AI Adoption

What it is. A practical 90-day programme, with a 6-to-12-month horizon, that integrates five workstreams which most organisations handle separately:

Who it is for. Organisations of any size: small and mid-size companies, large businesses run by small teams, and large corporations or groups. The method is the same and each organisation receives its own instance; what changes is the scale, the entry point and the governance. It also arrives tuned for the sectors where CEMI already has projects and knowledge: financial services, education, technology, professional services, legal, media, arts and culture, and social impact (§27).

The goal that unites the five workstreams: the company’s Business Intelligence. At the close of the 90 days, the company has its own Business Intelligence created, configured and being tuned: its data, its shared folders, its mail and its meetings, connected with reviewed permissions, so that any authorised person can ask about an account, an opportunity, a commitment or a document and receive an answer with its source, and so that management reads the business on a dashboard and receives timely alerts. It is not a sixth workstream or a product that is bought: it is what the five workstreams build together (§22; D78, D79).

How the work is done. Week by week. Each of the 12 weeks has deliverables, an owner role, a milestone and a win of the week: something anyone in the company can see, not a report that someone claims to have written.

What is not done. The ERP is not replaced, no CRM is bought in the first 90 days, no process is automated or integrated that is not yet standardised, and no intelligence is added to one that is not written. First the current work flows with the tools that already exist; new tools are evaluated afterwards, with written processes in hand.

Phases.

How it is measured. With indicators defined in week 1, a baseline in week 2 and a reading every Friday or every month depending on the indicator. The targets each instance sets are targets, not promised results.

1. Process first, tool second. A process that is not written is neither automated nor “improved with AI”: it is written down. AI helps to write it. Then it is standardised, and only then is what is rule automated and intelligence added to what is judgement (process maturity ladder, below).
2. Do not replace what works. The work is done on the tools the company already pays for and uses. Changing systems is a decision for the horizon, not for the first 90 days. When the company’s office suite already offers an integrated AI assistant that meets the conditions of the approved tool (§4.2), that assistant is adopted first, inside the applications where the work happens; it is not an obligation for companies that use another suite, and the licensing decision and its cost always belong to the client.
3. A single source of truth per piece of data. Each important piece of data (an opportunity, a cost, a due date) lives in one place with one owner. Everything else is a copy.
4. AI does not invent the numbers. Prices, costs, margins and contractual dates come from the system that calculates them (ERP, cost sheet, manufacturer’s configurator). AI drafts, summarises, compares and reviews; it does not calculate the price.
5. A person answers for what goes out. Everything sent to a client with the aid of AI is reviewed and signed by a person.
6. Data rules before the first use. Nobody pastes a client’s information into an AI tool before there are rules signed by management.
7. Visible or it does not count. Each week ends with something that can be seen: a document in its folder, an email sent, an indicator on screen.
8. Learning by doing the real work. The learning experiences use the company’s own documents, emails and processes (with protected data), not generic exercises.
9. Bounded load. The programme cannot compete with operations. Each instance sets a limit of hours per person per week.
10. Quick wins first. This is the service’s default sequence. People adopt AI when they see it serve them in their own work, and that happens in days, not at the end of a diagnosis. That is why the first 30 days begin with quick wins and bounded pilots: a one-page minimum rule that management signs on days 1 and 2 (the traffic light of §4.1, work accounts only, nothing “red” and a person who reviews everything that goes out to a client), which satisfies principle 6 while the full rules are signed in week 2; the approved tool in the hands of everyone who already has a licence from week 1, with a short session and ready-made prompts per role; more efficient day-to-day communication in week 2 (summarised threads, drafted replies, minutes with agreements, follow-up templates); and a few pilots with an owner, on real work. It is measured from week 2 with simple readings (people who use AI every week, one win of the week per person, declared hours freed and response time to clients), which are adoption readings and not a measured return. The structural work (process mapping, platforms, CRM) is not dropped: it runs in parallel with or behind the wins, with the same deliverables, and the week-8 gate decides whether to keep, adjust or drop each pilot. Those who work in the field and without the licence receive a simpler path on the company phone. Principle 1 does not change: quick wins are people using AI in their task, with human review (rule 4 of the ladder), not processes automated ahead of time.

Design position: the Impact Arc

This service’s way of intervening follows the Impact Arc (Engage → Enable → Inspire → Empower → Connect), created by Carlos Miranda Levy as his personal motto, “NEVER HELP: Engage, Enable, Inspire, Empower and Connect”, and adopted by CEMI as a group. It is an organisational philosophy and a design position, not a framework validated by research, and it is used here only as such:

Working method: the process maturity ladder

Each of the company’s processes is worked rung by rung, in this order. It is a practical working method and a design position of ibizai.ai, not a framework validated by research: it serves to decide what is done first with each process and to make progress visible.

