-resolution imagery or additional field information may then be brought in.
Data selection is therefore not a one-off product recommendation. It is part of evidence design:
What is the minimum reliable evidence an engineer needs to make this decision? What information is still missing from the current business judgement? Which type of data can fill that gap most effectively?
The answers determine which satellites are required, what timescale should be used, which analytical methods are appropriate, and what information should ultimately enter the workflow.
Let Real Use Turn Tacit Expertise into Solvable Problems
When the first risk-prioritization outputs are placed in front of engineers, judgements that were previously difficult to articulate begin to become concrete.
The system may already detect a set of changes consistently. Yet engineers may point out that some are merely normal seasonal patterns, while others—although real—result from known construction or operating activity and do not require immediate investigation.
Other signals may appear subtle in the imagery but become far more important when considered alongside previous anomalies, recent environmental conditions, and the risk profile of the facility itself.
Real outputs make previously tacit judgement visible, allowing it to be examined, challenged, and—where appropriate—formalized. Discovery does not end when building begins; it deepens through building and use.
Du and An would not simply record this feedback and hand it to another development team. They would review the results with the engineers: Why would this item not trigger an investigation? What evidence is missing? What would change its priority? Which judgements can be expressed reliably as rules, and where must human involvement remain?
As they develop a deeper understanding of these business decisions, they also modify the system directly.
A change detected by satellite no longer automatically becomes a risk alert. Instead, it becomes one element in a broader body of evidence. Historical trends, environmental information, facility characteristics, known operating activity, and previous records are progressively incorporated into the prioritization logic.
AI can help extract, connect, and interpret this information, but it does not independently replace professional judgement in every decision. Deterministic business rules define system boundaries, while human review addresses cases in which evidence is incomplete or uncertainty remains high.
The system initially answers:
“Where has something changed?”
Real-world feedback pushes it toward a more useful question:
“Which location should be investigated first—and why?”
This is the progression from detection, to prioritization, to decision support.
Deliver the Result Where the Work Actually Happens
Even as the prioritization results begin to align more closely with engineering judgement, the project may still fail at the last mile if the results remain isolated in a separate platform while the engineering team spends each day working in GIS, an asset-management system, email, or another collaboration tool.
The FDE therefore needs to embed the results into the existing way of working. Where do engineers begin their day? How do they receive information? Who confirms an anomaly? How is an investigation task created? How do conclusions from the field return to the system?
The new capability is then connected to this existing chain. This might mean linking geospatial analysis to asset records so engineers can examine the evidence through their established workflow. Or it may mean pushing high-priority items into tools they already use every day, rather than asking them to remember to open another system.
When the system flags a signal for attention, users should see more than a location and a priority score. The result should also show the supporting evidence, relevant data-quality information, the degree of uncertainty, and the recommended next action.
In the resulting workflow, high-priority information reaches the right people, results with insufficient evidence enter human review, and findings from field investigations are fed back into the system.
A closed loop begins to form:
Observe → Screen → Prioritize → Validate → Investigate → Feedback
From recurring observation, through initial screening and prioritization, to expert validation, follow-up investigation, and feedback, every stage is connected to a real operational action.
At this point, geospatial analysis is no longer an isolated output. It has become part of an operational decision process. That is the last mile of Earth observation.
Once the System Enters Production, the Real World Keeps Asking Questions
Once a system enters everyday use, the real world exposes boundaries that are difficult to see in a demonstration environment.
A data source may become temporarily unavailable. Some observations may lack the quality required to support a decision. Certain normal operating activities may repeatedly trigger anomalies. The same rules may perform differently across regions. Facility operating conditions evolve, and new business situations may fall outside the original samples and rules.
At this point, an FDE’s job is not simply to keep “tuning the model.”
Du and An would determine whether a problem lies in the data, algorithm, rules, interaction design, or workflow, and then adjust the relevant layer directly. They might intercept poor-quality data before analysis, use confidence and evidence completeness to determine whether a priority should be assigned automatically, introduce a degraded mode when data is missing, and incorporate new edge cases into continuous evaluation.
Sometimes the system’s most important improvement is not a more precise answer, but the ability to tell the user clearly:
“There is not enough evidence to make a determination. Human review is required.”
A trustworthy system does not always produce an answer. It knows when it should not pretend to be certain.
This is also why FDEs must remain involved as the solution enters real use. Only when the system begins to bear the complexity of everyday operations do its boundaries, risks, and genuine product requirements fully emerge.
FDE Success Is Not Measured by Model Metrics Alone—It Must Produce Outcomes
Model metrics still matter, but they are not the final answer. Evaluation must continue to ask:
- Is the engineering team identifying investigation-worthy assets earlier?
- Has unproductive screening been reduced?
- Are limited field resources being allocated more effectively?
