Data readiness · solution selection · implementation · adoption

Prepare the data, select the right AI and teach the team to use it effectively.

I help move from an idea to a solution with a clear objective, credible data and quality criteria. I can lead selection or represent the business during delivery by an internal team or technology supplier.

I do not start with the model. I compare built-in AI, enterprise assistants, API solutions, RAG and automation without AI.

Describe the process and data Estimate AI potential
Data ready for usescope, quality, access, owners and evaluation set
A solution matched to the processquality, risk, integration and total-cost comparison
A team ready to workroles, rules, practice, playbook and adoption measures

Common starting point

The organisation experiments with AI but has no shared path from trial to daily work.

The hardest problems rarely come from model access. They sit between data, process, responsibility, security and user capability.

Data is not ready

Sources lack owners, documentation is stale and knowledge remains dispersed.

Selection begins with a supplier

Models and demos are compared before the process, quality criteria and limits are agreed.

The demo has no acceptance test

A few impressive answers replace representative cases, quality metrics and a go/no-go decision.

Business and technology speak differently

A supplier builds features while users cannot translate needs into data and acceptance scenarios.

Rules appear after the pilot

Confidential data, retention, access, human oversight and logging are considered just before launch.

Training ends with prompting

The team knows techniques but not when to trust, verify or escalate an answer.

Four delivery streams

Technology is one part of the change, not the whole project.

Each stream has its own outcome and closure criteria. The scope can cover all four or fill a gap in an existing programme.

01 · Data

AI data readiness

Inventory, quality, access, classification, metadata, knowledge and test cases.

Outcome: data preparation plan
02 · Selection

Solution selection

Options compared across quality, security, integration, cost, maintenance and flexibility.

Outcome: scorecard and recommendation
03 · Implementation

Business assurance

Requirements, backlog, supplier communication, acceptance tests, go/no-go decisions and risk.

Outcome: solution accepted by the business
04 · Team

AI adoption

Roles, usage rules, scenarios, exercises, human oversight, champions and measurement.

Outcome: playbook and adoption plan
Problem before toolselection starts with the process and business outcome
Test before scalequality and risk are measured on representative cases
Practice before declarationadoption means changed behaviour, not training attendance

Data readiness

AI needs more than access to data: it needs context and a definition of a correct answer.

Readiness is assessed against the use case. A knowledge assistant, a prediction and automated decision-making require different sources, controls and tests.

  1. 01

    Scope and system of record

    Required data, its source and the reference definition.

    Sources · grain · freshness
  2. 02

    Quality and completeness

    Gaps, inconsistencies, duplicates, stale knowledge and refusal cases.

    Quality profile · exceptions
  3. 03

    Access and sensitivity

    Personal data, trade secrets, user roles, retention and supplier use.

    Classification · access control
  4. 04

    Context and metadata

    Glossary, owners, document source, version, relationships and interpretation rules.

    Metadata · provenance
  5. 05

    Evaluation set

    Representative questions, expected answers, difficult cases, prohibited data and expert criteria.

    Golden set · evaluation rubric

Solution selection

The best answer does not always require a custom application with the newest model.

I compare the whole solution: user experience, process integration, data access, control, quality, implementation cost and maintenance.

Solution classes considered during AI selection
OptionWhen to considerPrimary conditionTypical trade-off
AI in an existing systemThe process already runs in CRM, BI, helpdesk or an office suite.The feature covers the use case and respects permissions.Less flexibility, faster adoption and simpler maintenance.
Enterprise assistantThe team needs secure support for text, files and daily tasks.Identity, data policy, tenant configuration and usage rules.Broad use, limited process specialisation.
Knowledge assistant or RAGAnswers must use a controlled document and source base.Content quality, metadata, index refresh, citations and retrieval tests.More knowledge control, a pipeline to maintain.
API workflow or agentAI performs process steps and uses company systems or tools.Limited permissions, validation, logs, human approval and error handling.Highest automation and operational responsibility.
Automation without AIThe task has fixed rules and no unstructured interpretation.The process and exceptions can be specified clearly.Less flexibility, predictable cost and outcome.
Quality on company dataSecurity and privacyProcess integrationTotal costEvaluabilityMaintenance and exit

Consultant during delivery

I ensure the supplier builds against the process, data and business criteria.

I can support delivery without replacing a software house, translating needs between sponsor, users, data, security and the technical team.

Before launch

Requirements and pilot charter

Objective, scope, users, data, risk, metrics, constraints and responsibilities.

During build

Backlog and decisions

Features, priorities, changes, integrations and consequences for users and data.

Before release

Evaluation and acceptance

Golden set, expert tests, security, cost, latency, failure modes and go/no-go.

After release

Monitoring and development

Feedback, incidents, quality, cost, adoption, model changes and controlled backlog.

Team readiness

Effective AI use requires different capabilities for users, managers and system owners.

Users

Work on real tasks

Task formulation, context, result verification, sources, confidential data and escalation.

Managers

Design the way of working

Task choice, human oversight, quality, responsibility, effect measurement and learning.