Rules of the ladder:

1. One rung at a time. What is not standardised is not automated, and no intelligence is added to what is not documented. If a process has no rule-based steps worth automating, rung 3 is considered met with that decision written in the process catalogue (D23).
2. Process by process. The level is tracked per process in the catalogue (D23). For a company to have processes on different rungs at the same time is normal.
3. Using AI to climb a rung is not adding intelligence. AI helps to document (a procedure drafted from a recording) and to standardise (a procedure reviewed against the template); that is the work of rungs 1 and 2. A process has had intelligence added when AI works inside the standardised procedure, in a step of judgement or language, with its signatory.
4. People can go ahead of the process. Someone can use AI in their task (workstream B) before their process is standardised, with human review. That counts on the per-person usage scale (§9), not on the level of the process.
5. Automation with an owner. Each automation is in the register (D25) with its owner, what it does and what it does not do, the account it runs under and how it is switched off. It does not run on personal accounts.
6. Evidence, not declaration. As on the usage scale, each rung is accredited with evidence.

Roles are named by function. In a small team, one person may hold more than one; in a large organisation, each role may be a team or an existing committee (§27).

How Management leads: open the way, ask, and raise the bar

Sponsoring the programme is not signing and waiting. Adoption advances at the pace of three gestures by General Management, which no other role can make on its behalf. It is a design position of the service, not a framework validated by research.

Governance rules:

• The weekly meeting is never cancelled; if someone is missing, it is held with whoever is there.
• A “blocked” pending item is a call to Management the same day, not a note for Friday.
• Every agreement is written in the programme’s tracking tool, not in a chat.

Management’s five questions

The Friday AI block has a script. General Management asks these five questions, always the same ones and in this order; each one points to a deliverable or an indicator that already exists, so that the answer leaves a trace and not an impression. There is no need to answer all five every week: it is enough that none goes unanswered two weeks in a row.

Each instance adapts this template to its sector and its contracts, and General Management signs it before the first use. In regulated sectors (banking, health, public sector) the company’s legal counsel also reviews it.

4.1 Information traffic light

4.2 Rules of use

1. Approved tool. Only the AI tool that Management approves is used, with an enterprise licence and a contractual guarantee that the vendor does not train models on the company’s data. Personal or free accounts are not used for work, not even when the consumer version of an assistant carries the same name as the enterprise version.
2. Check the contract before the tool. Some tender documents and contracts prohibit or limit the use of third parties to process their information. If the document prohibits it, it does not go in, even if the tool is approved.
3. Anonymise by default. When the task does not need the client’s name, it is replaced (“Client A”).
4. The numbers come from the system. No price, cost, margin or contractual deadline figure is generated with AI; it is copied from the source system and verified.
5. Mandatory human review. Every text that goes out to a client or a supplier is read and approved by a named person.
6. Recordings with consent. A meeting is recorded or transcribed only if all the participants know and accept it.
7. Messaging chats stay out. Conversations in messaging groups with clients are not exported to AI tools.
8. Incident log. If someone pastes red information by mistake, they report it that day to IT Support; it is recorded and corrected without punishment. Hiding it is the offence.
9. Quarterly review. The rules are reviewed every quarter with the accumulated incidents and questions.

4.3 The organisation’s IT and AI policies

The programme writes, or reviews and updates, the organisation’s IT and AI policies. ibizai.ai’s recommendation is not a restrictive one: it starts from the organisation’s real context and the tools it already has, and says clearly what can be done, where and how. A policy that only prohibits pushes use towards personal accounts, where nobody sees it.

• Sensitive and private material: when, where and how. Using sensitive information about clients or about the organisation itself with AI is not prohibited on principle: it depends on the tool. The policy carries a tools-and-data matrix: for each tool the organisation has or is evaluating, which traffic-light levels it admits, with which account, under which conditions, and who verified it. A tool is safe for a level when the data stays in the organisation’s environment or under a contract that guarantees it does not go to third parties or train models, when it respects each person’s permissions, and when IT Support can grant and withdraw access.
• What stays out of every AI tool. Passwords, keys, credentials and encryption keys, and whatever a contract or a law prohibits. Until the matrix is signed, the minimum rule applies: red information goes into no tool.
• The tools that are not safe, by name. The policy also names consumer versions, personal accounts and services without a contract, and says what they are good for: green information.
• Naming convention and order of files. One short format for naming files and folders, where each document lives and how it is versioned (D11).
• Accounts, access and devices; recordings, messaging and incidents. The work account, who grants and withdraws access, what is done from a personal device, and rules 6, 7 and 8.
• How it is done. It starts from what the organisation already has in writing; if there is nothing, it is drafted. IT Support verifies what each tool really does under the organisation’s configuration, not what its advertising says. General Management signs them (D02); in regulated sectors legal counsel reviews them; each person confirms having read them (I01); they are reviewed every quarter. They go into the learning experiences: each person knows which tool to use for what.

This is a design position of ibizai.ai, not legal advice.

Each week declares deliverables, owner role, milestone and win of the week. Each client’s instance replaces the generic ones with its own. Recommended load limit: two hours per person in weeks 1 and 2, and up to four hours per person from week 3 onwards, in addition to the weekly meeting.

Quick wins first (principle 10). By default, each instance places before this template a section with its first 30 days: the one-page minimum rule (days 1 and 2), the approved tool for everyone with a licence and the first learning experience in week 1, efficient communication in week 2 and the bounded pilots of weeks 2 to 4. Where a piece of the template starts later (D07 in week 3, D09 and D10 in week 4), the instance brings it forward in a small version and keeps its full version in the week indicated.

Each instance chooses its own, sets its baseline in week 2 and its targets in week 3. Targets are targets: they are reviewed at each monthly review and the changes are recorded.

