CONSITE · AI adoption consulting
Effective AI Adoption Pathway
Five steps from scattered AI experiments to measurable business value.
AI in companies today
5 in 100
company AI pilots ever show up in the P&L.
The other 95 have no measurable impact on profit. Not low. None.
Source: MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (300+ public deployments, 150 leader interviews).
Shadow AI
One in two
employees already use AI their company never approved.
With company data, often on free accounts, outside any control. Banning it doesn’t help; it only pushes it further into the shadows.
58%
of those using unapproved AI rely on free versions.
51%
of employees have connected AI to work systems without IT approval.
Source: BlackFog / Sapio Research, January 2026 (2,000 employees at companies with 500+ staff, UK and US).
Why pilots stall
Three gaps. None of them technical.
Workflow gap
95%
of pilots never reach the P&L. AI sits beside the work, not inside it.
MIT Project NANDA, 2025
Control gap
51%
of staff have connected AI to company systems without IT knowing. Leadership can’t see what’s happening.
BlackFog, 2026
Skills gap
7.5%
of employees have had thorough AI training. Licences without training go unused.
Licences close none of them. A well-planned pathway closes them in the right order.
What sets the 5% apart?
They don’t have better tools. They have a better path.
Effective AI Adoption Pathway: five steps in the right order, each ending with a measurable result.
Effective AI Adoption Pathway
Five steps from experiments to measurable value
T1 · Captain
Appoint an AI captain
One owner, a clear direction and three jobs worth fixing.
T2 · Crew
Train the crew
Approved tools, connected data and training on real work.
T3 · Map
Share the map
Company memory and shared methods instead of personal tricks.
T4 · Oversight
Switch on oversight
Value, cost and risk on one page. Every month.
T5 · Autopilot
Turn on the autopilot
Agents handle the routine; people approve what matters.
Phase 1 · first results in about 6 months
Phase 2 · prove, protect and multiply the results
Where are you today?
Five statements, one honest answer
Tick what is really true for you. Count in order, for as long as each statement holds.
Your level
00
Random experiments
People try AI on their own, with no owner and no rules. That’s not a failure; it’s the normal starting point.
Next step: T1 · Appoint an AI captain
Discuss it in a consultationPhase 1 · T1 → T3
From experiments to first results
T1 · about 6 weeks
Appoint an AI captain
Survey, 10–15 interviews and a workshop: value × effort × risk.
You take home
- An AI opportunity map
- One-page rules
- A starting score
- A plan and a go / no-go decision
T2 · about 8 weeks
Train the crew
Rollout in waves, role-based training, a champion in every team.
You take home
- Approved tools for everyone
- Connections to CRM, tickets and wiki
- Company memory v1
- A library of proven methods
T3 · about 10 weeks
Share the map
Personal tricks become one shared system. This is where results start to compound.
You take home
- Company memory with sources
- A catalogue of company methods
- Live data connections
- A central registry: what we have and who owns it
Before AI touches your data
AI is a magnifying glass. It enlarges the mess you already have.
Without preparation
“How much did we earn last month?”
€412,000
Fast. Confident. But whose number is it?
- The project manager means hours delivered.
- The accountant means booked profit.
- The CEO means cash and margin.
With company memory
“How much did we earn last month?”
“Do you mean hours, invoices or booked profit? September closes on the 10th and three invoices haven’t been sent yet.”
Illustrative example · sources: accounting · CRM · timesheets
- One truth per question: client = CRM, profit = accounting
- The business defines what data means, not the column names
- Every fact cites its source
We know how to do it.
Our LTM (Long-Term Memory) solution shows it in practice: every night it reads what changed in the company, every morning it finds duplicates and contradictions, and every answer comes with a verifiable source.
Phase 2 · T4 → T5
Prove, protect and multiply the results
T4 · about 10 weeks
Switch on oversight
You can’t steer what you can’t see. Value, cost and risk on one page.
- AI board and rulebook v2
- Risk traffic lights based on the EU AI Act
- Monthly value report in money and hours
- Logs and budgets ready for audit
T5 · about 8 weeks per process
Turn on the autopilot
Agents take over the routine. Each next step only when the numbers allow it.
- Shadow: the agent suggests, a person does the work.
- Supervised: the agent works, a person approves every case.
- Autopilot within limits: a person approves only the exceptions.
Without Phase 2, nobody can prove the value and the AI budget is the first to be cut. Then the first audit or incident stops everything.
The question every CEO asks
When will you see results?
~10 weeks
first hours saved
~6 months
results across the company
Indicative targets. The exact plan is set in T1, based on where you start.
How we work
Start where you are. Buy only the steps you need.
Step 1 · start
Intro consultation
We look at where you are today, what already works and where AI will save you the most time.
Step 2 · about 6 weeks
T1 diagnostic
Opportunity map, rules, starting score and plan. Useful on its own.
Step 3 · at your pace
Guidance all the way
Phases 1 and 2 step by step, with a clear decision after each one on whether to continue.
- Independent of technology vendors
- Experience in regulated industries: ISO, GxP, GAMP5
- From strategy to implementation
References
We have worked for
- Microsoft
- MSD
- Novartis
- ING
- GE
- Home Credit
- PPF
- Taiwan Semiconductor
- Sotio
- Czech Television
- TV Nova
- Datart
A selection of references across banking, pharma, industry, technology, media and retail. Details available on request.
Next step
Let’s find out where you are today.
An intro consultation shows where to start. First hours saved in about ten weeks.