Many companies want to “do something with AI” before they know what AI should do, what data it should use, who should be allowed to use it, or how the output will enter a real workflow. That order creates weak projects. A better starting point is AI readiness: the practical questions that decide whether AI can become useful software capability rather than an isolated demo.
AI readiness is not only a strategy exercise. It is about systems, data, permissions, workflows, review points and operation.
Start with the work, not the model
The first question is not which model to use. It is what recurring task or workflow should improve.
NoA Ignite’s task-based approach to generative AI is useful here: start with people, roles and tasks rather than model features. Which tasks are repeated? Which tasks consume expert time? Which tasks require finding, summarising, classifying, drafting, routing or comparing information? Which tasks are high volume but low judgement? Which require human approval?
This makes AI easier to evaluate. A vague ambition such as “use AI in customer service” is too broad. A specific workflow such as “summarise incoming support emails, classify request type and prepare a draft response for review” can be analysed, prototyped and measured.
Review the data foundation
AI systems are only as useful as the information they can safely access. Before building, companies should ask:
- Where does the relevant data live?
- Is it structured, semi-structured or locked in documents and emails?
- Is it current enough to trust?
- Who owns it?
- Can it be retrieved through APIs, exports or databases?
- Are there duplicate or conflicting sources?
- Are access rights clear?
- Can sensitive information be separated or masked?
Twoday’s AI-ready data framing is relevant because it connects AI value to data quality, governance and business use. Poor data does not merely reduce model quality. It can create operational risk if an AI workflow acts on outdated, incomplete or unauthorised information.
Clarify permissions and access
AI readiness requires access control. A useful internal assistant may need to retrieve documents, tickets, CRM records, product data or system logs. But not every user should see the same information.
Access questions include:
- Which user roles exist?
- Which data can each role access today?
- Should the AI inherit existing permissions?
- Are there audit requirements?
- Should prompts and outputs be logged?
- Can the AI tool expose information across teams by accident?
- Are external providers allowed to process the data?
These questions are not blockers. They are design inputs.
Map the workflow around the AI
AI should sit inside a workflow, not next to it. A practical AI workflow usually includes five parts:
Data → Task/workflow → AI assistance → Human review → System actionThe human review step matters. Some outputs can be low-risk suggestions. Others require approval before they affect customers, financial records, legal documents, operational decisions or production systems.
Useful readiness questions include:
- What does the AI receive?
- What should it produce?
- Who reviews the output?
- What happens when the output is wrong?
- Which system receives the final action?
- How will failures be handled?
- How will quality be monitored over time?
Decide what “useful” means
A working AI demo is not the same as business value. Before implementation, define the success criteria.
Possible measures include:
- time saved in a recurring workflow;
- reduction in manual copying or classification;
- better information retrieval;
- fewer missed cases;
- lower review effort;
- faster reporting;
- improved consistency;
- safer handling of sensitive data.
If the workflow cannot be measured, it may still be worth exploring, but it should not be treated as production-ready.
Memory(One) perspective
For Memory(One), AI readiness belongs at the intersection of software, data, integration and operation. The work is not “add AI” in the abstract. The work is to understand where AI can support real workflows, what data it needs, how permissions should work, where humans stay in control and how the resulting system can be maintained.
Sources and inspiration
- NoA Ignite — How we plan for GenAI task by task: https://noaignite.com/insights/how-we-plan-for-genaitask-by-task/
- Twoday — AI-ready data becomes business critical: https://www.twoday.com/blog/ai-ready-data-becomes-business-critical
- NNIT — AI in the public sector: from pilots to responsible operations: https://www.nnit.com/insights/articles/ai-in-the-public-sector-from-pilots-to-responsible-operations
- NIST — AI Risk Management Framework: https://www.nist.gov/itl/ai-risk-management-framework