What this scale measures is judicious use: that a person knows what to ask AI for and with what context, how to read what it returns, check it against the source, integrate it into their work and recognise when it is better not to use it. Using AI a lot is not moving up a level; using it well, and letting others use it well, is.

A practical scale for reading individual progress. Each level is accredited with evidence from real work (an email sent, a document, a prompt in the library), not with a self-assessment.

Workstream A relies on products of the CEMI group; they are not rebuilt.

• Skaills (skaills.ai) measures the starting point and the progress: competency assessments per profile, the AI readiness assessment of its transformative layer (in active development; its availability is confirmed in each instance) and the personal growth plan. The company uses its organisation account.
• ibizai.ai (ibizai.ai/practice/ai-readiness) measures the AI readiness of the organisation with its in-depth organisation assessment (§21); it also offers individual assessments, quick and in-depth, which each person may use as a complement, without replacing the Skaills baseline.
• Smoother Experiences (experiences.ailearning.global) hosts the company’s own academy with its learning experiences, organised into Learning Units. The participants learn on their own work; the Learning Facilitator accompanies, and does not lecture.

Principles of workstream A:

1. The subject is the person who learns and what they become able to do; the experience is designed from their daily work.
2. Each Learning Unit ends in a real task done with AI, with data protected according to the rules.
3. Progress is measured with the scale of §9 and with the evidence from workstream B, in addition to completion of the experiences.
4. The experiences are ordered by role: everyone shares a foundation (rules, judgement, email and documents) and each role has its own path.
5. Those who multiply are recognised. The person who documents a use, adapts it and accompanies another person in repeating it counts as much as the person who discovers it. Levels 3 and 4 of the scale are accredited only with another person’s use, and the win of the week names the person who shared, not only the person who found (I66).

Typical path:

Workstream C uses what the company already has and, where a piece is missing, CEMI group products that are already available:

The team’s documents live in a shared site of the office suite, with permissions per role, and are synchronised with each device; never in personal folders, which are lost when someone leaves (D51, I45). The platform, the order of the archive and the design of the sites, in §28.

Golden rule of workstream C: each piece of data has one single home. Each instance’s tool map says, for each type of data, which is its single source and who is its owner.

The last three rows are traps with a name of their own: they make the company appear to advance without the work changing. Having a short name for each one makes it possible to say it in the Friday meeting without it sounding like a reproach.

1. Sources. The client’s documents, interviews, plans under way. Every fact about the client is marked with its source or as “to be validated in week 1”.
2. Confidentiality. If the instance is published, it does not include names of the client’s clients, amounts, sales figures, prices, names of people, emails or credentials. Roles are named by function.
3. Adaptation. The generic deliverables are replaced with the client’s, indicators from §8 are chosen and targets are set, labelled as targets.
4. Calendar. The date of week 1 is set and adjusted for the client’s public holidays and closures.
5. Validation. The list of facts and assumptions to validate is resolved in week 1 and the instance is updated.

This programme integrates into one single journey services that ibizai.ai also offers separately:

• AI readiness assessment — ibizai.ai’s in-depth organisation assessment is the reading of the organisation’s readiness (D53 to D55, §21); each person’s baseline is the Skaills one (D04).
• AI adoption and transformation — defines the framework of phases and the gap report.
• Workflow automation sprint — reserved for the horizon (months 7–9) and only on processes that the catalogue (D23) shows on rung 2 or higher.
• Learning programmes — delivered through the academy in Smoother Experiences.

ibizai.ai’s three phases (Diagnose, Transform, Scale) correspond to Phase 1, to Phases 2–3 and to the 6–12-month horizon of this canon.

Many organisations do not only sell: they carry out projects for their clients, such as software developments or rollouts of equipment and technology across several sites. When that is the case, the programme adds a line for those who coordinate and those who take part in those projects, within workstream B, without creating a project office or imposing a methodology.

• What AI does. It drafts and keeps up to date the documents that consume the most time: the project charter from the approved proposal, the work breakdown, the register of risks, assumptions, issues and dependencies, the minutes with actions, the status report to the client, the presentation to the project committee, the impact summary of a change request, the user stories with their acceptance criteria, the test cases, the release notes, the per-site rollout checklists, the lessons learned and the portfolio view for management.
• What it does not do. Dates, hours, budgets, prices and scope are set and signed by the person who manages the project (principle 4). Nothing that goes to the client leaves without their signature (principle 5). Network or security configurations, credentials and real data of end customers never go in, not even in rollout checklists or test cases (§4.1, red level).
• Where each thing lives. The documents, in the project folder with the playbook templates (D28). The plan and the tasks, in one single execution tool: tAIm, which receives each project from StrategAI with a view by phases, agile or by action lists, or the project tool that the company’s office suite already includes. The portfolio, in StrategAI (D30).
• How it is introduced. First a pilot on a real project for about five weeks of phase 2, with the minimum playbook (charter, risk register, status report); then, with the lessons from the pilot, the full playbook becomes standard for new projects in the horizon. It is measured with I19 to I21.
• Learning. A short programme on project management with AI that is followed at the same time as the pilot, in which each Learning Unit ends in the real project document (§10).

In many organisations, client information arrives through online meetings, calls and messaging, and stays in the memory or on the phone of whoever handled it. When that is the case, workstream C adds a line to turn it into actionable information, without buying new tools in the first 90 days.

