Activate · Decision walkthrough

One decision walked through
the Knowledge Corridor

The type of decision many businesses make. The framework gives you a structure to work through the information. It names the layers of knowledge the decision draws on, and shows which AI approach can help.

The scenario

Should we invest to save this customer?

A significant customer is signalling churn.

The customer has started to absorb more effort than they return, impacting margins. Support costs are up, engagement is down, and the main contact has moved on. Renewal isn’t due yet, but the team has to decide now: invest to save them, match the effort, or let them go?

Making the right decision depends on the knowledge the company has. It’s just not in one place.

Step 1 · Classify

Which layers of knowledge does this decision draw on?

Every layer contributes something. What matters is where the decisive knowledge sits, and whether AI can draw on it.

Customer-retention fundamentals. Any experienced account manager recognises the early-warning signs: the main contact leaving, support tickets climbing, growth conversations stalling. The pattern is the same across any software business. The model already knows the pattern. Whether the AI has the data to apply it to this customer is what Step 2 shows.

Retention economics in this sector. What winning a customer actually costs, what they return over time, the cost of saving them versus letting them leave, how today’s discount sets precedent for the next renewal. Takes years in the sector to read correctly. Partly in the model, increasingly covered by specialised tools.

L3a

This company’s retention playbook. Which accounts count as top-tier, who can sign off retention spend, what discount needs leadership approval, renewal template language, past retention attempts that worked. Written down in SharePoint, a CRM playbook doc, an operations wiki. Scattered, but retrievable.

L3b

The judgement call. Will this customer respond to a retention offer? Can the new sponsor be won over? How does the regional VP weigh reference value against margin? Which past retention attempts worked? Unwritten. Held in the account manager’s head, the VP’s pattern recognition, and the Slack threads where concerns get flagged. Only people have it. That’s where the decision is made.

Step 2 · Locate in the corridor

Where across your systems does this knowledge actually sit?

The corridor doesn’t centralise knowledge. It shows how scattered it is. For this specific decision, here’s which systems hold which pieces.

CRM

L3aAccount history, renewal terms, support-case summary, opportunity pipeline, contact change-log. Partly available to AI if CRM is integrated.

Data & analytics platform

L3a + L2Usage trend dashboards, feature-adoption rates, comparable customers’ churn patterns, engagement-score models. AI can use it if analytics feed an AI layer.

Document repositories

L3aRetention playbook, tier-1 account SOP, past customer case studies, discount-authority matrix. AI can use it with a RAG setup that indexes the right spaces.

ERP & back-office

L3aBilling history, discount precedents, AR aging, payment reliability. Available to AI if integrated. But most retention AI systems never are.

Collaboration suite

L3a + L3b tracesAccount review decks, draft save-plans, board updates about strategic logos, email trails with the customer. Partly available. Depends on permissions and indexing.

Real-time communication

L3b — highest densityThe account manager’s running commentary, the VP’s one-liner judgement in Slack DMs, internal channels where risk is flagged long before it lands in a system-of-record. This is where the decisive judgement is most often visible. Mostly outside what AI uses today.

Reading it. The retention playbook is documented. The usage data is modelled. The billing history is clean. None of that tells you whether this customer is savable. The judgement that decides the call sits in Slack threads and in the account manager’s head. AI working only from the documented systems gives you a confident answer with only part of the picture. That’s Context Debt: confident output, partial grounding.

Step 3 · Plot on the Activation Map

The Activation Map for this decision

Which deployment option actually helps?

Four ways AI could support this one decision. Each behaves very differently. The knowledge each one draws on shows what your architecture must make available for the AI to add value.

↑

Outcome value

High value · Far from knowledge · Context Debt

AI recommendation without organisational context

Option C · Context Debt

Generic AI produces a confident save/don’t-save call from a thin slice of context. No L3b, no precedent, no VP read.

L1L2L3aL3b

High value · Close to knowledge · L3 Advantage

Retention agent that forms a view

Option D · L3 Advantage

A reasoning agent synthesises CRM, usage, past saves and Slack threads into a hypothesis the AM can interrogate. Not a briefing, a view.

L1L2L3aL3b

Low value · Far from knowledge · L1 Commodity

Copilot drafts a renewal email

Option A · L1 Commodity

Generic outreach drafting. Saves time. Tells you nothing about whether to save this customer.

L1L2L3aL3b

Low value · Close to knowledge · L3a Foundation

Retrieval-only briefing

Option B · L3a Foundation

AI indexes CRM notes, tickets and past success plans into a clean briefing. Shows what you already know. Doesn’t form a view.

L1L2L3aL3b

←Far

Proximity to knowledge

Close→

ReachedPartialMissed

Reading the four options. Option A (L1 Commodity) saves time but tells you nothing new about the customer. Option C (Context Debt) makes the call on thin context. Confidence high, grounding thin. Option B (L3a Foundation) shows you what you already know, organised. Option D (L3 Advantage) is categorically different: it synthesises across CRM, usage, past saves, and Slack to form a view, a point the AM can interrogate rather than documents the AM has to read.

What Knowledge Strata makes visible

Without the framework, these four options look like four vendor pitches, each claiming to solve retention. With it, they separate: one saves a bit of time, one is risky, one gives the account manager a better briefing, one synthesises the knowledge the decision actually depends on.

More importantly, the option that fits tells you where your architecture has to invest. Drawing across CRM, usage data, past-save precedents, and the team’s working chat isn’t a vendor decision. It’s a programme of work at L3a (integration) and L3b (extraction). The Activation Map is the planning tool for that work.

Run the Corridor and the Activation Map together for every decision that matters. The Corridor shows what’s at stake. The Activation Map plans the investment that delivers it. That’s how architecture decisions get made on evidence, not on vendor demos.

Continue

Project walkthrough →

See the same kind of thinking applied to a whole AI investment, not just one decision. The Activation Map across an initiative.