An AI roadmap for business
What is the GenAI Opportunity About?
AI has been used as a decision-making tool for decades, resulting in breakthrough use cases such as spam filtering, fraud detection, and, more recently, self-driving cars and semi-autonomous robots.
Generative AI, which gained mainstream popularity in 2022 with the release of ChatGPT and Stable Diffusion, is a new type of AI that encompasses models capable of mimicking human behavior by creating content (text, images, audio, and video) in response to user prompts.
GenAI is made possible by several concurrent technological advances: novel training algorithms based on the Transformer architecture (2017), the broad availability of powerful chips, and the digital availability of vast human knowledge on the internet. These advances underpin foundation models, which can be further trained, fine-tuned, orchestrated, and prompted to achieve specific objectives.
Using GenAI, businesses can automate an increasing number of human activities, particularly those that do not require substantial interaction with the physical world. Companies can leverage GenAI to deliver better value to their customers at a lower cost. If they don't, they risk being displaced by AI-first competitors.
Most of the business value from GenAI initiatives is likely to come from software engineering and IT, customer operations/customer service, sales and marketing (along the entire customer acquisition and retention funnel), and product development operations (especially product documentation). Additional opportunities can be found in supply chain, procurement, risk management, and compliance.
That said, the technology is evolving rapidly, and many applications are not mature yet. GenAI transformation is a journey rather than a blueprint.
For businesses, especially mid-sized companies with limited resources, this means finding a balance between launching short-term experiments and building long-term internal capabilities that deliver business value.
In this blog post, we discuss a possible roadmap for GenAI adoption and highlight some key challenges to watch out for.
Roadmap overview
The GenAI adoption roadmap of a company consists of 3 steps, which may overlap with each other:
Individual productivity: This step involves employees adopting commercial GenAI apps to complete office tasks more efficiently. The tasks and tools are not company-specific: they can include image and video generation for marketing campaigns, coding, and online research.
Internal process optimization and automation: This step involves using GenAI tools to streamline internal administrative processes. GenAI can be used for data clean-up and transformation, decision support, and automated execution. It is important to always simplify processes before automating them.
Product and process innovation: This last step introduces transformative changes to a company’s products, services, and operating model. It requires rethinking the business from a blank slate, like an AI-first competitor would do. It may involve proprietary data or models leveraging the company’s unique expertise, but this is optional.
Many experiments and projects are typically launched as part of the above steps. Not all of them work out. Sometimes this is because recurring costs are too high, but more often, it is because the value proposition does not drive sufficient user adoption.
The company should learn from every initiative and build long-term knowledge and capabilities that can be leveraged across projects. Such capabilities include: chatbots, AI assistants, AI agents, observability and evaluations, APIs to data and third-party services, cloud infrastructure, and team competencies.
This does not mean, however, that every prototype must be built on top of production capabilities. Projects often begin with an ad hoc stack, subsequently transitioning to a more robust infrastructure as they move into production, as illustrated in the diagram below.
Let us review each step of the roadmap in more detail.
Individual productivity
This first step requires:
Training staff about AI technology, opportunities, and tools.
Trialing many commercial GenAI tools to settle on a core set of app subscriptions that best match the company’s requirements.
The range of tools to be evaluated includes chatbots, meeting assistants, coding assistants, text generators, image and video generators, voice and music generators, web scraping tools, and other tools as needed.
Internal process optimization and automation
This step requires:
Initial brainstorming. As outlined in a previous post, the question "How would we use cheap interns?" can be a good starting point.
Top-down tool prototyping and deployment. The reason why this usually starts top-down is that teams may lack both the bandwidth and the competencies to deliver meaningful projects organically.
Killing projects that are not used every day. Many projects won’t garner sufficient user adoption even if they look good on paper. Better to terminate them and move on.
It often makes sense to prototype internal process automation tools using no-code or low-code platforms, but teams must not shy away from writing code, given that vibe coding assistants make coding easier than ever. Zapier + Cursor + Digital Ocean is often a great choice, but for more options, read this previous post.
Product and process innovation
This step requires:
Use case brainstorming and prioritization
Prototyping. Often, this is done by a small team of AI-savvy employees.
Handover to production teams. For projects to work at scale, they must be absorbed into the relevant product or function.
The diagram below illustrates a possible framework for brainstorming.
Production deployments present numerous challenges that are often underestimated by influencers who post viral use cases on social media platforms.
Challenges include:
Use case selection: When use cases show low adoption, is it a user education problem, or is the business value just not big enough for now? When to kill projects?
Data quality: How much time is it worth spending on upfront data clean-up?
Cost control: How to launch apps at a limited scale before costs and infrastructure requirements start to escalate?
Team buy-in: What's the right balance between top-down push and organic innovation?
External dependencies: What's the right balance between in-house development and reliance on immature third-party service providers?
Long-tail risks: How to manage customer and reputation risks (e.g., hallucinations, tone-deaf outputs, jailbreak, leakage of private information)?
Long-tail risks, in particular, require both an observability and evaluation framework, as well as significant testing (all of which can be assisted by AI), which can add significantly to the project timeline if they aren’t anticipated.
Takeaway messages
GenAI transformation requires a structured approach that balances experimentation with sustainable capability building. Companies should start with individual productivity gains, progress through internal automation, and gradually move toward product innovation. Success depends on learning from failed experiments, building reusable infrastructure, and maintaining realistic expectations about adoption timelines. The journey is iterative—not every prototype will succeed, but each attempt should contribute to organizational AI literacy and technical foundations that enable future breakthroughs.