Part 1 argued that a functional grid is an inventory, not a roadmap, and that good use cases are discovered through analysis rather than named in a workshop. It closed on a foundation the grid assumes and rarely funds. This is part 2, on that foundation and how a single piece of work produces both the ranking and the knowledge layer underneath it.
The precondition everyone draws and almost nobody builds
Look at the foundation layer on those architecture slides, the one labelled something close to enterprise knowledge management. Right instinct. Least funded box on the page.
What it actually requires is every process, policy, SOP, SLA, decision right, and exception rule sitting in one current, consistent, queryable model. Most banks are a long way from that. Documentation is scattered, last accurate at the end of a consulting engagement. The real version of the process is an undocumented workaround in three people's heads.
This is governance in disguise. An agent cannot be pointed at a process that lives in three people's heads. Neither can a governance framework. Centralised AI governance sitting over undocumented processes is a committee, not a control.
The uncomfortable part is that this layer cannot be bought as a deliverable. A consultant can hand you a documented process, but the day the process changes, the document is wrong again. The knowledge layer has to be a living output of the work itself, or it decays straight back into the same three heads. AI process mapping is what keeps it alive: the model is generated from the process and updated as the process moves.
How ESSAM closes both gaps at once
ESSAM produces the ranking from part 1 and this knowledge layer as byproducts of the same work.
Use cases surface from analysis, not a workshop. ESSAM baselines the process, identifies waste, applies the E-S-S-A-M framework to each wasteful activity, and ranks opportunities against measured waste. No consultant is needed to go and find them.
Every artefact is generated from the live model. SOPs, policies, SLAs, OLAs and RACI charts stay in sync, versioned and retrievable. The knowledge foundation becomes a consequence of the work rather than a programme of its own.
Risk and compliance are engaged at design time. FMEA runs before go-live and the approval workflow sits in-platform. You co-design the control instead of discovering the design in week nine.
Adoption is planned, not hoped for. A change gauge quantifies how much change a redesign introduces, so communications and training can be sized honestly rather than guessed at.
The standard reaches the people doing the work. Deployment and feedback run over WhatsApp. In markets with near-universal daily WhatsApp use, that is a materially better last mile than an intranet nobody opens.
The proof: the use case nobody nominated
A Kuwait bank took procurement from 139 days to 57 days, a 59% reduction that retired 82 days of work at a 106.9% efficiency improvement, with no added headcount. That process was not on anyone's AI use-case grid when the work started. It surfaced from the analysis. That is the point. The highest-value process in a bank is often the one no function nominated, because the people living inside it had stopped seeing the waste as waste.
AI process mapping is what surfaced it. The grid from part 1 would never have listed it, because no workshop volunteers its own worst process.
An inventory tells you what is possible. A process model tells you what is next.
Part 1 said keep the grid as a map of ambition. Part 2's addition is that the sequence lives somewhere else, in a measured, living process model rather than a slide. Order is a measurement question, and it is answerable in a session rather than a quarter.
Pick any box. Bring us the process behind it, and we will return the redesign, the ranked opportunities, and the documented standard.
Book a demo on your process. Your actual workflow. No hypothetical use case required. Start at https://apac.essam.ai/contact.
Related reading: