01 · The situation
Good research that did not compound.
The research was not bad. It just did not accumulate. Findings lived in whichever tool captured them, so the effort of compiling a picture fell on whoever needed it, every time they needed it. The predictable result: teams defaulted to the most recent thing they remembered, because that was cheaper than assembling the truth.
This is a research operations problem wearing a data problem's clothes. More research would not have fixed it.
"I know the insights exist somewhere, but pulling them together takes more time than the decision itself."
"Every time we start a new initiative, it feels like we're rediscovering the same insights instead of building on existing knowledge."
"We're not lacking data. We're lacking a way to connect it meaningfully."
Three of nine conversations. The same complaint arrived from product, design and research independently, which is what moved this from a nuisance to a system to build.
02 · The call
Connect the tools. Replace none of them.
The tempting version of this project is a single system of record that everyone migrates to. That project fails. It asks four teams to change where they work in order to solve a problem none of them owns.
BotDojo reads from the tools people already use and leaves them exactly where they are. It normalises what it finds, answers questions in plain language, and links back to the source artifact so an answer can always be checked. Speed was chosen over completeness deliberately: a fast, usable, traceable answer beats a perfect analysis that arrives after the decision.
Sources, unchanged
BotDojo
Consumers
Read only connectors are the whole reason adoption was possible. No team had to move, agree on a schema, or change a workflow to benefit.
03 · Where it landed
What is proven, and what is not.
Measured
3 wks to 1.5
Research cycle time, a 50% reduction
Internal figure from the GRC design function. The same number reported on the design practice page: BotDojo is what produced it, so the two are one result, not two.
Observed
Research became self serve. Product and support teams retrieve customer signal without a researcher in the loop, which is what stopped the same understanding being rebuilt every quarter.
Adoption came from the tool being easier than the alternative rather than from a rollout. It was embedded in workflows people already ran instead of arriving as a new ritual.
Too early to claim
Whether faster retrieval produces better decisions, or just faster ones. Cycle time is a proxy, and it is the honest limit of what this number shows.
Whether teams verify the source links or trust the summary. That behaviour is the difference between a research system and a confident guessing machine, and it needs watching rather than assuming.
04 · Status
In use, and still the interesting part.
BotDojo is running and is part of how the design function works. It also powers the Ask AI assistant on this site, which is the same idea pointed at a much smaller corpus.
The lesson that transferred: the problem was never a shortage of data, it was accessibility and synthesis. Treating research as a system rather than a series of projects is what let it scale. AI made the shift possible, but the value came from embedding it in a workflow, not from the model.