Built an agentic AI system that made an expert-dependent consultancy method scalable, and created a standalone product that opened a new revenue stream alongside their consulting.
Built an agentic AI system that made an expert-dependent consultancy method scalable, and created a standalone product that opened a new revenue stream alongside their consulting.
Built an agentic AI system that made an expert-dependent consultancy method scalable, and created a standalone product that opened a new revenue stream alongside their consulting.
CHALLENGE
Scaling a method built on judgment no one had written down
Skill Factor's consultancy method helped organizations map their values and resolve the communication tensions that build up across generations and roles. It worked, but only when a specific expert ran it: the judgment behind it, the criteria they used to read a company and decide what mattered, lived in their head and was never written down. Generic AI couldn't stand in for it, because without that interpretive criteria the output was inconsistent and shallow. The hard part was capturing how the expert reasoned well enough that a system could reason the same way.
CHALLENGE
Scaling a method built on judgment no one had written down
Skill Factor's consultancy method helped organizations map their values and resolve the communication tensions that build up across generations and roles. It worked, but only when a specific expert ran it: the judgment behind it, the criteria they used to read a company and decide what mattered, lived in their head and was never written down. Generic AI couldn't stand in for it, because without that interpretive criteria the output was inconsistent and shallow. The hard part was capturing how the expert reasoned well enough that a system could reason the same way.
CHALLENGE
Scaling a method built on judgment no one had written down
Skill Factor's consultancy method helped organizations map their values and resolve the communication tensions that build up across generations and roles. It worked, but only when a specific expert ran it: the judgment behind it, the criteria they used to read a company and decide what mattered, lived in their head and was never written down. Generic AI couldn't stand in for it, because without that interpretive criteria the output was inconsistent and shallow. The hard part was capturing how the expert reasoned well enough that a system could reason the same way.
SOLUTION AND IMPACT
Expert judgment captured and turned into a product
The system captures how the expert reads a company and runs that judgment across four specialized agents, so the method works without depending on the person who created it. It was validated against expert judgment through repeated testing, holding the depth that made the original method work. One of the four agents became a standalone product Skill Factor can sell on its own, opening a new revenue stream alongside their consulting. A first use case with a major company is now in progress.
SOLUTION AND IMPACT
Expert judgment captured and turned into a product
The system captures how the expert reads a company and runs that judgment across four specialized agents, so the method works without depending on the person who created it. It was validated against expert judgment through repeated testing, holding the depth that made the original method work. One of the four agents became a standalone product Skill Factor can sell on its own, opening a new revenue stream alongside their consulting. A first use case with a major company is now in progress.
SOLUTION AND IMPACT
Expert judgment captured and turned into a product
The system captures how the expert reads a company and runs that judgment across four specialized agents, so the method works without depending on the person who created it. It was validated against expert judgment through repeated testing, holding the depth that made the original method work. One of the four agents became a standalone product Skill Factor can sell on its own, opening a new revenue stream alongside their consulting. A first use case with a major company is now in progress.
DESIGN DECISION: FORMALIZING EXPERT KNOWLEDGE
How do you get an expert's reasoning out of their head and into a system?
In discovery workshops I mapped the method step by step, tracing the inputs and outputs of each activity to see where judgment was actually being applied. From there I worked out which activities an agent could take on, and what each one needed to know to do it well. I gathered everything written down, past case studies and existing material, and worked in constant alignment with the experts to fill the gaps their documents didn't cover. The result was the expert's knowledge restructured into instructions, criteria, and value ontologies an agent could read and act on.
DESIGN DECISION: FORMALIZING EXPERT KNOWLEDGE
How do you get an expert's reasoning out of their head and into a system?
In discovery workshops I mapped the method step by step, tracing the inputs and outputs of each activity to see where judgment was actually being applied. From there I worked out which activities an agent could take on, and what each one needed to know to do it well. I gathered everything written down, past case studies and existing material, and worked in constant alignment with the experts to fill the gaps their documents didn't cover. The result was the expert's knowledge restructured into instructions, criteria, and value ontologies an agent could read and act on.
DESIGN DECISION: FORMALIZING EXPERT KNOWLEDGE
How do you get an expert's reasoning out of their head and into a system?
In discovery workshops I mapped the method step by step, tracing the inputs and outputs of each activity to see where judgment was actually being applied. From there I worked out which activities an agent could take on, and what each one needed to know to do it well. I gathered everything written down, past case studies and existing material, and worked in constant alignment with the experts to fill the gaps their documents didn't cover. The result was the expert's knowledge restructured into instructions, criteria, and value ontologies an agent could read and act on.
