Tier 01
Company outcomes
Is the business healthy?
- Revenue growth
- margin
- cash position
- retention
- capacity
Undefined process in, unreliable automation out.
The reason AI fails inside most companies is not the tool. It is that nobody has ever written down how the work actually happens. There is no definition for an agent to run on.
Definition before automation
So the method starts where the failure starts: with definition. We do not ask a company to change what it already has — we optimise what it has, by adding the connection and architecture that makes it AI-native. A new tool is introduced only into a genuine gap.
In this guideStarting point
01Starting point
You have already started. Nobody has structured it.
Your team already pastes things into a chat assistant. Someone drafts replies with it; someone else summarises a document with it. That is not nothing, and it is not a strategy — it is level one of seven, and it is ungoverned.
Which is the useful place to begin. You are not behind, and you are not starting from a blank page. You are somewhere specific on a ladder, and the first job is to find out where.
02Seven levels & gates
Seven levels — and what earns the next one
Every area of a business sits somewhere between fragmented and continuously optimised. The levels matter less than the rule that moves you between them:
You do not climb because time passed. You climb because the previous level's numbers earned it.
LevelWhat earns the next one
01FragmentedAn audit: what exists, what is undefined.
It depends who’s doing it. Information lives in inboxes, spreadsheets and people’s memory. Someone already uses AI privately; there’s no rule about it.
02DigitisedProcesses written down, each with an owner.
The software exists, but everyone uses it a little differently. No one owns the process.
03ConnectedClean records; data reliable enough to act on.
Systems exchange data. Reports are still assembled by hand.
04AutomatedAutomations stable; exception rate low.
Predictable work moves on its own. Someone can name exactly what runs automatically — and what breaks.
05AI-assistedMeasured accuracy and acceptance rate.
AI drafts the real work; a person approves before it ships. Someone can state the approval rule out loud.
06Agent-enabledLow override rate; zero unauthorised actions.
Multi-step work runs within defined authority — with limits, logging, and a named owner.
07Continuously optimisedThe top of the ladder.
Performance data changes the rules, and someone reviews those changes.
That is what makes measurement something worth wanting rather than something we have to justify. The gate is the mechanism, not the paperwork.
03The threshold
The line that matters is four to five — Automated to AI-assisted
04
Automated
AI proposes.
05
AI-assisted
AI produces work that leaves the building.
Below it, AI proposes. Above it, AI produces work that leaves the building. Every real governance conversation lives on that line — not at the top of the ladder, where the vendors put it.
Most companies should stop at five or six — and should be told so. Everyone else is selling level seven. For most businesses of twenty to two hundred people, a human approving each output — or governing a bounded agent by exception — is the right place to stop, and staying there is a decision rather than a failure.
04Twelve business areas
Twelve areas, checked against a standard
Coverage is derived from a recognised process-classification framework rather than invented, which is how we answer the question "how do you know that is everything?" Each area carries the same two questions: what has to exist before AI touches it, and what you would measure afterwards to know whether it worked.
Inspect an area
Area 01 / 12
Strategy & market intelligence
Must exist · required structure
- Objectives
- target market
- competitors
- positioning
- an annual plan
Measure · evidence returned
- Revenue growth
- forecast accuracy
- strategic milestones
Area 02 / 12
Products & services
Must exist · required structure
- Offer catalogue
- pricing
- packaging
- a development process
- quality standards
Measure · evidence returned
- Gross margin by offer
- adoption
- time to launch
- defect rate
Area 03 / 12
Marketing & demand generation
Must exist · required structure
- Brand system
- audience database
- content calendar
- lead capture
- attribution
Measure · evidence returned
- Qualified leads
- conversion rate
- cost per lead
- attributed revenue
Area 04 / 12
Sales & CRM
Must exist · required structure
- A CRM
- lead stages
- qualification rules
- pipeline
- quoting
- follow-up
Measure · evidence returned
- Pipeline value
- win rate
- sales cycle
- forecast accuracy
Area 05 / 12
Operations & service delivery
Must exist · required structure
- Scheduling
- capacity
- written procedures
- task ownership
- quality control
Measure · evidence returned
- Delivery time
- on-time completion
- utilisation
- rework
Area 06 / 12
Supply chain & inventory
Must exist · required structure
- Purchasing
- vendor records
- inventory
- orders
- logistics
Measure · evidence returned
- Inventory turnover
- stockouts
- order accuracy
- fulfilment time
Area 07 / 12
Customer service & success
Must exist · required structure
- Ticketing or a shared inbox
- service levels
- knowledge base
- escalation
Measure · evidence returned
- Response and resolution time
- retention
- churn
- expansion
Area 08 / 12
Finance & administration
Must exist · required structure
- Accounting
- invoicing
- receivables
- budgeting
- approvals
- cash forecasting
Measure · evidence returned
- Cash flow
- margins
- days receivable
- budget variance
Area 09 / 12
People & organisation
Must exist · required structure
- Org chart
- roles
- hiring
- onboarding
- training
- performance
- policies
Measure · evidence returned
- Time to hire
- onboarding time
- turnover
- training completion
Area 10 / 12
Knowledge & information
Must exist · required structure
- A central repository
- naming rules
- permissions
- version control
Measure · evidence returned
- Search success
- duplicate documents
- time spent searching
Area 11 / 12
Technology, data, security, integrations
Must exist · required structure
- App inventory
- identity
- permissions
- integration architecture
- backups
Measure · evidence returned
- Uptime
- integration failures
- duplicate records
- recovery time
Area 12 / 12
Risk, compliance & resilience
Must exist · required structure
- Policies
- contracts
- approvals
- audit trail
- incident response
- continuity
Measure · evidence returned
- Open risks
- policy exceptions
- incidents
- audit findings
05The weakest-area rule
An area is capped by the weakest area it depends on
This is the rule that makes the picture honest. Automating on top of an undefined process does not raise the level — it raises the blast radius. A sales function running at level five on top of records nobody trusts is not at level five; it is at the level of the records.
