AI Services

AI adoption for organisations

Strategy, policy, training, and systems. We take an organisation from scattered individual AI use to a governed programme: a written adoption plan, a use policy staff can follow, training for every team, and workflows in production with human review built in.

Adoption plan and AI use policy written from a staff usage survey
Training delivered inside each team's standing meeting, leadership edition included
Workflows where the model proposes, a person decides, and the record shows why

Discuss an AI programme See the four service lines

Today: individual use

Chat assistant

personal account

Translation

unknown data rules

Summaries

no verification step

Spreadsheet helper

mixed accounts

Drafting

no disclosure

Programme: governed use

1
Adoption plan
written, three years
2
AI use policy
adopted, organisation-wide
3
Training
every team, leadership edition
4
Workflows
in production, reviewed

Human review share

per workflow

100%

70%

40%

What we deliver

Four service lines, taken together or one at a time

Most organisations already use AI; the gap is usually confidence about what data is safe to share and which uses are worth the effort. Each line below closes part of that gap and can be commissioned on its own.

AI strategy and adoption roadmap

A multi-year AI adoption plan written for your organisation: where AI is already used, where it should be, and in what order.

Staff AI usage survey: tools in use, account types, data entered, how outputs are checked
Use-case inventory ranked by operational value, data sensitivity, and effort
Three-year roadmap with owners, budget lines, and measurable milestones
Platform choice: which model providers, where data is processed, what is excluded

Deliverable

Adoption plan and prioritised use-case list

AI use policy and governance

A policy staff can apply on a normal working day, written from how they already use AI, and adopted as the organisation-wide norm.

Approved tools and account rules, including no-training guarantees from approved providers
Green / Amber / Red data classification with a step-by-step checkpoint for Amber data
Human verification of every output, disclosure of AI use, handling of identifiable records
An AI practice group that meets on a fixed cadence and decides which new uses go ahead

Deliverable

AI use policy, data classification, practice group charter

AI training for staff and leadership

Practical sessions delivered inside each team's standing meeting, so the whole team attends rather than volunteers.

All-staff session (90 minutes): the rules, the data classification, prompt anatomy, hallucination cases, scenario quiz
Leadership edition: approval decisions, risk, budget, and what to ask of vendors
Champions track: one person per team who keeps the prompt library and answers first questions
Recording, slides, and policy text sent to anyone who missed a session

Deliverable

Trained teams, session materials, champions in place

AI workflows, agents, and ERP integration

Production workflows where the model proposes, a person decides, and the record shows why.

Opportunity and document screening over portal feeds, with an analyst deciding on each hit
Transcription and translation of field recordings with reviewer sign-off before use
Retrieval-based knowledge platforms that answer with citations to your own documents
AI inside the ERP: applicant analysis, document extraction, exception handling, approvals

Deliverable

Working integrations with review gates and audit logs

The programme

How an adoption programme runs

Five phases, each with a defined output. The order matters: the survey informs the plan and the policy, the training explains both, and the first workflows go live under rules staff already know.

1
Survey
Measure current use
A short staff survey establishes which tools are used, on which accounts, with what data, and how outputs are checked. The plan and the policy are written from these answers.
↳ Baseline report
2
Plan
Adoption plan and policy
The roadmap and the AI use policy are drafted together, so every approved use has a data rule, an owner, and a place in the sequence.
↳ Adoption plan, AI use policy
3
Train
Staff and leadership sessions
Sessions run inside existing team meetings over two to three weeks. Materials stay with the organisation for new joiners.
↳ All staff trained
4
Build
First workflows in production
The highest-value use cases are built with review gates and audit logs, and measured against the result the manual process used to give.
↳ Workflows live, fully reviewed
5
Operate
Practice group and review ladder
The practice group approves new uses on a fixed cadence. Review shares step down as measured accuracy holds, per workflow.
↳ Governed, scaling programme

Human review

Review starts at 100 percent and steps down as measured accuracy holds

Every workflow we put into production launches with a person checking every output against the result the manual process used to give. Reviewer feedback becomes explicit rules, and the share under review steps down only when the measured accuracy supports it.

The steps and their pace are set per workflow by its complexity and the sensitivity of the data. The end state is a sampled check, not a person on every item, and the measurement never stops.

Share of outputs reviewed by a person, per workflow

1Launch: Every output checked against the manual result
2First iteration: Prompts and rules corrected from reviewer feedback
3Accuracy holds: Review focused on the categories that still fail
4Steady state: Sampled check. The measurement stays; the person leaves routine items

Training curriculum

What a staff AI training session covers

Ninety minutes, delivered remotely or in person, inside the team's own meeting slot. The examples come from the organisation's work, and the policy is taught as part of the session rather than circulated afterwards.