• One single entry point. Everything that is captured reaches one single inbox, with its account, opportunity or project and its area or division. The rule for people fits on one line: if it creates a commitment, it is captured (a commitment, a request, a complaint, a condition mentioned, a decision or a risk), within an agreed time (proposal: 24 hours). AI proposes the classification and a person confirms it and turns it into work (D31, D32).
• One route per channel. First, what the company’s office suite already includes with an enterprise licence: meetings with transcription and summary inside the suite, with everyone’s consent (principle 2); a short note at the end of each call; the messages that create a commitment, forwarded to the inbox; email, summarised by the suite’s assistant or forwarded to a capture mailbox (D35). Third-party note-takers, recorders and messaging integrators are external tools: they are evaluated with the rules of §4, after a quarter of evidence.
• Classify and prioritise. A short taxonomy and an explicit priority rule (for example, urgency times impact, with extra points for disqualifying deadlines and affected services), which is presented as a rule of thumb and is reviewed against real cases. Filters by account, division, owner, type, priority, date and channel.
• The prioritised brief. A daily brief per person (their top priorities, what is falling due, what is new in their accounts, what they are waiting for from others) and a weekly one for management (the area’s priorities, what is overdue per owner, the decisions requested, the risks). It is piloted with what the group’s tools already show and is built afterwards, with an approved specification (D34). It is measured with I22 to I25.
• Client conversations by messaging. Company numbers, at least two company administrators per group, agreed response and escalation times, handover before someone leaves, and what is not done by chat (closing prices or sending final documents outside the formal channel) (D33, I26, I27).
• Learning. A nano learning experience for everyone on the discipline of capture, once the inbox is already in use, and the capture of agreements within the email and meetings experience (§10).

When management asks that opportunities not be dropped, that contact with clients be regular and that commitments not lapse, workstream C turns it into three guarantees: working rules with an owner, a place in the register, a weekly review and an indicator. They add no meetings or tools: they are reviewed in the weekly meeting and in the monthly review, and they rely on capture and the brief (§16).

• No opportunity is dropped. Each open opportunity has four mandatory pieces of data: owner, next step, next-contact date and estimated closing date. None goes longer than an agreed time without activity (proposal: 10 working days). A list of opportunities at risk is reviewed every week and each one leaves with a decision; every loss is closed with its reason. It begins with a clean-up of the register in the first week (D36, I07, I28).
• Regular contact and visits. Each account has an owner and a cadence of contact and in-person visits according to its tier, proposed by the support team and confirmed by the commercial area; a visit calendar per quarter for the main accounts; each contact and each visit recorded within the capture time; and a stakeholder map for each main account (D37, I29, I30).
• No commitment or deadline lapses. Each commitment with owner and date; alerts before the due date in the brief; escalation to management of what is overdue beyond a short time; weekly review of what is falling due; and the tender and renewal deadlines in the commitments calendar with advance alerts (D38, D16, I31, I32).
• Honesty of the targets. The cadences and the targets are proposed targets and are adjusted with the evidence of the first month; if a cadence cannot be sustained by the team, the frequency is adjusted, recording does not stop.

When an organisation grows, changes leadership or discovers that its work is not written down, staff alignment stops happening by itself. Workstream A then adds an institutional onboarding for all staff, not only for new people, with CEMI’s Smoother Onboarding method and its vocabulary (participants, Learning Units, Learning Facilitator).

• Two journeys, the same content. Current staff go through an Update in the first weeks of the programme (proposal: weeks 3 to 6), never presented as an exam or as a correction; each new person goes through the Institutional Orientation within a short time (proposal: 30 days), followed by the Role Induction of their path (§10).
• Six Learning Units. Who we are; where we are going (vision, mission, values and the perspectives of the plan, in plain language); how we serve clients; how we use technology; who does what; and each person’s place. Each one ends in a real task and links to the other programmes instead of repeating them (D39).
• What management presents is presented honestly. If the vision, the mission or the values are still a proposal, they are presented as a proposal and the welcome session gathers input from the team; never as something approved when it is not (D40).
• Acknowledgements before data. Values, data rules and rules for messaging with clients, per person; a new person does not touch client information before their acknowledgements (D41, I35).
• New people: 30, 60 and 90 days. Contact from the direct manager before the first day, an onboarding buddy who is not their manager, a first real contribution in the first two weeks, success criteria that are agreed and not imposed, a fresh-eyes review between days 30 and 45 and a first-100-days plan written with their manager and their buddy (D42, I34, I14).
• Periodic re-alignment. A short session for all staff every quarter and an Update when the organisation changes (values, leadership, rules, tools or position), with an alignment pulse (I36) that is a reading, not a validated instrument.
• Design position. Smoother Onboarding orders the journey with the five stages of the Impact Arc; like the Arc, it is a design position, not a framework validated by research.

Every organisation works with several platforms: its office suite, its ERP, its manufacturers’ configurators, messaging with its clients and, in this programme, the applications of the CEMI group. When nobody decides which platform is the source of each piece of data or how they are connected, the same data is typed by hand in several places and none is entirely reliable. Workstream E puts that in order.