- Has the time between signal detection and human review been shortened?
- Are users genuinely incorporating the new outputs into their daily work?
If the system can narrow a large screening pool into a smaller set of higher-priority items—and engineers continue to use it—then proof of value begins to take shape.
Technology starts becoming a business capability only when it enters real work and changes operating behaviour.
That is what it means for an FDE to be accountable for outcomes.
A Successful FDE Eventually Steps Back
Once a solution has been validated and begins to stabilize, some of the components initially maintained by FDEs by hand must be re-engineered.
Temporary data adapters created for rapid learning should become reliable integrations. Where appropriate, repeatable business judgements should be converted into transparent, testable rules or evaluation cases. Repetitive manual operations should be addressed through automation and product capabilities that reduce operating costs. Tasks requiring repeated FDE intervention should, wherever possible, be handed over to the product, automated workflows, or the customer’s own team.
As data pipelines, business rules, evaluation mechanisms, monitoring, and exception handling mature, day-to-day operation no longer depends on Du and An being continuously present. The FDEs can reduce their involvement and turn their attention to new, higher-value problems.
This is an important distinction between FDE work and long-term on-site staffing or outsourced labour.
The goal is not to sustain the service by continually adding more people. Through engineering and productization, FDEs progressively reduce dependence on particular inpiduals.
An FDE’s “exit,” therefore, is not simply project completion followed by engineer departure. It means the solution no longer requires someone to remain permanently in the middle for it to function.
The FDE can genuinely step back only when the customer can continue using the solution, the system meets the agreed reliability criteria, and known issues have clear owners and resolution paths.
That is a true handoff.
Turn One Solution into the Starting Point for the Next
Solving a problem in the field is not the end of an FDE’s work.
When the same patterns of data processing, evidence aggregation, human review, confidence management, or prioritization recur across more business scenarios, they are no longer merely special requirements for one customer.
FDEs bring these recurring solutions back to STARPATH GLOBAL’s product and engineering teams.
Temporary capabilities validated through real use can be developed into data connectors, reference architectures, evaluation templates, industry rules, or standard workflows. The next time an FDE encounters a similar problem, they do not need to rebuild the same foundations. They can start from a more advanced position.
An FDE may first travel an unproven gravel road with the customer. When that route repeatedly demonstrates value, it should gradually be paved into a road more customers can use.
This creates a complete value loop:
Enter the real workflow → Identify the critical breakpoints → Build rapidly → Put the solution into daily use → Resolve edge cases → Validate value → Productize for reuse
One engagement solves one specific problem.
A mature FDE model ensures that every problem solved strengthens the company’s ability to solve the next one.
STARPATH GLOBAL: Taking Geospatial Intelligence All the Way Through the Last Mile
Satellite data is becoming increasingly accessible to businesses. The real challenge is making that data operational.
For STARPATH GLOBAL, this means closing two gaps at once: the gap between satellite observation and reliable evidence, and the gap between evidence and operational action.
STARPATH GLOBAL’s FDEs do not begin with a predetermined satellite or a fixed product. They begin with the business outcome the customer wants to improve. Our engineers enter the real workflow, identify the problem most worth solving, and then organize EO data, AI, the customer’s existing information, and engineering capabilities into a solution that can be built quickly, validated, and continuously improved.
The process can begin with one concrete question:
What decision do you want to improve?
At STARPATH GLOBAL, the path from problem identification to value validation, scale-up, and continuous operation is structured around four stages of our FDE service: Discovery, Deployment, Expansion, and Operation. Each step is built on evidence of value established in the previous one.
Customers do not need to select a satellite, model, or complete technical architecture in advance. A meaningful pilot begins with a clear operational question, access to the relevant domain experts, and enough representative information to evaluate the proposed workflow. Businesses do not always need more data. What they need is someone willing to enter a problem for which no standard answer yet exists—and organize satellite-based geospatial intelligence, AI, and engineering capabilities until they produce a real change in business outcomes.
Not simply closer to the customer. Closer to the problem—and accountable for turning satellite intelligence into operational value.
If your organization is exploring applications of satellite-based geospatial intelligence and AI but is not yet certain where to begin, STARPATH GLOBAL’s Forward Deployed Engineers can work with your team to identify high-value opportunities, establish a pilot, and validate potential business value and ROI before committing to deployment at scale.
For qualifying Pioneer Partners, STARPATH GLOBAL currently provides early-stage support spanning opportunity assessment, solution design, engineering enablement, and value validation. Once the potential value has been demonstrated, both parties can decide whether to proceed to a formal commercial engagement.
Have an operational decision that could benefit from better, more timely evidence? Talk to STARPATH GLOBAL’s FDE team to clarify the target decision, assess whether Earth observation is suitable, and define what a focused pilot would need to prove.