Module	Audience	Content
Five golden rules	All staff	The rules that cover most daily decisions about AI use
Green / Amber / Red	All staff	What data goes where, and the Amber checkpoint step by step
Prompt anatomy	All staff	Role, context, task, format, constraints; six techniques; a weak-to-strong rewrite
Hallucination cases	All staff	Real failures, how they were caught, and the verification habit
Scenario quiz	All staff	Six situations from the organisation's own work, decided as a group
Leadership edition	Senior management	Approval decisions, vendor questions, risk register, budget lines
Champions	One per team	Prompt library upkeep, first-line questions, feedback to the practice group

Format

Inside the team's standing meeting, so attendance is the whole team

After the session

Recording, slides, and policy text go to everyone, including absentees and new joiners

Follow-through

Champions keep the prompt library; the practice group reviews new uses

Systems

Where AI sits in the operation

AI workflows sit between the existing systems and the teams. They read, classify the data, draft a proposal, and route it by the policy: Green runs, Amber waits for a person, Red never enters.

Existing systems

ERP

CRM

Documents and SOPs

Portal and mail feeds

Field recordings

AI workflow layer

1
Reads the record, the document, or the feed item
2
Classifies the data by Green, Amber, or Red before anything leaves
3
Retrieves the relevant policy or reference passages
4
Drafts a proposal with its evidence
5
Routes by data class and the rule for that step

Routed by data class

Green

Public or internal data, low risk

Runs, result logged, sampled check

Amber

Sensitive data or a consequential action

Checkpoint, then a person decides

Red

Identifiable records, restricted sources

Does not enter the workflow

Every route writes an audit entry: input, data class, proposal, decision, who, when.

Retrieval and agents

Answers grounded in your own documents, with citations

Retrieval-augmented generation gives a model controlled access to your policies, contracts, reports, and procedures. Each answer cites the passages it used, so a reviewer can check it, and the quality is tracked in an evaluation loop with the people who know the subject.

Agent workflows use the same retrieval for grounded reasoning, while approval rules decide which actions run on their own and which wait for a person.

A grounded answer, as the reviewer sees it

What does our travel policy allow for field per diems when a trip crosses a month end?

The per diem is claimed per night at the rate of the country where the night is spent¹. A trip that crosses a month end is split across the two monthly claims, each with its own approval². Nothing in the policy allows a flat rate for the whole trip.

Sources used

1 Travel and expenses policy, section 4.2, per diem basis
2 Finance manual 2025, monthly claims, page 12

Evaluation loop

1Reviewer reads

the answer next to the cited passages

2Feedback

what was missing, wrong, or over-stated

3Rule written

an explicit instruction, not a one-off fix

4Next version

measured on the same questions

Step 4 feeds step 1 until the measured quality holds; the loop runs with the people who know the subject.

Agent workflow: the model proposes, a person decides, the record shows why

One applicant screening, as it runs inside the ERP. The same pattern applies to opportunity screening, document extraction, and approvals.

The model proposes

Applicant 0418, Field Officer, Syria

Meets: four of five stated requirements
Gap: French, stated at basic level
Evidence: CV page 1, job description section 3

Draft only. No score shown as a verdict; the rule that ranks it is fixed and written down.

A person decides

HR officer, recruitment panel

ShortlistReturn

Reason

Shortlisted. French at basic level is acceptable for this post; the panel will test it at interview.

The record shows why

09:14 proposal drafted · model · inputs: CV, JD
09:14 data class: Amber · checkpoint passed
09:31 decision: shortlist · HR officer
09:31 reason recorded · visible to the panel

Every item can be traced back to who decided, on what evidence, and under which rule.

Related services

Connect AI to the systems that run the organisation.

AI workflows hold up when they are grounded in reliable records: the ERP, the integration layer, and the custom software that executes the work.

ERPERP ImplementationPut AI inside the system of record: applicant analysis, document extraction, and approvals in the ERP.SoftwareCustom Software DevelopmentBuild the applications, APIs, and orchestration layers that AI workflows need to execute work.WebWebsite DevelopmentSupport AI-enabled customer experiences with fast, reliable web platforms and support workflows.

Related insights

Review and hardening of AI-generated code before production

Teams that build with AI coding tools move fast and often reach a working prototype without the controls a production system needs. We review and harden that code before go-live: architecture, security, dependencies, observability, and release quality.

