i2Business — Consulting & Training : Intelligence to your business
Home/AI services/Strategy and roadmap

AI strategy and roadmap

A clear AI strategy and a 12- to 24-month roadmap

An AI strategy answers four questions: where AI creates value for you, what architecture it runs on, which models and vendors it uses, and in what order and with what budget to deploy it.

Build my roadmap Measure my AI maturity

Updated October 11, 2026

A trainer talking with participants during an in-room workshop

The reality

AI budgets are rising: better to manage them

5 → 17%

AI's share of the IT budget, more than tripled according to the BCG Applied AI 2026 survey (respondent estimates).

Source: BCG, October 2026

2.8%

of revenue spent on average on AI initiatives funded outside the IT budget: without an overall plan, spending gets scattered.

Source: BCG, October 2026

86% / 19%

of Tunisian companies consider technology essential to innovation, yet 19% report having innovated: a plan closes the gap.

Source: ITCEQ, May 2026 (via WMC/TAP)

The frameworks

Three value plays, six dimensions to manage together

Two well-known frameworks structure our work. They are aids to thinking, not proof of effectiveness.

What kind of value?

Deploy, reshape, invent

  1. 1. DeployApply AI to your current operations: meeting notes, writing, support, coding assistance.
  2. 2. ReshapeRethink processes and ways of working around AI.
  3. 3. InventCreate new products, services or business models. The rarest and most ambitious play.

BCG framework. Sources: BCG · BCG Executive Perspectives.

What capabilities?

Six dimensions to manage together

  • 1Strategy
  • 2Talent
  • 3Operating model
  • 4Technology
  • 5Data
  • 6Adoption and scaling

McKinsey's “Rewired” framework: a transformation stalls if you only rebuild the tools. McKinsey

See change management

Prioritize

Pick the right use cases, in the right order

You probably have more ideas than resources. We score each use case on four criteria, then decide: fund, put on hold or drop.

  • Business value
  • Technical feasibility
  • Data maturity
  • Risk level

A high-value use case can be a trap if the data is poor or if an error has regulatory consequences: that's why we add these criteria to the basic value-versus-effort grid.

Sources: GetWidget · CIO Index. A consulting method, not a standard.

Prioritize my use cases
Value ↑
Quick winsHigh value, low effort: start here.
Strategic betsHigh value, high effort: plan carefully.
DeferLow value, low effort: later, if time allows.
DropLow value, high effort: let it go.

Effort →

The roadmap

Five workstreams moving together over three horizons

A 12- to 24-month roadmap is more than a list of projects: it coordinates use cases, technology, people, compliance and budget. A sample template, to adapt to your situation.

Use casesHorizon 1: Quick wins and prototypesHorizon 2: Pilots and first production releasesHorizon 3: Scaling, new uses
Data and technologyHorizon 1: Baseline and minimum foundationHorizon 2: System connections, first agentsHorizon 3: Stable shared architecture
Skills and changeHorizon 1: Awareness, sponsor appointedHorizon 2: Role-based training, internal championsHorizon 3: Embedding and community of practice
Governance and complianceHorizon 1: Usage rules, inventoryHorizon 2: Risk tracking, training (EU AI Act)Horizon 3: Audit, ISO/IEC 42001 readiness
BudgetHorizon 1: Prototype budgetHorizon 2: Budget lines per use case, operating costsHorizon 3: Value-based management, cost per operation

Typical benchmarks: first pilot results in 6 to 12 months, impact at scale in 12 to 24 months (single source, to adapt). CIO Index. Budgets are defined with you: we don't put forward a standard figure.

Build mine with a consultant

Target architecture

A shared foundation, not a pile of projects

Defining the target architecture up front means you don't rebuild for every new use case. It is organized in layers, supported by a cross-cutting foundation.

Use casesThe apps and assistants your teams use.
OrchestrationAgents and search across your documents.
System connectionsMCP and APIs to your ERP and CRM.
ModelsOne or more models, chosen per task.

Cross-cutting foundation

  • Data and quality
  • Security and access rights
  • Governance and compliance
  • Monitoring, costs and quality
See production monitoring

Models and vendors

Choose by testing, not by leaderboard

The best model is the one that meets your quality bar at the lowest cost and risk. We compare several options on your use cases and plan for the ability to switch.

