The AI Impact Framework

Parth Chhaparwal
7 min readMar 3


It took 32 years for computers to evolve from the first-generation vacuum tubes to the fourth-generation VLSI microchips. Although the term fifth-generation computing was coined in 1982, computers took a long time to pass the fourth generation, finally outpacing Moore’s law in the 2010s and making computers incredibly powerful. The tenet of fifth-generation computing is to develop artificial intelligence and make machines intelligent. In the last 50 years, science fiction films have helped us imagine what this would look like. Cut to the 2020s — when the technology around AI finally passed the research phase and organizations started implementing AI in their applications.

With the industry now at a turning point, Generative AI and Applied AI have become important areas of interest for new-age entrepreneurs and key investment domains for venture capitalists. Most of the resources on the internet teach us how to evaluate the technology and product of an AI startup. Frameworks to evaluate AI startups are yet to be established. Measuring the business impact and feasibility of new technology is a difficult task. In this article, I attempt to answer 2 critical questions while evaluating AI startups.

Is AI a suitable technology for a particular use case?

How do you create a right-to-win in AI products?

Measuring the Business Impact of Artificial Intelligence

As an early-stage venture capitalist, one of the first things I do while evaluating a startup is to check the size of the total addressable market (TAM). This is usually difficult for products implementing emerging technologies such as blockchain or artificial intelligence. In simple terms, TAM corresponds to the impact created by a product, may it be revenue increases, cost savings, time saved, etc. Modeling this impact for something untested and unused may be very difficult.

To define the impact of such products, it is important to identify the characteristics of the business that would exist around the product. This would include the industry (e.g. education, agriculture, chemicals, technology, design, etc.) and the function (e.g. manufacturing, sales, marketing, finance, accounting, etc.) which the product is serving, followed by the product type (SaaS, application, platform, SDK, etc.) and most importantly, the use case (computer vision, audition, linguistics, planning, and forecasting). Having this information handy would quantify the parameters of impact mentioned in the formula below:

All four parameters are critical in establishing the impact created by AI for a particular use case and can be translated into standard product terminology, leading to the creation of the 4-P framework for measuring AI business impact.

The 4-P framework states that AI impact depends on four parameters — the people involved in a use case, the programmability of their workflow, the productivity improvement created for them, and the precision required in a particular use case. The four parameters can be translated into standard, quantifiable, and measurable business metrics.

Let’s assess the framework with an example of a product that helps create an application’s UI/UX design based on the parameters provided by a product manager. The first step involves identifying characteristics around the potential business the product may build. The industry is technology, the function is product design, the product type may be SaaS, and the use case utilizes computer vision. Let’s look at the four parameters:

People: The key metric used here is # of designers utilized for UI/UX design, ranging from 1–2 for a startup to 10–20 for a large-scale enterprise.

Programmability: The key metric used here is % digitized workflow. For the case of a UI/UX designer and the product design and management function in an organization, the number is usually close to 100%. All product design teams can benefit from AI in some way.

Productivity: The key metric used here is # of weekly hours spent by a freelance designer and the corresponding salary per hour.

Precision: The product may have the capability to eliminate the need for a designer, but the organization may still want a human touch. Productivity improvement need not always translate entirely to monetary benefits. It is important to identify what precision and ROI are required by the customer. In this case, it may be cutting down employment time for a freelance designer from 8 hours to 2 hours per day, thereby saving 6 hours of salary paid to a freelancer.

Putting it all together, the impact created by this tool, defined as the savings in salaries of freelance designers (to the required precision), may be calculated by the product of the average number of freelancer designers employed by an organization, the number of hours reduced per month due to the tool (in this case, 6 hours per day) and the average designer wage per hour.

The Right-To-Win in Applied AI

A large impact (TAM) is not enough to build a scalable business. There should a clear moat to capture market share. Usually, moats are manifold — product, technology, distribution, insight, etc. However, in the case of applied AI, moats are specific to the technology and product. The hype around Generative AI has increased the excitement about the technology. Organizations have realized the importance of data in making decisions and are now waiting with open arms for tools that can help them do so. This strikes distribution off the list of moats.

In applied AI, the moats are the ones that solve for technical barriers to adoption — access to data, portability to new tools, user experience, and ROI.

Higher ROI

The monetary returns created by paying for a particular AI tool are the first and foremost moat that every applied AI product should develop. AI models providing sub-par results will not succeed. Period. ROI is closely related to precision. The usual thought process is that more the precision the AI is creating, the better would be the ROI. However, that is not the case. ROI has a linear relationship with the precision till the minimum required threshold is reached. After that, the effect of an increase in precision on the ROI is subdued. To put this in an example, if my business just demands that I can segment apples and oranges, a model that does exactly this would be extremely useful for me. If the model goes one step above and helps me segment red apples and green apples, this extra level of precision is good to have but not a necessity.

Access to Data

Another must-have for applied AI tools is access to high-quality data. Model libraries provided by Open AI, Stability AI, and Hugging face have commoditized access to general-purpose data for computational linguistics and vision. Startups building applied AI tools can now use these libraries, and further contextualize them to their particular use case by adding high-quality, user-specific data. Most startups today have their narrative built around the use of proprietary datasets. However, with time, as data becomes more commoditized, the effect of access to data as the right to win will decrease.


The ability to quickly transition from old, non-AI workloads to new AI workloads would, in my opinion, become a massive differentiator for winners in the market. Today, the entire AI infrastructure is in place. Startups building in this space should streamline all the pieces to create a simple and fast porting process for organizations, accelerating adoption.

Startups that can quickly put together components from all 5 layers in place and package their product on top will significantly reduce the onboarding time for organizations.

User Experience

Artificial intelligence and machine learning are complex technologies that use advanced mathematics and computer science concepts. Limited to developers and data scientists, AI can only reach massive scales when people from non-technical backgrounds have the right tools. An intuitive UX can become a differentiator for AI startups in potentially competitive categories.


The wait for practical usage of artificial intelligence is over and the technology presents a large opportunity for entrepreneurs and innovators to build the next generation of products at scale. While building, make sure that you think through why AI is needed to solve a particular problem and what could be a differentiator to make your product the most lucrative one. To all builders out there — if you ever want to brainstorm or discuss what you’re building, please reach out to me at



Parth Chhaparwal

VC @ Venture Highway | Ex-Bain & Co | IIT Kanpur