Owners and administrators

Rules and control

Approved tools, access, logging, retention, use register, incidents and change cycle.

AI champions

Support close to the team

Example library, office hours, feedback, feature testing and good-practice exchange.

Adoption programmerole-based trainingusage and verification playbookscenario librarychampions and office-hours planusage and outcome measures

Engagement process

From deciding whether AI makes sense to controlled use by the team.

  1. 01

    Process and use-case map

    Tasks, audiences, scale, business pain, frequency, risk and non-AI alternative.

    Outcome: use-case priorities
  2. 02

    Data and organisational readiness

    Sources, quality, access, owners, policies, capabilities and technical dependencies.

    Outcome: readiness assessment
  3. 03

    Solution selection

    Demo scenario, scorecard, options, total cost and supplier risk.

    Outcome: recommendation and architecture
  4. 04

    Pilot and evaluation

    Prototype on representative data, golden set and quality, risk, cost and UX tests.

    Outcome: go/no-go decision
  5. 05

    Implementation and acceptance

    Backlog, integrations, security, monitoring, failure scenarios, documentation and tests.

    Outcome: solution ready for use
  6. 06

    Adoption and improvement

    Role training, playbook, champions, feedback, effect measurement and controlled development.

    Outcome: AI embedded in work

Tangible outcome

A decision, test and control package that lets the project proceed safely.

Document scope follows the stage. Not every project needs every item, but each decision needs an owner and evidence.

01AI readiness assessment

Data, process, capabilities, policies, risk and gaps.

02Solution scorecard

Criteria, demo scenario, comparison results and recommendation.

03Pilot charter

Goal, scope, users, data, metrics, roles and exit conditions.

04Evaluation pack

Golden set, rubric, results, errors and go/no-go.

05Risk and control register

Risks, controls, owners, monitoring and incident path.

06Adoption playbook

Usage rules, roles, scenarios, verification and capability development.

Acceptance criteria

AI must be useful, measurable and controllable in the target process.

There is no universal quality threshold. Metrics and tolerances follow the use case, audience and consequence of error.

Business effect has a baselinetime, cost, quality or result is compared with the current process
Quality passes the golden setrepresentative, difficult and prohibited cases have expert-reviewed results
Data follows agreed rulesaccess, retention, logging and suppliers are documented
Failure modes are knownthe system can refuse, identify missing data and escalate to a human
Cost and performance stay within limitslatency, volume, usage cost and limits are measured
The team can use and verify outputusers complete practical scenarios and feedback has an owner

Honest limitations

What I do not promise before understanding the process and data.

  • AI will not fix ungoverned data.It may accelerate access to chaos but cannot create a credible definition without an owner.
  • A PoC does not prove production readiness.Integrations, access, monitoring, cost and failure appear in the target environment.
  • The best model depends on the task.A generic ranking cannot replace testing on company data, language and constraints.
  • One training session does not ensure adoption.Scenarios, manager support, feedback and time are required.
  • AI quality changes over time.Models, data, prompts, integrations and user behaviour require observation and regression tests.
  • Consulting does not replace legal assessment.Regulated or high-risk use cases require the appropriate legal, security and compliance roles.
Krzysztof Surowiecki

Data, analytics and organisational change

I connect business, data and the real work of users.

I support organisations with data, reporting, automation and AI. That makes it possible to assess not only model capability but source quality, process integration, acceptance and team readiness.

20+years of experience
150+analytics and data projects
50+clients

Figures refer to the broader analytics, data and digital-transformation practice, not only production AI implementations.

Before we talk

Frequently asked questions.

Where should an organisation start with AI?

Start with one process and a defined outcome, then assess data, users, risk, the current alternative and effect measurement before comparing solutions.

What does AI data readiness mean?

It covers systems of record, scope, quality, freshness, ownership, access and sensitivity. Knowledge assistants also need metadata, versioning, retrieval tests and an evaluation set.

Do you select a specific AI model?

I can compare models, but the recommendation covers the whole solution. Built-in AI, an enterprise assistant or automation without an LLM may be the better choice.

Can you support a project delivered by another supplier?

Yes. I can represent the business by refining requirements, leading the scorecard, reviewing backlog, preparing acceptance tests and supporting the go/no-go decision.

How do you prepare the team?

Training uses real tasks and roles. Users practise formulation and verification, managers design oversight and measures, and owners work on policy, access and incidents.

Does the programme include AI literacy?

Yes. It covers opportunities, limitations, potential harm, responsible use, verification and accountability, matched to the users' knowledge and context.

First step

Describe the process, users and data intended for the AI solution.

I will follow up on the business outcome, current alternative, data quality, risk, technology supplier and team readiness.

Discuss AI implementation Estimate AI potential

Prepare the decision

Tools for an initial idea assessment.

AI potential calculatorteam time, process cost and potential effectEstimate potential AI solution selectorrequirements, data, compliance and integrationsCompare options AI models hubcapabilities, providers and comparison resourcesExplore models