DESIGN DECISION: AI AS DESIGN MATERIAL
How do you design a system that doesn't behave predictably?
Agents don't work like normal software: the same input won't always return the same output, so the usual approach of designing fixed logic falls apart. Instead of trying to control every result, I designed the system to catch problems: input validation at the start, defined checks at each step and a final review that holds the output to specific constraints. Each agent runs as a pipeline of smaller, checkable steps rather than one opaque action, so a wrong turn is visible and correctable. Human review stays in the loop at key points, so the system stays accountable rather than running unchecked.
DESIGN DECISION: AI AS DESIGN MATERIAL
How do you design a system that doesn't behave predictably?
Agents don't work like normal software: the same input won't always return the same output, so the usual approach of designing fixed logic falls apart. Instead of trying to control every result, I designed the system to catch problems: input validation at the start, defined checks at each step and a final review that holds the output to specific constraints. Each agent runs as a pipeline of smaller, checkable steps rather than one opaque action, so a wrong turn is visible and correctable. Human review stays in the loop at key points, so the system stays accountable rather than running unchecked.
DESIGN DECISION: AI AS DESIGN MATERIAL
How do you design a system that doesn't behave predictably?
Agents don't work like normal software: the same input won't always return the same output, so the usual approach of designing fixed logic falls apart. Instead of trying to control every result, I designed the system to catch problems: input validation at the start, defined checks at each step and a final review that holds the output to specific constraints. Each agent runs as a pipeline of smaller, checkable steps rather than one opaque action, so a wrong turn is visible and correctable. Human review stays in the loop at key points, so the system stays accountable rather than running unchecked.
DESIGN DECISION: A NEW REVENUE STREAM
A standalone product built from one piece of the system
The first three agents analyze and diagnose, so they need a consultant's interpretation around them to be useful. The fourth was different: it produces finished, usable output on its own, adapting a company's text to its values and target audience without anyone in the loop. Working through the design with the client, we saw this could stand alone, and built it to be customizable so each company could adapt it to its own context. That made it a sellable product, giving Skill Factor something to offer clients directly and a new revenue stream alongside their consulting.
DESIGN DECISION: A NEW REVENUE STREAM
A standalone product built from one piece of the system
The first three agents analyze and diagnose, so they need a consultant's interpretation around them to be useful. The fourth was different: it produces finished, usable output on its own, adapting a company's text to its values and target audience without anyone in the loop. Working through the design with the client, we saw this could stand alone, and built it to be customizable so each company could adapt it to its own context. That made it a sellable product, giving Skill Factor something to offer clients directly and a new revenue stream alongside their consulting.
DESIGN DECISION: A NEW REVENUE STREAM
A standalone product built from one piece of the system
The first three agents analyze and diagnose, so they need a consultant's interpretation around them to be useful. The fourth was different: it produces finished, usable output on its own, adapting a company's text to its values and target audience without anyone in the loop. Working through the design with the client, we saw this could stand alone, and built it to be customizable so each company could adapt it to its own context. That made it a sellable product, giving Skill Factor something to offer clients directly and a new revenue stream alongside their consulting.
LEARNINGS
AI removes the ceiling on iteration, and shifts the work to judgment
Building this system changed how I think about AI in a product. It behaves less like software and more like a material: you shape it, test it and adjust, but you never fully control it, so the work is in setting the right constraints rather than fixing every output. What struck me most is how much it opens up iteration. Design has always wanted to try more options and refine further, and has always been limited by time and budget. Working with AI removes a lot of that ceiling, which means the constraint shifts from how much you can produce to how well you can judge what is worth keeping.
LEARNINGS
AI removes the ceiling on iteration, and shifts the work to judgment
Building this system changed how I think about AI in a product. It behaves less like software and more like a material: you shape it, test it and adjust, but you never fully control it, so the work is in setting the right constraints rather than fixing every output. What struck me most is how much it opens up iteration. Design has always wanted to try more options and refine further, and has always been limited by time and budget. Working with AI removes a lot of that ceiling, which means the constraint shifts from how much you can produce to how well you can judge what is worth keeping.
LEARNINGS
AI removes the ceiling on iteration, and shifts the work to judgment
Building this system changed how I think about AI in a product. It behaves less like software and more like a material: you shape it, test it and adjust, but you never fully control it, so the work is in setting the right constraints rather than fixing every output. What struck me most is how much it opens up iteration. Design has always wanted to try more options and refine further, and has always been limited by time and budget. Working with AI removes a lot of that ceiling, which means the constraint shifts from how much you can produce to how well you can judge what is worth keeping.
Next up
Next up
Next up