The output is a system the organisation can operate and maintain, with the fallback paths and runbooks written down.

Security and dependency risk checks

Code quality and maintainability review

Performance and scalability assessment

Production deployment readiness validation

Prompt and model behavior traceability checks

Operational runbooks, fallback paths, and incident playbooks

Discuss an AI programme Explore all services

AI Services

AI adoption for organisations

Adoption plan and AI use policy written from a staff usage survey
Training delivered inside each team's standing meeting, leadership edition included
Workflows where the model proposes, a person decides, and the record shows why

Discuss an AI programme See the four service lines

Today: individual use

Chat assistant

personal account

Translation

unknown data rules

Summaries

no verification step

Spreadsheet helper

mixed accounts

Drafting

no disclosure

Programme: governed use

1
Adoption plan
written, three years
2
AI use policy
adopted, organisation-wide
3
Training
every team, leadership edition
4
Workflows
in production, reviewed

Human review share

per workflow

100%

70%

40%

What we deliver

Four service lines, taken together or one at a time

AI strategy and adoption roadmap

A multi-year AI adoption plan written for your organisation: where AI is already used, where it should be, and in what order.

Staff AI usage survey: tools in use, account types, data entered, how outputs are checked
Use-case inventory ranked by operational value, data sensitivity, and effort
Three-year roadmap with owners, budget lines, and measurable milestones
Platform choice: which model providers, where data is processed, what is excluded

Deliverable

Adoption plan and prioritised use-case list

AI use policy and governance

A policy staff can apply on a normal working day, written from how they already use AI, and adopted as the organisation-wide norm.

Approved tools and account rules, including no-training guarantees from approved providers
Green / Amber / Red data classification with a step-by-step checkpoint for Amber data
Human verification of every output, disclosure of AI use, handling of identifiable records
An AI practice group that meets on a fixed cadence and decides which new uses go ahead

Deliverable

AI use policy, data classification, practice group charter

AI training for staff and leadership

Practical sessions delivered inside each team's standing meeting, so the whole team attends rather than volunteers.

All-staff session (90 minutes): the rules, the data classification, prompt anatomy, hallucination cases, scenario quiz
Leadership edition: approval decisions, risk, budget, and what to ask of vendors
Champions track: one person per team who keeps the prompt library and answers first questions
Recording, slides, and policy text sent to anyone who missed a session

Deliverable

Trained teams, session materials, champions in place

AI workflows, agents, and ERP integration

Production workflows where the model proposes, a person decides, and the record shows why.

Opportunity and document screening over portal feeds, with an analyst deciding on each hit
Transcription and translation of field recordings with reviewer sign-off before use
Retrieval-based knowledge platforms that answer with citations to your own documents
AI inside the ERP: applicant analysis, document extraction, exception handling, approvals

Deliverable

Working integrations with review gates and audit logs

The programme

How an adoption programme runs

Five phases, each with a defined output. The order matters: the survey informs the plan and the policy, the training explains both, and the first workflows go live under rules staff already know.

1
Survey
Measure current use
A short staff survey establishes which tools are used, on which accounts, with what data, and how outputs are checked. The plan and the policy are written from these answers.
↳ Baseline report
2
Plan
Adoption plan and policy
The roadmap and the AI use policy are drafted together, so every approved use has a data rule, an owner, and a place in the sequence.
↳ Adoption plan, AI use policy
3
Train
Staff and leadership sessions
Sessions run inside existing team meetings over two to three weeks. Materials stay with the organisation for new joiners.
↳ All staff trained
4
Build
First workflows in production
The highest-value use cases are built with review gates and audit logs, and measured against the result the manual process used to give.
↳ Workflows live, fully reviewed
5
Operate
Practice group and review ladder
The practice group approves new uses on a fixed cadence. Review shares step down as measured accuracy holds, per workflow.
↳ Governed, scaling programme

Human review

Review starts at 100 percent and steps down as measured accuracy holds

The steps and their pace are set per workflow by its complexity and the sensitivity of the data. The end state is a sampled check, not a person on every item, and the measurement never stops.

