Research and Innovation: Workflows
The research to knowledge to product to service to revenue chain, where each arrow fails, and the handoffs that keep findings moving.
Markdown. No sign-up, no email.
What arrives#
| From | What | Becomes |
|---|---|---|
| Every function | Questions blocking a decision | The queue, filtered by whether an answer changes anything |
| AI Strategy | Capability gaps in the transformation portfolio | Directed research |
| Sales | What customers keep asking for and we cannot yet do | The most commercially loaded questions available |
| Engineering | Technical unknowns blocking a design | Time-boxed investigations |
| Scanning | What changed in the eight domains | Candidates, heavily filtered |
Questions from Sales carry the most commercial weight, because a customer asking for something we cannot do is a market signal with a name attached.
What leaves#
| To | What |
|---|---|
| AI Engineering and Product | Replicated findings ready to build on |
| AI Strategy | Capability assessments feeding the sequencing decision |
| Marketing | Publishable research, which is the strongest authority content available |
| Knowledge Management | The negative-results record, which is the compounding asset |
| CEO | Anything that changes the strategic picture |
The chain#
The second lane is the one that decides whether this function compounds or restarts every year. A finding that exists only in the head of whoever ran the experiment leaves the company when they do.
Handoff contracts#
With AI Strategy. They own sequencing; research supplies whether a capability is ready. The useful answer is frequently "not yet, and here is what would have to change", which is more valuable than a yes because it is checkable later.
With AI Engineering and Product. They productise; research does not. Proving a thing works and judging whether it is ready are different assessments, and doing both in one place leaves nobody to disagree with the first.
With Marketing. Research output is the strongest content the company can publish, because nobody else can publish it. Negative results are publishable too and are unusually well read, since almost nobody publishes them.
With Knowledge Management. Every answer, positive or negative, lands in the record in a form someone who was not there can use.
Cadence#
| Weekly | Running experiments, anything past its time box |
| Monthly | Answers, negative results written up |
| Quarterly | Queue review with the CEO and AI Strategy |
| Annually | Chain conversion: how much reached product, service and revenue |
The failure this design is built against#
A research function that is genuinely busy, reads widely, prototypes constantly, publishes occasionally, and produces nothing the company can sell, while the same questions get investigated again every two years because nothing was ever written down.
The defences are the intake test at the front, the kill criterion in the middle, and the negative-results record at the back. All three are unpopular, and the last one is the one that makes the function an asset rather than an expense.