For the last 15 years, I'm that person in the company who always said "let's not build it ourselves".

In the last 6 months, I'm that person in the company who always said "let's just build it in a few days".

I agree with you 100%. Fin and products like this simply do NOT solve the hard part of providing support in 2026. Basically, the hard parts are (1) coming up with the tools for agents to use, like searching for data, making updates, etc. (2) reviewing the logs of actual usage and adjusting prompts, docs, tools based on the real feedback. (3) tuning human escalation procedures.

This process is an ongoing effort, with an upfront engineering commitment which depends entirely on the product, but can be months of work. But if you have your own backend, I would argue this hard works is made HARDER by implementing something like Salesforce/Fin, because you have to now pipe a bunch of data and structure over to them, which is a pain.

LLM models capable of doing this are a commodity, the UI for customers and support teams is pretty trivial, the database/backend is trivial.

Outside of some cases, if you have your own app, and you have a given support volume, build your own.

	▲	aurareturn 34 minutes ago \| parent [-]
		Agreed. A recent example is that we replaced our support ticket system with an in-house built one. The new system lives in the same database as our app. Every ticket now has real live customer data. You can't get this kind of integration easily using a 3rd party tool. It was surprisingly easy to build. Just took 2-3 days for us. Massive improvement in productivity. Took about $100 in tokens to build. Maybe an hour of maintenance work per week. This larger company took 48 hours: https://tradecore.com/resources/blog/we-replaced-zendesk-in-... Doing this would've be seen as crazy in 2023. But in 2026, it's often an advantage. Better, more integrated, cheaper, faster. I'm happy to buy if it's something I know we don't have the expertise to build. For example, you'll never catch me saying we need to build our own database. But for something like Fin, I know exactly how to build it for my company and I think I can build a better agent with better context, faster iteration, cheaper, and more tailored to my company's business.