The problem I found
While working on customer-service systems at Insaito, I kept finding tools but no useful map of the category. Vendor pages compared feature lists. They did not explain what each platform can actually change in a customer's systems, what counts as a billable resolution, or what the bill becomes at a real support volume.
What I built
- An independent directory of 78 AI customer-service tools. Nobody paid to be listed.
- 58+ editorial posts and 531 comparison pages built around buying questions, not generic definitions.
- A side-by-side comparison that puts headline price next to the cost at a chosen number of contacts, agents and AI resolution rate.
- A cost calculator covering 21 published-price platforms. It separates tools that can query or change systems from tools that only answer from documents, and excludes vendors whose AI rate is not public rather than estimating it.
- A billing-definitions dataset showing why a $0.69 rate can cost more than a $0.99 rate when one vendor bills customer abandonment and another bills only confirmed resolution.
The research standard
Every price has a source and a checked date. “Not published” is a valid result. In the site's pricing-transparency study, 43 of 83 vendors had an actionable public rate; voice AI and agent-assist were among the least transparent categories. When sources disagree, the site prints the disagreement instead of averaging it away.
The calculator also says what it cannot know: implementation, integration engineering and the state of a company's documentation. That matters because a neat software quote is not the same as the cost of making the system work.
How it is finding readers
Traffic is coming through referrals, direct visits and search. Pages have also appeared in answers from ChatGPT, Claude, Gemini and Google. I am leaving counts out until I have a clean reporting window.
What changed while building it
I used my own directory to research what was actually agentic about a tool and found that the answer was missing. “Can it change something in your systems?” became a first-class field across the directory. Tables now generate from the data files, and a retraction blocks the build until the prose changes too.