• Boundary with workstream D. Workstream D decides when a process is ready: it is automated or integrated only from rung 2 of the ladder, when it is already standardised (§1). Workstream E provides the means: platforms, APIs, integrations, accounts, credentials and their governance. An integration that automates a step counts as rung 3 of the process.
• One source of record per type of data. Each type of data is created and corrected in one single system; the others read it, link to it or receive a read-only copy (D43, I37). It is the golden rule of workstream C (§11) carried over to the connections.
• Platform map and API records. For each platform: what it is for, who its owner is, what it records, whether it has an API and how it authenticates, verified in the vendor’s documentation or in the code, with source and date; what could not be verified remains “to be validated” (D44, I40).
• Phased roadmap. Phase 0 (first weeks): manual exports and one-way uploads, which also measure the baseline of manual handovers (I13). Phase 1 (months 3 to 6): scheduled read-only synchronisations and no-code flows with the connectors the suite already includes. Phase 2 (months 6 to 12): API integrations, decided one by one (D46, D47, D48).
• A priority can be brought forward. If management prioritises a high-impact integration on a process that already has a written protocol (for example, having AI monitor the mailboxes and create tasks, alerts and notices that a person confirms), it enters the 90 days as a bounded pilot, with the consent of each person whose mailbox is read and minimum permissions, and it is rolled out generally after a decision gate (D49, I41, I42).
• Build or buy. First what already exists (standard connector or native function), then a partner, and building last; only on processes on rung 2 or higher; according to the manual work it eliminates, the data it moves, who maintains it and how it is switched off, and its total cost and what it ties the company to.
• Security. Service accounts belonging to the area, never personal ones; credentials in a secrets store, rotated; least privilege; a log of each run and monitoring with a notice when an integration fails or stops running (I38, I39). No integration writes figures into a document for a client or moves red information (§4).
• The file platform and the workspace. Where the team’s documents live, how what already exists is organised and how it is presented to each area are decided with §28.
• Governance. Integrations register (D45) in the format of the automations register (D25); management approves, IT Support operates, the process owner confirms the fit with the process; it is reviewed every month and every quarter with the platform map.

• Process mapping (workstream B, with links to A, D and E). Before deciding where AI comes in, each person in the programme is interviewed, one by one and without their direct manager in the room, to understand what they do, how they do it and what hurts. Per activity, the following are recorded: frequency, hours, tools and data, what hurts (retyping, waiting, correcting, searching), the AI potential (none, assist or automate) and the use case. The output is a map of work per role validated by each role and a backlog of use cases prioritised by hours saved and risk (D50, I43, I44), which feeds the cases per role (D18), the intelligence map (D27), the assessment profiles and learning paths of each person (workstream A), the process catalogue (D23) and the baseline of manual handovers (I13).
• Checklists (workstream D). Each workflow and each process in the catalogue has its checklist, which is the standardise rung made concrete: steps, owner per step, inputs and outputs, definition of done, controls and where the evidence is filed. First the critical processes, then the rest; the ones that already exist are gathered in one single library (D52); AI can write the draft from a recording or a procedure, and the process owner approves it; it counts only if it was used in real work (I46, I47).

• Two readings, two instruments, workstream A. The organisation is measured with ibizai.ai’s in-depth organisation assessment; each person, with their Skaills baseline (D04), which is not duplicated. The reading of the organisation indicates which gaps each workstream addresses, with links to workstream B (use cases) and to workstream E (platforms and data governance); the reading of each person feeds their growth plan, their learning path and the map of work per role (D50).
• Two modes. The online self-assessment is free and open to anyone: quick or in-depth, for one person or for the organisation, with its result and action plan at once. The in-depth accompanied assessment is part of ibizai.ai’s consulting and solution services, which are paid: it adds the consolidation of several respondents, the check against evidence, the comparison with the sector and the report for management (D54, D80). Inside an adoption programme, it is the baseline of workstream A.
• The organisation’s instrument. 24 statements, three per dimension, on a scale of 0 to 10, across eight dimensions: strategy, data, infrastructure, talent, culture, processes, governance and return on investment; about 15 minutes. Each dimension and the total are scored from 0 to 100 (average of the answers) and fall into a level: Early Stage (0 to 34), Developing (35 to 59), Advanced (60 to 79) or Leading (80 to 100). It returns the two strengths and the two gaps, a reading and an action plan per dimension over five time frames (first week, 30, 60, 90 and 360 days), written in advance for each level, and it can be downloaded as PDF, Word, Markdown or HTML. The result stays in the ibizai.ai account of the person who answers (or in an anonymous session in their browser); ibizai.ai does not consolidate several people from one organisation. The quick organisation assessment (four dimensions) does not serve as a baseline.
• The process. (1) Guide (D53): who answers (management and the area leads), how the report is delivered, consolidation template and evidence guide per dimension. (2) Each person answers on their own, about the organisation and not about themselves, while signed in, and delivers their downloaded report in a restricted-access folder. (3) The support team consolidates by dimension: mean, lowest, highest and where the readings differ, without attributing scores to anyone. (4) In a cross-check session, management goes through each dimension with the evidence from the process mapping (§20), the platform map (§19), the process catalogue (D23) and the Skaills aggregate: the evidence sets the ceiling; the score is recorded as it came out and, if the evidence does not support it, the plan works with what the evidence shows (D54). (5) The two main gaps become work with owner and date in workstreams B and E, and the others in their own workstream; ibizai.ai’s action plan is an input, not a plan. (6) Re-measurement in month 3 and in month 9 with the same people and the same process (I48).
• Each person. The Skaills AI readiness assessment, if it is enabled for the organisation account; if not, the role-profile assessment with its readiness index. Baseline in weeks 2 and 3 and re-measurement in month 3 and in month 9 (I02, I49, I50). No client information is written in the open answers, because the synthesis is drafted by an AI model.
• For development, not for sanction. The person and Human Resources see the individual result; management and the leaders see aggregates with a minimum number of people per group; the result is not used in performance, compensation, promotions or sanctions, and any use outside its purpose is an incident (D55, I51). Each person receives in writing, before answering, what it is for, who sees what and how long it is kept. The organisation report of each person who answers is seen by that person and the support team; the consolidated report does not attribute scores.
• Two comparisons. With itself over time: the organisation is measured again with the same instrument, the same people and the same process, and the change is read (I48). With its sector, in the accompanied assessment: each dimension is read against published sector references (studies, surveys and data from regulators or industry bodies), each with its source, date and scope in the report (D80, I69); where there is no verifiable reference for a dimension, the report says so and does not estimate one.
• Method, not a validated instrument. The two assessments are self-assessments, and the comparison with the sector is a reading against published references, not a ranking or a certification.
• What to validate in each instance. Who answers the organisation assessment and how they deliver their report; in Skaills, whether the readiness assessment is enabled for the organisation, whether it is linked to it, with invitations, aggregate view and comparison between measurements, and who administers it.