Fit with your tasksWe test models on your own use cases: a public leaderboard doesn't tell you what works for you.
CostPay-per-use or dedicated hosting, batch processing for non-urgent work, cost per operation.
Speed and throughputResponse time, rate limits, predictability at peak hours.
Data and hostingWhere your data is processed, under which agreements, with which logs: to check in the vendor's documentation.
PortabilityYour prompts, tests and connections stay with you so you can switch models.
ComplianceEU AI Act, GDPR and Tunisian law built in from the selection stage, not after.

Criteria drawn from practitioner guides: RaftLabs · Interview Kickstart · ModelCompare. Pricing and rankings change fast: we check them at the time of selection.

Test on a rapid prototype

A Tunisia-specific point

Sending data abroad: what the law says

Many AI models are hosted outside Tunisia. If personal data is sent to them, loi organique n° 2004-63 (Tunisia's Organic Law No. 2004-63 on personal data protection) sets out specific rules. They weigh on vendor selection.

  • Art. 51The destination country must ensure an adequate level of protection.
  • Art. 52Authorization from the INPDP is mandatory for any transfer abroad; it rules within one month at most.
  • Art. 90A transfer without authorization is punishable by one year of imprisonment and a 5,000-dinar fine.

Sending personal data to a service hosted abroad may constitute a transfer, depending on what is sent and how. We build this question into vendor selection.

General information, not legal advice: have your choices validated by a lawyer. A reform of the law is under discussion; its adoption is not confirmed. Source: loi organique n° 2004-63 (INS).

See AI governanceTalk to a consultant
Illustration: two employees at a table reviewing a contract and a diagram of data flows between a server, the cloud and several countries

Our approach

From baseline to roadmap

  1. 01

    Assess

    Company maturity, data, skills, constraints.

  2. 02

    Set the ambition

    What kind of value to target: deploy, reshape, invent.

  3. 03

    Prioritize

    Score use cases and decide: fund, wait, drop.

  4. 04

    Design and select

    Target architecture, models and vendors tested on your use cases.

  5. 05

    Plan

    A 12- to 24-month roadmap: milestones, budget, governance.

What you get

  • A three-horizon roadmap
  • A target architecture
  • A shortlist of models and vendors
  • A budget framework
Start my roadmap See all our services
Illustration: employees placing sticky notes on a whiteboard split into four quadrants by impact and effort, to prioritize use cases

FAQ

AI strategy and roadmap: your questions

What is an enterprise AI strategy?

It is a plan that sets out where AI creates value for you, what architecture and models it relies on, in what order and with what budget to deploy it, and how to manage risk and change.

Why a 12- to 24-month roadmap?

Because first pilot results typically arrive within 6 to 12 months and impact at scale within 12 to 24 months. This range is a benchmark, to adapt to your situation.

How do you choose priority use cases?

By scoring each use case on value, technical feasibility, data maturity and risk, then deciding to fund, put on hold or drop it. Quick wins come first.

Should you pick a single model vendor?

Not necessarily. It's better to test several models on your own use cases, keep your prompts, tests and connections in-house, and plan to be able to switch or combine models.

Can I send personal data to an AI hosted abroad?

It may constitute a transfer under Tunisian law 2004-63, which requires INPDP authorization and an adequate level of protection. The answer depends on what is sent: have your case validated by a lawyer.

What budget should I plan for?

It depends on your use cases. Cost items to plan for are models and APIs, data, integration, skills and change, governance and monitoring. We build the budget framework with you, without a standard figure.

Go further

Related services

AI Maturity & Opportunity AssessmentData assessment, process mapping and identification of high-ROI use casesLearn more Rapid Prototypes (PoC)Working proofs of concept built and validated on your real data in 4 to 8 weeksLearn more Fractional Chief AI Officer (CAIO)Ongoing strategic guidance and executive leadership of your AI transition, on a recurring subscriptionLearn more Governance, Compliance & AI ActRegulatory compliance (EU AI Act, GDPR), risk management and ISO 42001 readinessLearn more

Move from intent to a plan

Let's talk about your priorities: we'll build a realistic, costed roadmap with you.

Talk to a consultantMeasure my AI maturity