Introducing Chat990

I’ve always found Data-as-a-Service, or DaaS, fascinating.

DaaS is a business model in which two or more organizations buy, sell, or trade machine-readable data in exchange for something of value. To put it simply: selling data. Whenever you hear “third-party data enrichment,” you are likely interacting with a DaaS.

Nielsen, ZoomInfo, and FactSet are a few relatively well-known DaaS companies. Nielsen collects individual and household data across media and retail services, then sells this packaged data to help marketers better understand the fruits of their labor. ZoomInfo is the go-to for go-to-market business intelligence, aggregating public and private business and contact info to help sales teams build and move prospects down a sales funnel. FactSet centralizes financial market data, such as public company quarterly earnings and P&Ls, enabling investment managers to make, well, questionable investment decisions.

DaaS is inherently unlimited: new systems will always need to be measured. While writing this newsletter, I gave myself five minutes to conjure a few DaaS ideas. I came up with:

1. Small business owners could use pedestrian traffic data to build accurate conversion funnels and better project sales. 

2. Professional tennis players could use shot-by-shot match data to inform strategy against upcoming opponents.

3. Policymakers could use household survey data to assess the drivers behind declining birth rates and aid population growth efforts.

Not bad, right?

With the advent of AI, even traditional SaaS products now leverage DaaS models for additional revenue streams. For instance, companies like Reddit and StackOverflow are turning their community forums into training data for Google and OpenAI.¹ Regardless of the moral ambiguity of this (usually discreet) tactic, if your choice is between monetizing your data versus not, it’s safe to say that your bottom line has no moral qualms with monetization, nor does your CEO.²

That said, (stay with me here), selling data is not inherently bad.

Take the compensation tool Pave, for example. Not only does Pave “sell” your data, but it is a required condition of using the service. Pave aggregates compensation data across its customers at dimensions like job title, seniority, company maturity, etc., to provide employees with compensation benchmarks and reduce compensation inequity within and across organizations.

In this case, the end outcome is fairer compensation across an organization. Who can argue with that?

🌊 The Evolving Data Landscape

Switching gears from the supply side to the demand side, I hold a strong conviction that the future of interfacing with data will be routed through LLMs and not people.

Do I think that data professionals are going away? No. However, I do believe their roles will map closely to today’s system architects and software engineers, with newfound responsibilities like designing the semantic layer, data warehouse, and overall system for users to interact with data (sound familiar?). Meanwhile, the desk support role, such as responding to ad-hoc questions, will evaporate. I’m not alone in this sentiment - it’s a fairly popular belief among data leaders that is already starting to take hold.

For example, in eCommerce, Shopify’s native "ShopifyQL” and Stripe’s AI-powered Sigma Assist enable your operations and finance managers to build analytics using natural language. In the CRM world, Salesforce (and Matthew McConaughey?) built Agentforce, a customer support bot, to replace human handling of customer requests.

If the private sector is making huge headway in AI-powered BI, my next thought is, what about the philanthropic sector?

💰 DaaS: Philanthropy Edition

It is a known fact in philanthropy that there is structural inequity in nonprofit funding decisions. I am no expert on the topic, but having worked with those directly in the space, I have some (even if minor) perspective on the problems both funders and nonprofits face. 

The TL;DR: data is not democratized across the philanthropic space, leading to inequitable, inefficient, and asymmetric funding decisions.

For one, funders often have trouble finding new nonprofits to fund that fit their mission. In my most recent full-time role, I led the data team at Resilia, a capacity-building and impact-tracking software solution for nonprofits. In short, foundations like the Bill and Melinda Gates Foundation would sponsor a cohort of nonprofits to use our product. However, we often experienced situations where funders had the budget for our offering but lacked a full suite of organizations to support. I found that baffling! You would think the main issue in today’s world would be the lack of funding itself and not the organizational deficiencies of funding delivery. Imagine if Oprah had multiple cars to give away but had forgotten to invite the audience… crazy, right?

On the other side of the same coin, nonprofits are bogged down in a never-ending cycle of grant applications for potential funders that may or may not be strong fits. Nonprofits spend fifteen (!!) hours per grant application. That is a lot of time for a bootstrapped executive director who is juggling board development, staff management and development, budgeting, and oh, don’t forget about programming towards their actual mission as a nonprofit. The kicker? It’s all for a 10-20% chance of success. Let’s say your organization only needs one grant per year to survive; that’s still 75 to 150 hours spent annually on grant writing. There must be a better way to filter out low-likelihood opportunities and increase the win rate to over 50%.

Lastly, the individual donor does not know where to look to support small contributions. Scenario: I’m a Millennial who finally has discretionary cash in the bank after paying off my student loans.³ I’m interested in improving low-income students’ access to technology, and I want to donate to this cause. Where do I turn?

This leads me to the question: Why is there no central repository of nonprofit and foundation data in the U.S.?

Candid has wide coverage, but is expensive. ProPublica is free, but bottom-up, not top-down. Newer players like Impala are moving in this direction, and I’ll be watching their approach to leveraging AI against their first- and third-party data. Standard grant management systems like Instrumentl “proactively find the most relevant funders and grant opportunities” but don’t support funders.

📖 Chat990

Enter: chat990.org

To stay tax-exempt each year, nonprofits must submit an IRS Form 990 to the government. Different reporting requirements are required for different-sized organizations: those with under $50k in assets, for example, submit very basic information like point of contact and geographic information, while large foundations (like a Bill and Melinda Gates Foundation) are required to submit detailed financial information, like assets, expenses, income, distributions, and much more.

Chat990 uses publicly available Form 990 data, RAG (retrieval-augmented generation), and Claude 3.7 Sonnet to make nonprofit information more accessible, founded on the two earlier points of discussion:

1. With the advent of AI, how users interact with a data platform will fundamentally shift; all you need to do is design the right system.

2. Structural inequity in nonprofit funding decisions is driven by asymmetric information.

And this is just the beginning! There’s a huge opportunity to add user value by incorporating the remaining 990 data and other publicly-scrapable data from nonprofits’ websites. Further, AI-powered chat is no longer fetch. Whether it be exposing a nonprofit MCP server on top of this data, building agentic workflows to improve funder and nonprofit matching, or some other feature, myriad updates could level up the funders and nonprofits seeking to impact their communities.

Stay tuned… 👀

1  Exclusive: Reddit in AI content licensing deal with Google; Stack Overflow signs deal with OpenAI to supply data to its models

2  Having spent over half my life with the internet, I now assume that 100% of my personal data is recorded on an AWS server in Ohio.

3  This is a (surprisingly) real scenario 🤔