The company’s Business Intelligence is the organisation’s ability to ask its own data and receive answers with their source, timely alerts and proposed actions. The programme names it as a component and as a goal: it does not add a workstream, because it is what the five workstreams build together, in layers (data sources, integration, commercial registry, search and agents, results), and this section orders those layers and adds what is missing to complete them.

Create, configure and tune. Business Intelligence is not bought ready-made: it is assembled on the tools the company already has, in three movements.

What it is and what it is not. It reads only what each person can already open; numbers come from the system that calculates them (principle 4); a person decides and signs (principle 5); nothing enters without data rules (principle 6). It is not a new data warehouse or one more dashboard: it is the company being able to consult, with judgement and with the source in view, what it already knows.

• Business outcomes. Five outcomes, each with the workstreams, deliverables and indicators that produce it: more sales (opportunities identified earlier, higher win rate, competitor tracking, more accurate register), greater efficiency (faster decisions, less time searching for information, automatic task follow-up), better client service (full view, proactive resolution, compliance with service agreements), knowledge retention (procedures, past proposals, onboarding) and new growth opportunities (trends, new accounts and verticals, data-based strategy). None is promised: it is measured.
• Catalogue of agents and advisors (D56). Executive, commercial intelligence, full view of the client, operations and service level agreements, meetings, proposals and tender documents, processes and checklists, onboarding, projects, email monitoring and conversational advisors in StrategAI: customisable advisors (persona, knowledge base, write scope) that converse about the live strategy, the plans, the accounts and the brief, and whose proposed changes a person approves, one by one; for example, one for strategy and innovation and one with the perspective of the company’s management. For each agent: what it does, where it runs, what data it reads, its owner, when it comes in and its status. Ladder rule: an agent comes in only on a standardised process and on approved sources. A person decides: no agent replies to a client, sets prices or changes dates without confirmation.
• Full view of the client (D62). The history of interactions from all captured channels, per account, in the commercial register; a relationship risk signal (sentiment analysis) only on what has already come in through authorised routes, referring to the account and never to a person, confirmed by the account owner and with prior legal validation.
• Enterprise search and knowledge graph (D57). The office suite’s assistant, grounded in the organisation’s data, searches only what each person can already open; the connectors bring content from other systems with the source permissions; the commercial register provides the business graph (accounts, contacts, opportunities, projects, commitments and people). Permissions reviewed before extending search and for each connector.
• CRM layer (D58). No CRM is bought or migrated in the first 90 days; a decision matrix is prepared with criteria and weights set before scoring, and the decision is made in the horizon with written processes. Whoever operates one of the options informs, but does not score.
• Data governance in a single view (D59). Classification, one owner per type of data, permissions, approved tools, retention, quality, processing and enrichment, registers and incidents, on one page.
• Impact and return (D60, D61). Hours freed per case in production times the loaded hourly cost of the role, opportunities recovered, compliance with service agreements, response times and programme costs, against a baseline from the first weeks; report at the close of the 90 days, in month 6 and in month 12. What could not be measured is stated; it is not estimated.
• Market intelligence (D63). Competitors, trends and new accounts and verticals with their source, in the commercial register, reviewed every quarter.
• A four-phase route, in parallel. Plan and design, connect the sources, build the agents and deploy, scale and measure are covered by workstream and in parallel within the 12 weeks, instead of in series. What the programme adds to any platform: prepared people, written processes, signed data rules, support with decision gates and no dependence on one single vendor.
• Execution plans. Each instance loads the programme into StrategAI as execution plans per workstream, with owners, dates, metrics and log, and the instance lists what each plan leaves. Visibility of execution: the list of plans shows the progress as a percentage, the owner, the dates and the overdue and blocked items of each plan, and each plan opens in its dashboard (dashboard, phases, decisions, indicators and log) with filters by person responsible, status, priority, type, phase or only what is overdue; it is reviewed every week together with the business dashboard (I61).

