The problem
A financial data company had spent years building something hard to replicate: standardized figures and research covering thousands of public companies, all in one consistent format. It sold that data through an API and through its own website.
Then its customers started asking their questions somewhere else.
Analysts, advisors, and individual investors began putting financial questions to ChatGPT and Claude instead of logging into a portal. The assistant answered from whatever it had picked up during training — sometimes out of date, sometimes wrong, and never the company’s own licensed research.
Every one of those was a sale that did not happen. The data existed. The customer existed. The moment of intent existed. What was missing was any path between them.
This is not a financial data problem. It is happening to every company whose product is information: market data, research, pricing, benchmarks, logistics, compliance. Customers are asking assistants first, and if your data is not there, someone else’s answer is.
The solution
Put the company’s data inside the assistants its customers already use, with the commercial rules intact.
A customer adds it to ChatGPT, Claude, or Microsoft Copilot and starts asking questions. No integration project on their side, no developer required. They get the company’s real, current, licensed research instead of a guess — and the company gets paid for it.
Three things made it work commercially.
It is a genuinely new channel, not a feature. The API sells to engineering teams. This sells to the professional who wants an answer. Those are different buyers, and the second group was previously unreachable. It is additive: customers who wanted an API still want an API.
The paywall stays up. Plan rules are applied before anything reaches the assistant, so a free user and an enterprise customer asking the same question get different depth, coverage, and history. The free tier is a deliberate instrument rather than a giveaway — enough to prove the data is real, structured so that the obvious follow-up question requires a paid plan.
It doubles as discovery. Someone who has never heard of the company can encounter its data mid-conversation and be shown that more exists. Very few channels put a paid product in front of a qualified buyer at the exact moment they are asking the question it answers.
The first release deliberately covered one slice of the catalogue rather than everything, which got it in front of real customers in weeks. Coverage and pricing tiers were then extended based on what people actually asked, rather than on what had been guessed in advance.
The Google Cloud partnership
Node8 is a Google Cloud reseller and implementation partner, and this engagement shows why that matters: we did not only design the product, we stood up and ran the platform underneath it.
- Google Cloud — the foundation the service runs on, including the search layer that makes the analyst research library findable
- Google Workspace — resold and implemented for customers standardizing on Google (see also our Gemini Enterprise practice)
- Google security tooling — available through the same partnership
- One commercial relationship — a single partner for the product and the platform it runs on, rather than several suppliers pointing at each other
Why Node8
We have taken this through to the other side. Getting data into an assistant is the easy part. Getting it through Anthropic’s and Microsoft’s approval processes, with the packaging, security posture, and access behaviour each one expects, is where projects stall. Those reviews are iterative and run on the platform’s calendar, not yours. We know what they ask for, which means starting the process early instead of discovering the requirements at the end.
We know where the time actually goes. On this engagement the connection itself was about 15% of the effort. Security, reliability, and operations were the rest. Anyone scoping this as a connector project will underestimate it, and the overrun lands in exactly the places that matter to a business putting its name on a paid product.
We protect the commercial model first. The most common failure is treating access rules as something to bolt on later. We keep enforcement where your existing systems already handle it, so a pricing change takes effect everywhere at once and you never end up maintaining two versions of the truth about who may see what.
We build once and distribute repeatedly. The same work reaches ChatGPT, Claude, and Copilot. What differs between them is packaging and review, not the underlying build — so the second and third channels cost a fraction of the first.
We did the homework on this market. We catalogued every app in the public ChatGPT directory, and the findings are published in The State of ChatGPT Apps. The short version: almost nobody has built a second one, and very few do more than answer a simple question. The bar is lower than most companies assume, and it will not stay that way.
If your business sells information, the question is not whether your customers will ask an assistant. They already are. The question is whose answer they get.
Node8’s connector practice is at /mcp, and the differences between the distribution channels are covered at /mcp-platforms.