The Improzone Podcast Episode 1 - AI That Actually Gets Used: A Startup’s Take on Pharma Innovation

Indeprit Combo, CEO of Improzo, shares the inspiration behind starting the company. He observed that many biotech and pharma programs often stalled at the planning stage, lacking real-world execution. The core issue was a gap in last-mile implementation despite significant investments in data, digital, and technology. Improzo was founded to bridge this gap, focusing on impactful, scalable, and cost-effective AI solutions for the biopharma sector.

Improzo differentiates itself by strategically focusing on commercial and medical aspects within biopharma. Their platform, ISO, is modular, built on swapable layers, and trained on pharma-specific datasets. Unlike traditional black-box Software-as-a-Service (SaaS) models, Improzo integrates with existing pharma tech stacks (e.g., Data Bricks, Snowflake, Veeva, Salesforce), avoiding the need for replatforming and leveraging current infrastructure to accelerate ROI.

Improzo employs tailored AI solutions to address specific pharma challenges, such as creating analytical workbenches or agents for territory design and forecasting. These solutions consume data from multiple internal and external sources. The company customizes agents for each client's needs without rebuilding from scratch, integrating with existing tech stacks and algorithms to reduce time to insights by up to 70%.

Improzo emphasizes an outcome-based model to demonstrate value, focusing on measurable KPIs such as faster turnarounds, better accuracy, and increased usage. An example cited is a medical team (MSL) solution that structured unstructured notes, generated insights, and fed them back into the system, enabling market share analysis and leading to a multi-year deal. The goal is to provide tangible impact beyond mere data collection, especially for field-based roles where impact is harder to quantify.

A key learning is that insights don't always equal impact. Indeprit shares an anecdote about a large pharma company with over 1,800 dashboards, highlighting the inefficiency of data overload. Improzo aims to simplify this by transitioning to integrated, live, and trustable data platforms that consolidate information, enabling faster decision-making and cutting down on internal and external dependencies.

Improzo navigates large pharma by focusing strategically on AI and technology, starting with small proof-of-concept projects to demonstrate ROI, and securing executive sponsorship. This approach builds trust and fosters a collaborative partnership, which is crucial for driving significant transformational change within complex organizations.

Indeprit addresses the critical need for 'trust in AI', emphasizing transparency, explainability, and responsible AI. Improzo's ISO platform incorporates an industry-unique trust metric score, partnering with a company that supports major names like Netscape to benchmark for toxicity, vulnerability, and red teaming. This ensures AI agents operate within guardrails, preventing brand misalignment and reputational damage, especially when dealing with sensitive pharma data.

Success for Improzo is measured by customers asking for more, positive feedback, and strong referencability from senior leadership. They aim to be seen as a new category in pharma—an execution layer that delivers solutions in weeks, not months or years. Indeprit envisions Improzo driving innovation in commercial and medical sectors, with future potential expansion into R&D. His "magic wand" wish for pharma is to shift focus from measuring insights to measuring actual outcome and demonstrable usage.

Indeprit advises aspiring founders to prioritize solving problems effectively rather than leading with shiny AI technologies. For AI trends, he points to small language models (SLMs) as an underestimated area, especially for industries like pharma with rich contextual data and specific nuances. SLMs, embedded with AI, can provide highly contextualized and canonical answers, potentially offering more practical impact than general large language models (LLMs).