Share of outputs reviewed by a person, per workflow

1Launch: Every output checked against the manual result
2First iteration: Prompts and rules corrected from reviewer feedback
3Accuracy holds: Review focused on the categories that still fail
4Steady state: Sampled check. The measurement stays; the person leaves routine items

Training curriculum

What a staff AI training session covers

Module	Audience	Content
Five golden rules	All staff	The rules that cover most daily decisions about AI use
Green / Amber / Red	All staff	What data goes where, and the Amber checkpoint step by step
Prompt anatomy	All staff	Role, context, task, format, constraints; six techniques; a weak-to-strong rewrite
Hallucination cases	All staff	Real failures, how they were caught, and the verification habit
Scenario quiz	All staff	Six situations from the organisation's own work, decided as a group
Leadership edition	Senior management	Approval decisions, vendor questions, risk register, budget lines
Champions	One per team	Prompt library upkeep, first-line questions, feedback to the practice group

Format

Inside the team's standing meeting, so attendance is the whole team

After the session

Recording, slides, and policy text go to everyone, including absentees and new joiners

Follow-through

Champions keep the prompt library; the practice group reviews new uses

Systems

Where AI sits in the operation

AI workflows sit between the existing systems and the teams. They read, classify the data, draft a proposal, and route it by the policy: Green runs, Amber waits for a person, Red never enters.

Existing systems

ERP

CRM

Documents and SOPs

Portal and mail feeds

Field recordings

AI workflow layer

1
Reads the record, the document, or the feed item
2
Classifies the data by Green, Amber, or Red before anything leaves
3
Retrieves the relevant policy or reference passages
4
Drafts a proposal with its evidence
5
Routes by data class and the rule for that step

Routed by data class

Green

Public or internal data, low risk

Runs, result logged, sampled check

Amber

Sensitive data or a consequential action

Checkpoint, then a person decides

Red

Identifiable records, restricted sources

Does not enter the workflow

Every route writes an audit entry: input, data class, proposal, decision, who, when.

Retrieval and agents

Answers grounded in your own documents, with citations

Agent workflows use the same retrieval for grounded reasoning, while approval rules decide which actions run on their own and which wait for a person.

A grounded answer, as the reviewer sees it

What does our travel policy allow for field per diems when a trip crosses a month end?

Sources used

1 Travel and expenses policy, section 4.2, per diem basis
2 Finance manual 2025, monthly claims, page 12

Evaluation loop

1Reviewer reads

the answer next to the cited passages

2Feedback

what was missing, wrong, or over-stated

3Rule written

an explicit instruction, not a one-off fix

4Next version

measured on the same questions

Step 4 feeds step 1 until the measured quality holds; the loop runs with the people who know the subject.

Agent workflow: the model proposes, a person decides, the record shows why

One applicant screening, as it runs inside the ERP. The same pattern applies to opportunity screening, document extraction, and approvals.

The model proposes

Applicant 0418, Field Officer, Syria

Meets: four of five stated requirements
Gap: French, stated at basic level
Evidence: CV page 1, job description section 3

Draft only. No score shown as a verdict; the rule that ranks it is fixed and written down.

A person decides

HR officer, recruitment panel

ShortlistReturn

Reason

Shortlisted. French at basic level is acceptable for this post; the panel will test it at interview.

The record shows why

09:14 proposal drafted · model · inputs: CV, JD
09:14 data class: Amber · checkpoint passed
09:31 decision: shortlist · HR officer
09:31 reason recorded · visible to the panel

Every item can be traced back to who decided, on what evidence, and under which rule.

Related services

Connect AI to the systems that run the organisation.

AI workflows hold up when they are grounded in reliable records: the ERP, the integration layer, and the custom software that executes the work.

Related insights

Review and hardening of AI-generated code before production

The output is a system the organisation can operate and maintain, with the fallback paths and runbooks written down.

Security and dependency risk checks

Code quality and maintainability review

Performance and scalability assessment

Production deployment readiness validation

Prompt and model behavior traceability checks

Operational runbooks, fallback paths, and incident playbooks

Discuss an AI programme Explore all services

AI adoption for organisations

Four service lines, taken together or one at a time

AI strategy and adoption roadmap

AI use policy and governance

AI training for staff and leadership

AI workflows, agents, and ERP integration

How an adoption programme runs

Review starts at 100 percent and steps down as measured accuracy holds

What a staff AI training session covers

Where AI sits in the operation

Answers grounded in your own documents, with citations

Connect AI to the systems that run the organisation.

Further reading on practical AI adoption.

Review and hardening of AI-generated code before production

AI adoption for organisations

Four service lines, taken together or one at a time

AI strategy and adoption roadmap

AI use policy and governance

AI training for staff and leadership

AI workflows, agents, and ERP integration

How an adoption programme runs

Review starts at 100 percent and steps down as measured accuracy holds

What a staff AI training session covers

Where AI sits in the operation

Answers grounded in your own documents, with citations

Connect AI to the systems that run the organisation.

Further reading on practical AI adoption.

Review and hardening of AI-generated code before production