Diagnostic
The cheapest way to find out whether there is anything here worth doing.
- Interviews
- A map of where the work and the knowledge sit
- A ranked list of what is worth building
- Cost and risk
- A written recommendation
AI is advancing faster than organizations can absorb it.
IDG is the AI department you do not have to build. You operate your organization. We operate its AI capability.
We plan and build your AI systems, run them where you choose, under rules your leadership sets, and teach your people to use them.
- Blue: the AI capability
- Ink line: a rule leadership set
- Hairline: a measurement
- Boxed label: a person decided
Fig. 1Blue dropped on unruled paper. It beads up and runs off without soaking in.
LeadershipFinanceOperationsResearchFront deskRecords
Fig. 2An unprepared organization. Use placed at random, no rules drawn.
It's already in the building, just not on your terms.
Staff are already pasting work into public tools leadership can't see.
Most organizations should not become AI companies. They should have an AI function. It does not have to be one they build.
Leadership should set the rules, not staff using tools nobody approved.
The blooms start in the rooms furthest from leadership.
OriginFrontDiagnosticDesign & buildInfrastructureGovernanceTrainingManaged operations
Fig. 3One spot separates into six bands. Each reads alone or together.
Six services, one relationship.
Each one stands on its own. Bought singly or together.
Design & build
Delivered into production rather than into a demo. Nothing gets ripped out.
- Architecture
- Integrations
- Search and permission layers
- Interfaces
- Testing
- Rollout
Infrastructure
We create the secure environment AI needs.
- The environment
- Model hosting
- Identity and network
- Keys
- Monitoring
- Backup
Governance
Who may see what, what needs a person, and paperwork a regulator can read.
- Access model
- Approvals
- Audit logs
- Retention
- Documentation a regulator can read
Training
Adoption is most of the work.
- Leadership briefings
- Role-specific sessions
- Written guidance
- Refreshers
Managed operations
A named contact rather than a ticket queue.
- Monitoring and incident response
- Security upkeep
- Cost control
- Model checks against your own work
- Regulatory watch
- A quarterly written review
- A named contact
Leadership's wellFoundationKnowledgeResearchDocumentsProcessesFinanceIn your tools
Fig. 4One well, one trunk, six zones. Each new branch is shorter.
One foundation underneath. Each new system builds on it.
Leadership sets the rules once, at the well. Every system drawn from it inherits them.
A private AI workspace for your people. Your people use it three ways: ask it. assign it. or let it run.
Six kinds of system can draw from the same foundation, each one building on what is already there.
Ask it. Assign it. Or let it run.
Ask it
A question, answered from what your organization already holds, with its sources.
How do we send files to someone outside the organization?
Through the secure portal, once a manager has signed off.
Source: Information handling policy, section 3
Assign it
A task handed off, with a point where a person approves it.
Draft the renewal letters for this month's contracts.
Drafts ready. Waiting at the approval point.
Let it run
A recurring job, such as the Monday report, that runs on schedule and reports back.
The weekly operations report, prepared every Monday at 06:00 for a person to approve.
- Mon 06:00Prepared, approved
- Mon 06:00Prepared, approved
- Mon 06:00Prepared, approved
Six kinds of system
- 1Knowledge with sourcesAnswers that show where they came from, or say they can't answer.
- 2Research gathered and cross-checkedInternal and approved outside material, checked before it is presented.
- 3Document workReading, extracting, drafting and checking.
- 4Processes that runThe Monday report, prepared every week for a person to approve.
- 5Finance and reconciliationMatching records, invoicing and reporting.
- 6Help inside the tools people already useAssistance where the work already happens, not another app to open.
Where a system runs is a decision, not a default.
Your choice, decided workload by workload. Three perimeters, each one yours to choose.
Fig. 5An office plan. The work spreads from one machine under a desk and stops at the outer wall.
A machine in your office
Your information does not have to leave the building. The office walls are the boundary.
Perimeter: yours
Your own cloud account
Fig. 6A perimeter in your own account.
Perimeter: run by us, for you
A private setup we run
Never a public tool.
Fig. 7A perimeter inside a perimeter.
Ink rule: what leadership set
Fig. 8The boundary, close up. Each line left behind is one entry in the record.
Three rules, and a record of every step.
- Nobody reaches anything through AI they couldn't already open.
- A person approves anything consequential.
- Every step is written down: who asked, what was used, what came back.
| Who asked | What was used | What came back |
|---|---|---|
| Finance lead | Supplier contracts folder | Summary of renewal terms |
| Front desk | Visitor policy | An answer, with its source |
| Operations | This week's figures | Draft report, approved by a person |
| Research | Approved outside sources | A cross-checked brief |
Leadership's wellFoundationKnowledgeResearchDocumentsProcessesFinanceIn your tools
- March
- April
- May
- June
- July
- August
- September
- October
- November
Fig. 9The same device. Zones take colour once people are taught. Later the model changes; the channels do not move.
Adoption is most of the work.
We would rather deploy less and have it used.
- Leadership briefings
- Sessions on the system you were actually given
- Refreshers when it changes
The thing you build in March needs someone answering for it in November.
A named contact rather than a ticket queue.
Models change. Your setup stays.
The model underneath can be swapped as better ones arrive, while your setup stays put.
It's yours. No lock-in.
Documented well enough that another engineer could take it over.
Organizations that can't afford to leak what they know, and have no AI department of their own.
- Example situation: a law firm
The ABA's ethics guidance (Formal Opinion 512) says a lawyer needs the client's informed consent before client information goes into a self-learning AI tool.
- Example situation: a tax practice
Preparers generally need the taxpayer's consent before using or disclosing return information (IRC §7216).
- Example situation: a company
Staff are already pasting work into public tools leadership can't see.
There is no industry edition.
Most begin with a diagnostic and one system, then keep us on to run it.
Retained the way you retain counsel, auditors or a security firm: named people, defined scope, a fee.
- Understand
- Design
- Build
- Adopt
- Operate
Fig. 10 · Every part of the organization reached
Show us where your organization gets stuck.
If AI isn't the answer for it yet, we'll say so.Or email hello@idg.consulting