Value levers: how the organisation earns and saves money (D74). Group rule (cemi-system/canon/06-business-case.md, block cemi-business-case): every instance says, before the price and with the same names in the document, the quote, the deck, the summary and the video, how the organisation earns more, how it spends less and what it stops risking. Each lever carries a mechanism, a formula, a baseline (source, owner and week) and indicators; the amounts come from the Finance baseline (D74) and are read with the impact and return method (D60, D61). None is promised. Reusable catalogue for an organisation that sells, delivers and maintains technology or services:

The instance chooses the levers that apply, names them with its own vocabulary and writes their mechanism with the organisation’s facts.

The commercial register and the execution plans live in StrategAI, which also brings the tools to diagnose the business, write the strategy and put together the business plan. Each instance uses them so that management completes, reviews and maintains its strategy where it is executed. It is not a new workstream: each piece goes into the workstream that already uses it.

• Diagnosis (D64). Of the environment: PESTEL and extended Porter’s forces, with the opportunities and threats matrix and competitive intelligence (D63). Institutional: McKinsey 7S, VRIO and SWOT, with the evidence from the process mapping (§20) and from the AI readiness (§21). Draft in the first weeks; review at the first decision gate; every quarter afterwards.
• Strategic vision and business model (D65). Balanced Scorecard with indicators from §8, blue ocean with the ERRC grid, Business Model Canvas, Lean Canvas and one value proposition canvas per customer segment. The canvases and the segments can be drafted with AI from the strategy’s data, with approval block by block; what has no data (for example, the cost structure without recorded expenses) remains “To be completed” and goes to the list of facts to validate.
• Strategic Analysis Document (D73). The readable edition of D64 and D65, generated from the strategic register, with an annex of the key questions and management’s answers; it makes visible what is approved and what is still in draft, and it is renewed after each decision gate.
• Commercial strategy (D66). Value proposition per sector and per product, strategies per sector and for marketing, and roadmap with its target.
• Scenarios, simulator and dashboard (D67). The simulator per revenue source and the scenarios are calibrated with the company’s own values and are read every month against the revenue target and against what was closed; the business dashboard (key indicators, opportunities per stage, funnel, won and lost, sales velocity and progress towards the target) is looked at in the weekly meeting together with the progress of the execution plans (I61): it is the visibility of the commercial strategy.
• Prospecting, sales and templates (D68, D69). Stages of the sales process adjusted to how clients buy, prospecting flow, objections with their answer and a library of templates (email, call, messaging, proposal follow-up, meeting preparation, letters and answers to objections), in the service of the commercial guarantees (§17).
• Three rules. AI proposes drafts and the owner of each piece decides; amber information goes in at most; what comes from a generic template is replaced with the company’s own before it is used to decide.

When management answers the diagnosis questions (§23) and the diagnosis is loaded, the instance adds a section that goes through its findings in a table: for each one, what the programme does, where (workstream, plan, deliverable, week) and whether it is new or was already covered. It is a working method, not a validated framework.

• Sources. Management’s answers are facts on the client’s side and are cited as such. What AI drafts from them is a draft: statements about the market or the regulator are verified against their primary source before being used, and the figures it proposes are not targets until management approves them with their baseline.
• What is new is small and measurable. Deliverables or indicators are added only where something is truly missing. The most frequent: measuring client satisfaction (D70, I62), meeting project deadlines (I63), reading on the dashboard the measures of success that management names (I64), analysing the growth ideas as options, not as commitments (D71), and watching a regulatory change that management flags (D72).
• The scope does not change without a decision. What is left outside the programme (the structure, hiring, new lines of business) is named and left to management, with the evidence the programme produces. If the organisation has populations other than the office, such as field staff, the instance decides explicitly through which routes they come in (onboarding, acknowledgements, capture) and through which they do not, and every extension remains as a validation with its decision gate.

ibizai.ai maintains a library of lessons from leaders (docs/research/lecciones-lideres/biblioteca.md): concrete practices of large companies and the available evidence on AI at work, each one verified against its primary document, with its result cited as its author publishes it, its transferability to a small team and its warning; and a list of legends and figures that are not used. The library grows through research rounds verified source by source (canon 02 of journalaism-system); nothing goes in without its source.

• In the instance. A section takes the most transferable practices and says, for each one, who documented it and how the client applies it: workstream, deliverable, indicator and week. It also adds two or three warnings about what went wrong. It creates no deliverables or indicators: it refers to the ones that already exist.
• It aligns, it does not prove. A practice aligns with a decision of the programme; it is never said that it proves it or that the programme is founded on it. The figures a company publishes about itself are cited with their scope and are not used as targets.
• In the academy. The records are supporting material for the Learning Units, with the source always visible, to discuss what the team would do with that practice, not to imitate the company that tells it.
• Vendor guides, in their place. The library admits, in a separate section (§2.11 of the library), guides published by software vendors when their ideas are transferable. They are read in full, recorded as what they are (a vendor’s material, with no published results) and taken in the programme’s terms and measurement, never in theirs. The first is Atlassian’s guide on the judicious use of AI in teams (record G-01), with which Management’s five questions (§3), the three traps with a name of their own (§12), the context folder (D77) and indicators I65 and I66 align.
• In StrategAI. The companies in the library can be loaded into the References and Benchmarks module of the client’s strategy, with their verified practices as referenceable strengths and their application in the strategy.

