What is employee-driven digital innovation with generative AI?

Key Takeaways

• The problem is not a shortage of ideas. Organizations have spent decades collecting ideas from their employees. What almost never happens is that those ideas end up as digital solutions that run, get maintained and produce measurable value.
• The distance has a name in the literature: the idea-to-implementation gap. The person who lives the problem proposes, and somebody else builds. Most projects are lost in that handoff.
• What generative AI changed is that the same employee who knows the problem can produce the specification, the prototype and the first working version without being a professional developer.
• The tool on its own is not enough. Without an involved decision-maker, data rules and a baseline metric to compare against, the result is fragile prototypes that nobody maintains.
• We call that configuration GenAI-enabled employee-driven digital innovation. It is a construct we are putting forward, not an established academic category.

Almost every company we know has a suggestion box, an innovation committee or a channel where people report what is not working. Very few have a way to turn that into software that runs the following Monday. The idea enters the system and stops there, waiting for budget, a vendor or priority.

That bottleneck is what this article is about.

What is employee-driven digital innovation?

It is innovation in digital products, services or processes that originates with employees outside the formal innovation, research or software development functions. Academic work calls it employee-driven digital innovation (EDDI), and the term was consolidated by Opland, Pappas, Engesmo and Jaccheri in 2022 after reviewing 58 studies at the intersection of employee-driven innovation and digital innovation.

The premise goes back further. Since 2010, research on employee-driven innovation has held that ordinary employees carry tacit knowledge of routines, exceptions and real operating costs that rarely reaches the formal decision-making centers. Eric von Hippel had shown something similar from another angle: users with intense needs tend to be better sources of design than the people manufacturing for them, especially when they have tools that let them iterate on their own.

The critical point, and the one most often misread, is that this is not about collecting suggestions. If the employee drops off an idea and disappears from the process, there is no employee-driven innovation. There is a suggestion box.

Why do employee ideas almost never get implemented?

Because proposing and building are separated by a chain of translations, and every link loses information. The employee describes the problem, somebody interprets it, a third person specifies it, a vendor builds it, and at the end the technology function decides whether it can be integrated.

Leible and colleagues studied exactly this gap in 2026 and point out that prior research concentrated on ideation, while the realization of employee-generated digital ideas remains underexplored. Their warning is useful for anyone selling enthusiasm: purely self-service approaches hit a ceiling in environments with complex systems.

In practice, the gap is recognizable by the phrases that repeat inside companies. The prototype looked good but nobody uses it. The agency would have to do the next version. We don't know which data came out. Nobody ended up responsible. The team went back to the old process.

What does generative AI change?

It changes who can build. Unlike traditional automation or predictive analytics, generative AI turns natural language, examples, documents and rules into digital artifacts: specifications, workflows, interfaces, code, documentation and tests.

That does not turn anyone into a software engineer, and claiming otherwise is selling smoke. What it enables is an intermediate form of building. Someone who knows their company's quoting process better than anyone can translate that knowledge into a first working version without waiting for development budget to be released.

Von Hippel used the word toolkits for tools that shift part of the design work to the user. Generative AI is a new generation of toolkit, with one important difference from visual low-code platforms: it does not force you to think inside the logic of a specific environment, it works through intent, generation, evaluation and correction.

The productivity evidence points the same way. Brynjolfsson, Li and Raymond found significant improvements in customer support work, concentrated among the least experienced workers. Generative AI appears to spread good practice downward, not just make the already-fast faster.

What is GenAI-enabled employee-driven digital innovation?

It is the organized process through which employees outside software development take part in specifying, building, testing and implementing digital solutions for problems they know first-hand, using generative AI as the building tool and under explicit rules of governance, accountability and measurement.

It is a construct we are putting forward from our own practice, not a label recognized in the literature. We say it that plainly because this field is full of people who invent a category and present it as though ten years of research sat behind it.

The configuration requires six conditions that hold each other up:

• A situated problem. Not a technology and not a generic idea, but something frequent, costly and observable that a team lives with every day.
• The problem owner inside. The person who knows the routine, the exceptions and the operating cost takes part in the building, not just in the opening interview.
• Someone who will sustain the solution. A person with knowledge of the company's systems and data, responsible for keeping it running afterwards.
• A decision-maker involved early. Whoever can approve the change, unblock access and prioritize time. When they arrive at the end, the prototype suffocates politically.
• Data rules before building. What information can be used, with which tools, who approves and what must not be automated.
• A baseline. A starting number to compare against. Without one there is no way to tell a real improvement from an entertaining demo.

