AI Readiness Assessment, a Scorecard and Checklist to Run Before You Build
Malay Parekh
CEO & Director, Unico Connect
In this article
- Quick Answer
- Key Takeaways
- Why Does AI Readiness Matter Before Development Starts?
- What Should an AI Readiness Assessment Evaluate?
- The AI Readiness Scorecard, Six Checks Scored 0 to 3
- Build Now, Delay or Build Differently?
- AI Readiness Assessment vs AI Maturity Assessment
- Who Should Be in the Room, and What to Bring
- What the Assessment Hands to the Project Brief and Proof of Concept
- Running This Assessment With Unico Connect
- Frequently Asked Questions
An AI readiness assessment is a decision process you run before the build to find out whether one AI use case has the conditions it needs to work in production. Unlike a general innovation exercise, it looks at one specific workflow and scores it on evidence, so the people paying for the build get a clear answer.
Quick Answer
An AI readiness assessment checks whether a specific workflow can support a production AI system, and you run it before you pick a model or a vendor. Score six pillars from 0 to 3 on evidence (business fit, data readiness, integration, governance and risk, people and adoption, and evaluation) and check each against its minimum. Build now when every minimum is met and the total reaches 65 out of 100 using weights we suggest. Otherwise delay until the blockers clear, or build differently with a narrower scope or a solution that does not use AI.
Key Takeaways
- Run the assessment on one workflow before any model or vendor is chosen, because the question is whether that workflow can support AI at all.
- Score only what someone can show, such as a data sample or a signed target.
- Minimums beat totals. Data readiness, governance and risk, or evaluation below 2 delays the build whatever the total says.
Why Does AI Readiness Matter Before Development Starts?
AI readiness matters before development starts because the checks a team skips come back mid build, as unclear ownership, poor workflow fit or output nobody trusts. CTOs, VPs of engineering, product leaders and operations leaders should give AI implementation readiness its own phase, because it tests the problem statement before anyone tests a model.
Over 80% of AI projects fail by some estimates that RAND cites, roughly twice the failure rate of IT projects that do not involve AI (RAND Corporation, 2024). The practitioners RAND interviewed most often blamed leaders who misunderstood the problem, then missing or poor data, and cited the limits of AI itself less often.
IDC documented the pattern in research with Lenovo, finding that about 88% of AI proofs of concept never reach widescale deployment, with only four of every 33 launched graduating to production. IDC reads that as a sign of "the low level of organizational readiness in terms of data, processes and IT infrastructure" (Lenovo CIO Playbook 2025).
What Should an AI Readiness Assessment Evaluate?
An AI readiness assessment should evaluate one use case on six pillars, and it leaves out the broad organizational audit. Each assessment question below needs evidence.
Enterprise AI readiness checklist
- Business fit. Does a named owner have a baseline and a target?
- Data readiness. Can the system reach the data it needs, with the right to use it?
- Integration. Do we know where input comes from, where output goes and what happens on errors?
- Governance and risk. Do we know the data and actions at risk, and who signs off?
- People and adoption. Have the people whose work changes seen the new flow?
- Evaluation. Can we tell a right answer from a wrong one before launch?
The AI Readiness Scorecard, Six Checks Scored 0 to 3
This working framework scores one use case on six pillars, using weights we suggest. Neither NIST nor ISO defines an AI readiness score. The voluntary NIST AI Risk Management Framework says that once the Map function is done, a team should know enough about the impacts of an AI system to inform an initial go or no go decision about whether to design, develop or deploy it (NIST AI 100-1, January 2023). NIST says AI RMF 1.0 is being revised as part of the White House AI Action Plan (NIST).
Business fit and data readiness weigh most because RAND interviewees most often blamed a misunderstood problem, then missing or poor data. Cisco uses the same six numbers for its company wide AI Readiness Index, but puts the largest one on infrastructure. For one use case, the problem and the data matter more. Score the highest level whose evidence someone can show. Multiply each score by its weight, add the results and divide by 3 for a total out of 100. Use the table as your spreadsheet template.