What it looks like on a Tuesday / illustrative example
The tool is rarely where the failure begins
Follow one ordinary CRM defect as it becomes an AI output, a correction task, and finally a problem another team inherits.
01Records & data · L2 DigitisedThe source is incomplete
The CRM exists, but nobody maintains one dependable version of the customer.
- Customer left in March
- Last note is four months old
- Company entered three ways
02AI layer · L5 promisedThe system reads what exists
It has no hidden memory of your company — only what your people wrote down.
03Confident outputThe defect returns as an answer
The output looks polished because the interface is better than the evidence.
- Warm follow-up to a lost customer
- Healthy deal marked at risk
- False next-best action
04Human correctionThe removed work returns
Every output must be checked and repaired by the person the tool was meant to free. The work was not removed; it moved, and it moved to the person least able to refuse it.
05Inherited consequenceThe weak spot travels downstream
Each team was promised level five and receives the level of the records.
- Customer success L2
- Delivery L2
- Finance L2
- Why it happens
- AI knows only what it can read.
- What the company sees
- Level 5 activity resting on level 2 evidence.
- What it creates
- Three levels of exposure, not capability.
5 promised → 2 supported
Sales is built to level five. The records it reads from are at level two. Those three levels of difference are not capability the company has — they are exposure: work it is trusting, resting on records nobody has defined.
06The operating circuit
The operating circuit — six roles, not logos
Every business runs the same six roles, whatever the vendors. We name roles rather than products deliberately. Naming products implies integrations we have not built, ages badly as those products change, and obscures the actual argument: the shape is universal; the vendor is interchangeable.
01
Channels
where information enters the business
02
Systems of record
where official information lives
03
Integration & workflow
what connects them and moves information
04
AI & decision support
what interprets unstructured information
05
Data & performance intelligence
what tells management what is happening
06
Governance & control
what determines what the systems may do
07Five measurement tiers
Measurement — five tiers, never one score
A single composite number hides the thing that matters: strong sales can conceal poor cash flow or rising churn. So measurement stays in five separate tiers, and the fifth is the one most providers skip.
Tier 02
Revenue engine
Is marketing → sales → success converting?
- Lead-to-meeting
- meeting-to-opportunity
- sales cycle
- cost of acquisition
Tier 03
Operational performance
Is delivery working?
- Cycle time
- on-time delivery
- backlog
- rework
- exceptions escalated
Tier 04
Customer performance
Are customers staying?
- Resolution time
- satisfaction
- churn risk
- renewal
- lifetime value
Tier 05Often skipped
Automation & AI performance
Is the AI actually working?
- Hours returned
- exception rate
- human override rate
- accuracy
- unauthorised actions
An automation that saves clicks but does not improve revenue, cost, customer experience, capacity, risk, or quality is not a success.
08Where to start
Where to start — and where never to start
Candidate work is scored on two axes: what it is worth, and how feasible it is. Worth counts frequency, time consumed, error cost and compliance risk. Feasibility asks whether the process is standardised, whether the data is reliable, whether the decision rules are clear, and the question that settles most arguments — what happens if it is wrong?
Worth →
High worth · low feasibility
High worth · high feasibilityCandidate start zone
Low worth · low feasibility
Low worth · high feasibility
Independent veto
What happens if it is wrong?
Bounded and recoverableStart
Consequential or uncontrolledNever
The two axes rank candidates. The consequence question can still veto one.
Usually start here
Examples, not plotted scores.
- Lead intake and routing
- Meeting notes into the system of record
- Follow-up reminders
- Customer onboarding
- Invoice reminders
- Project setup
- Weekly performance reporting
- Knowledge search
- Ticket classification
Never start here
The veto register.
- Autonomous financial commitments
- Hiring or termination decisions
- Legal conclusions
- Sensitive customer decisions
- Processes that are not yet stable
- Data nobody trusts
- Workflows nobody can explain
- Agents with unrestricted access
09Governance
Governance, stated whether or not you ask for it
Every engagement establishes the same minimum, requested or not.
Fixed structure
Sixteen controls transfer
The same minimum exists in every engagement.
- Data inventory
- Application inventory
- Access-control review
- MFA
- Least privilege
- Backup and recovery
- Vendor assessment
- Personal-information handling
- Retention rules
- AI acceptable-use policy
- Approved tools list
- Human-approval matrix
- Audit logging
- Incident response
- Continuity
- Employee training
Operating parameter
Obligations are local
We establish which regime the business operates under before applying the structure.
- Governance structure
- The portable control architecture.
- Applicable obligations
- The business's operating regime — not someone else's frame.
One governed engagement · correctly parameterised
Two boundaries come with that, and we state both up front. We deliver process, checklists, and policy scaffolding — not legal advice. And the rules themselves are a parameter, never a constant: the structure of a governance layer transfers between countries, the obligations do not. We establish which regime a business operates under at the start of an engagement rather than carrying someone else's frame into the room.
How this becomes an engagement — and what you end up owning — is on the consulting page. To place yourself on the ladder first, start there.
From method to engagementFind the constraint.
Find the constraint.
Know what comes next.
A Diagnose turns the framework into a document about your company — one you keep.