Remove one and the result degrades predictably. Without a decision-maker, orphan prototypes. Without data rules, legal exposure. Without a baseline, enthusiasm that burns out in three weeks.

How is it different from an open innovation challenge or outsourced development?

Each model solves something different and fails differently. The honest comparison is this:

Model	Logic	Strength	Typical failure
Open innovation challenge	The company publishes a challenge and outside players propose solutions	Access to outside talent and ideas	Context asymmetry, intellectual property negotiation, prototypes nobody adopts
Outsourced development	A vendor interprets the problem and builds it	Professional technical capacity and engineering control	Dependency, recurring cost, long translation cycles
AI strategy consulting	Experts diagnose and hand over a roadmap	Executive perspective and prioritization	The document never becomes software or operational change
Citizen development without governance	Business users build with whatever is at hand	Speed and closeness to the problem	Unaudited systems, technical debt, weak security
GenAI-enabled EDDI	The internal team knows the problem and builds with generative AI, with method and support	Less translation, more ownership, learning that stays inside	Requires picking the problem well and reviewing the quality of what gets built

None of them cancels the others. Professional development is still necessary for critical systems, and open innovation works when there is enough context and clear agreements. What changes is the starting point: instead of outsourcing the first attempt, the company makes it in-house and decides afterwards what to scale.

What are the real risks?

The same ones that come with any democratization of digital building, amplified by speed.

The most common is overconfidence. Seeing something work on a screen does not mean having a production-ready solution, and generative AI produces code that runs but can be fragile. The second is maintenance: a useful tool that nobody sustains becomes debt within months. The third is security, because people upload customer information to public tools long before anyone defines a policy.

There is also a risk of internal inequality. Research on AI at work shows heterogeneous effects depending on experience and thinking habits, so not everyone benefits equally from the same training.

None of these risks invalidates the model. They define its design conditions, which is why data governance is part of the method rather than an annex.

How do you know if it worked?

Through a comparison against the baseline, not through a demo. The indicators that matter are operational rather than emotional: time saved on the task that was attacked, fewer errors or rework, real adoption measured in active users, an assigned owner, and whether the solution survives at ten and at twenty-four weeks.

There is one more indicator that almost nobody measures and that matters more to us than the first. It is the cost of the second solution. If building the next case costs the company less time, less debate and less outside help than the first one, the capability stayed in the building. If every new problem requires repeating the same support, what happened was a service, not a transfer.

Frequently asked questions

Is this the same as citizen development?

No. Classic citizen development rests on company-approved low-code platforms and predefined visual interfaces. The difference lies in the tool and in the scope: generative AI works through intent in natural language and takes part in the specification, not only in the assembly. The warning attached to citizen development does apply equally, and it is the need for governance to avoid systems nobody audits.

Do I need a technology department to try it?

Not to start, yes to sustain it. We have worked with companies that do not have a single developer and that built their first solution. What they do need is one internal person with knowledge of their systems and data who takes ownership. Generative AI changes the point at which technical capacity has to step in, it does not remove the need for it.

How long does it take a company to become autonomous?

Longer than the market promises. The first solution can be built in weeks. Real autonomy, meaning solving a new problem without support, shows up around the third or fourth solution. Promising full independence after the first exercise is the fastest way to lose a team's trust.

What happens with company data?

That is the question to answer before building, not after. Every organization needs to define what information can go into a generative tool, what never leaves, who approves an automated execution and where data can be sent. It takes an afternoon and it prevents the most expensive problem of all.

Who does GenAI-enabled employee-driven digital innovation?

We do. At Suricata Labs we designed and run the method that gives this framework its shape, and we call it BuildInside.

The framework did not come out of a literature review. It came out of running programs with real companies in Latin America, finding the same pattern over and over, and realizing that the research already had almost every piece except the one generative AI contributed. The academic review came afterwards, and that is where we confirmed this specific configuration had not been named.

BuildInside has now run several cohorts in Colombia, El Salvador and Mexico, with companies from sectors that have little in common and that rarely have a technology department. The result we are after in each one is the same: that the company ends up with a solution built by its own people, and with the capacity to build the next one.

The most recent example is Ciudad Juárez. We ran the applied artificial intelligence component of Innova Lab V3.0, the program operated by Democratizamos la Innovación, and in three days eighteen companies from Juárez walked out with a working prototype, among them cement producers, maquilas, metalworking shops and retailers.

If you run a digital transformation program, or a company that has already tried the other routes, let's talk. You can also see how we work on AI strategy.