The kick-off charter is the document that the client’s management signs in week 1 to mark the start of the programme. It adds no commitments: it gathers, in some four to six pages, what the instance already defines, with a reference to the section of the instance for each piece of data. If the charter and the instance differ, the instance prevails and the charter is corrected in its next version. It is drafted in week 1 as a draft for approval and is signed before the first decision gate, ideally at the first weekly meeting.

Rules.

• Only what is already written. Each piece of data comes from the instance, from the plans that depend on it or from the commercial proposal; what is not confirmed is marked “to be validated in week 1”. No new fact, figure or commitment.
• Roles, not people. On the client’s side and on CEMI’s side, roles are named; who holds each one lives in StrategAI. The signature is also by role.
• No amounts in the publishable version. The investment, the terms and the payment schedule are cited by the number of the commercial proposal, never with figures. The money levers go by name, without amounts (§22).
• Targets as targets. The targets are proposed targets, with the week in which they are confirmed.
• Versioning. Each change approved through change control produces a new version of the charter with its change note.

Fixed sections.

The canon is one. The service is adjusted to each organisation in its instance (§13) and, before that, it arrives tuned for the scenarios and sectors where CEMI already works. Tuned means the starting point already carries the vocabulary, the data rules, the use cases and the indicators of that scenario or sector; it does not mean there are measured results that can be promised to another organisation. Each instance measures its own.

Three scenarios. What does not change in any of them: the principles (§1), data rules before first use (§4), the process ladder, measurement with evidence, and that a person decides and signs.

Eight sectors with prior CEMI work.

A sector that is not on the list is served with the same canon and without prior tuning: the instance builds it.

The shared archive (D11, D51) is where the team’s work lives and what the AI assistant can consult. Three groups of actions make it possible, in this order: which platform it lives on, how what already exists is organised, and how it is presented to each area. They belong to front E (§19) together with front C (§11), and they come before the AI assistant is opened on the files: the assistant finds everything each person can open, organised or not.

28.1 The platform that fits

The platform is chosen for its fit with the organisation, not out of habit (D81).

• If the organisation already works in an ecosystem (for example, Microsoft 365 or Google Workspace, with licences that include shared storage), that ecosystem is confirmed and nothing is bought. The analysis is short: what each licence includes, who administers it, and whether the approved AI assistant (§4) works on that platform.
• If it has none, or mixes several, the options are compared with the criteria below and a single platform is chosen as the system of record for documents. The others are retired by a date, or kept only for what they do best (for example, exchanging files with a client who works on another one), always linked and never as a second copy.

The decision fits on one page: options considered, score per criterion, platform chosen, what is retired and when, and who reviews it. Management approves it and IT Support operates it.

28.2 Organising the archive that already exists

Before moving, it is organised; and it is organised so that everything can be checked and undone (D82).

1. Inventory. Every file with its path, size and date, before anything is touched. Names are read, not contents.
2. Plan for management. Few top-level folders; one folder per client with all its history; one per opportunity or project, with fixed subfolders; the naming convention (§4.3); and the decisions needed, each with its recommendation.
3. Records that allow undoing. One record per type of change (rename, move, delete), with date, before and after, in a permanent folder within the library itself, never in a temporary folder.
4. Names first, in place. Renaming is not copying: the file keeps its history and its internal links. First the objective fixes (spaces, dots, symbols that fail on sync); then long names, to the agreed format, each proposal validated before it is applied.
5. Links. An index of which documents link to others, before anything changes; after each step, what broke is recalculated and fixed or flagged.
6. Verified move. Every move checked; at the end, everything is reconciled against the inventory and no file is missing.
7. Duplicates and empty files, with a decision. Copies are compared byte by byte; empty files are resynchronised before being given up as lost; nothing is deleted without a management decision, and what is deleted stays in the recycle bin.
8. Notice to the team. Before the archive moves to its platform: where everything is, which links may fail, and that sharing links are recreated.
9. Server-side move. Between sites of the same platform, with its own move function, which keeps the history and neither downloads nor re-uploads anything.

It is measured with I70.

28.3 Sites, teams and pages

Few spaces, by area and by workflow, and one front door for everyone (D83).

• A home site for the organisation: notices, news, links to each team and to the applications, policies and help.
• One team per area or per workflow. In Microsoft 365, each Teams team comes with its SharePoint site; in Google Workspace, a shared drive with its Chat space. Channels only where there is work to discuss; what is confidential, in a private channel or folder with the members management names.
• A designed home page for each site, not the default one:

• Those who work in the field or with a more limited licence. What their licence includes is checked (teams, libraries, phone app, offline work) and what they really need in their working day is decided (procedures, checklists, technical guides, service orders). It is put within their reach through the means their licence allows: a read-only channel or library, or the field application they already use. Nobody is left without the information they need to work, and nobody gets access to what they do not need.
• Rules. Spaces are created by IT Support; each has two named owners; the page is maintained by its owner; at each decision gate, which space is used and which is surplus is reviewed (I71).