AI value creation for investors and portfolio companies
by Dr. Phil Winder , CEO
AI value creation is the work of finding where AI will change a portfolio company’s revenue, costs or risk, testing that on the company’s own data, and building the system that delivers it. Venture capital and private equity investors bring in Winder.AI to do that work with their portfolio companies. We are an engineering-led AI consultancy, independent since 2013, and the senior engineers who assess each company build what the assessment recommends.
From portfolio screen to production build
The work starts with the investor and narrows to the companies where AI will pay:
- Portfolio screen. With your platform or operating team, we look at each candidate company’s data, products and largest costs, and rank where a model would change a number its board already tracks, such as cost to serve, forecast error, conversion or price.
- Company assessment. For the companies at the top of that ranking, an AI readiness assessment checks whether the data, the team and the infrastructure can support the use case. Where the data has gaps, data strategy consulting closes them first.
- Evaluation before build. When the return is uncertain, we test the idea on the company’s historical data before anyone commits to production, and report where it wins and where it loses.
- Build and handover. The same engineers build the system into production, then train the company’s team to run it.
The investor gets a ranking it can act on across the portfolio, and each company gets a plan its own engineers can follow.
Portfolio company case studies
Our published case studies are engagements with the kinds of companies investors back: an online marketplace, a hotel revenue platform and a generative AI company. Duetto came to us through one of its investors, which is how most portfolio work starts.
At Apartment List, an online rental marketplace, deploying a new model took months and the data each model trained on differed from the data it saw in production. In 2025 we mapped the machine learning pipeline, moved it to a unified Chalk feature store with automated Kubeflow pipelines, and trained the team to run it, after which the case study records no training/inference drift and data scientists deploying their own models.
Duetto, whose platform powers pricing decisions for thousands of hotels, wanted to know whether reinforcement learning (RL) could improve on its existing pricing optimisation before it built anything. We ran offline experiments on its booking data, in which the RL agent outperformed a baseline that imitated past pricing decisions and showed a positive revenue lift signal, and documented the gaps in the data that production would need closed first. The case study, published in 2026, records that the evaluation de-risked the initiative and avoided premature production investment.
“The willingness to confront hard problems head-on rather than optimising for superficial wins. The work was treated as true research, with careful debugging and transparent discussion of failures as well as successes. That rigour significantly increased our confidence in the conclusions. Recommendation Score: 9 / 10.”
Sr. Director of Data & ML, Duetto Research
With Stability AI in 2023, we built the product around Stable Audio, its text-to-music model: the web application, the backend services, payments, identity and the MLOps infrastructure that trains, deploys and monitors the model. Stable Audio was named one of TIME’s Best Inventions of 2023, and Stability AI’s Head of Stable Audio reported over 500,000 generations in its first two months.
We have also worked directly with a venture capital firm, developing applied AI frameworks that help its portfolio companies speed up product development and operational decisions, so early-stage founders can adopt AI safely as they scale.
Cost
Our pricing page publishes indicative rates of £150 to £300 an hour on time and materials, drawn from real projects, and we quote each engagement on its scope. One example on that page is a well-funded startup that hired a team of three software-oriented data engineers, on time and materials, to push the development and adoption of a new product.
Who this is for
This work is for venture capital, growth equity and private equity firms with a platform or operating team, and for the chief executives and chief technology officers of the companies they back. The work needs data a company already holds, such as transactions, bookings, prices, usage or support history. If a product on the market already solves a company’s problem, we will recommend it on the first call. The Duetto evaluation shows the same test applied to our own speciality: it set out where RL beats simpler approaches and where it does not.
Frequently asked questions
AI value creation is the work of finding where AI will change a company’s revenue, costs or risk, testing it on the company’s own data and building the system that delivers it. For venture capital and private equity investors it runs across a portfolio: a screen ranks the companies where AI would change a number the board tracks, an assessment checks the data and team in each top-ranked company, and engineers build the system into production. Winder.AI, an engineering-led AI consultancy independent since 2013, runs all three stages with the same senior engineers.
We work with the investor’s platform or operating team to choose which portfolio companies to start with, then with each company’s leadership and engineers on the assessment and the build. The investor gets one ranking across the portfolio, and each company gets a plan and, if it chooses, the team to deliver it. Our AI readiness assessment describes the company-level method.
Usually the investor’s portfolio or platform manager introduces us to the portfolio company, and the portfolio company engages and pays us directly. Whoever pays for the work owns what we build.
Winder.AI’s pricing page publishes indicative rates of £150 to £300 an hour on time and materials, drawn from real projects, and we quote each engagement on its scope. The scope depends mainly on how many portfolio companies are screened and how many go on to a build.
Our published case studies are engagements with the companies themselves. Duetto came to us through one of its investors, and we have also worked directly with a venture capital firm on applied AI frameworks for its portfolio companies. In 2025 we moved Apartment List, an online rental marketplace, to a unified Chalk feature store, after which the case study records no training/inference drift. For Duetto, a hotel revenue platform, offline experiments showed a reinforcement learning agent outperforming a baseline that imitated past pricing, and the 2026 case study records that the evaluation avoided premature production investment. In 2023 we built the product around Stable Audio with Stability AI, and it was named one of TIME’s Best Inventions of 2023.
A small team of senior engineers, working remotely from Europe and the USA and led by Phil Winder, author of O’Reilly’s Reinforcement Learning, runs the portfolio screen, the company assessments and the builds. There is no sales layer between you and them.
We build machine learning for forecasting, pricing, recommendation and anomaly detection, reinforcement learning, large language model applications and AI agents, and the data and MLOps work underneath all of them. We work across OpenAI, Anthropic, Google and open-source models and pick between them for each company’s problem and data.