Table 1. AI readiness scorecard, weights, minimums and evidence for each score
| Pillar | Weight | Minimum | 0 | 1 | 2 | 3 |
|---|---|---|---|---|---|---|
| Business fit | 25 | 2 | A goal like "use AI in support" | A named workflow, but no owner or baseline | A named owner, a measured baseline and a target | A signed target and stop rule, with value after running costs |
| Data readiness | 20 | 2 | Nobody has looked at the data | Data exists, but access or gap owners missing | A real sample reviewed, access path known, gaps owned | A representative sample reachable today, permissions written down |
| Integration | 15 | 1 | No known trigger or output destination | Systems named, but APIs missing or owned elsewhere | Documented APIs, latency need and write actions listed | Inputs, outputs, checkpoint and fallback without AI on one page |
| Governance and risk | 15 | 2 | Regulated data in scope, no risk owner | A risk owner named, but nothing written down | Data classes known, vendor terms checked, approval actions listed | Error tolerance per action, audit logging required, risk owner signed off |
| People and adoption | 10 | 1 | Future users never asked | Users named, but not shown the idea | Future users walked through the flow, a reviewer named | The new workflow written down, review time agreed |
| Evaluation | 15 | 2 | Success means "it works" | Metrics named, but no baseline or test set | A test set of real cases with known answers | Pass thresholds, a named judge and written acceptance criteria |
1. Business fit, the problem and its owner
An AI readiness checklist starts with the workflow problem, and the underlying model comes later. Three questions settle this pillar. Which operational bottleneck improves, who owns that process, and what number changes if the build works? A focused workflow assistant built for one task is easier to score, fund and test than a broad AI transformation program.
On our Highlands Community Charter engagement, the first phase "framed what to unify into one platform, what to keep as a discrete integrated tool, and where AI would actually move the needle versus add noise." A 3 needs value that holds after running costs, so compute the return first.
2. Data readiness, access, quality and rights
Check the source data for quality, API access, freshness, structure, permissions and fragmentation in every source the use case needs, whether documents, live transactions or images. Gartner said in February 2025 that the data "must be representative of the use case." Score from a sample you pulled, never from the data dictionary.
3. Integration with your systems and workflow
To judge AI implementation readiness, map where the system sits in the live workflow. Name the upstream inputs, downstream actions, API access, where the model runs (your cloud or a vendor API), acceptable latency, human review checkpoints and a fallback without AI. A proof of concept can run on an export, so a 1 meets the minimum, and a 2 goes on the fix list for production.
4. Governance and risk, including human oversight
Governance has to be a design requirement from the first session, not a compliance layer added just before launch. Score data privacy, role based access, audit logging and the error tolerance for each action. Decide here which actions need human approval flows, and review the attack surface against enterprise AI security risks.
In Colorado, SB26-189 applies from January 1, 2027 to automated decision-making technology (ADMT) that materially influences a consequential decision about an individual in education, employment, housing, financial or lending services, insurance, health care services, or essential government services and public benefits. A federal court order of April 27, 2026 in xAI v. Weiser bars the Colorado Attorney General from enforcing it for conduct up to 14 days after the court rules on a preliminary injunction. If your use case feeds such a decision, have counsel check that status and help score governance and risk.
5. People and adoption
Score the people whose daily work changes, not the company culture. Name who uses the output and who reviews the cases the AI gets wrong. In a Gartner survey of 432 respondents published in June 2025, business units trusted and were ready to use new AI solutions in 57% of high maturity organizations, against 14% of low maturity ones. Two or three future users walking through a mockup, plus a named reviewer, earns a 2.
6. Evaluation, how you will know it works
Defining success before the build starts is not optional. Pick measures such as retrieval accuracy, target latency, completion rate, exception routing rate, time saved and adoption, with a baseline where one exists. After launch, the same measures feed how you define each AI KPI.
Build Now, Delay or Build Differently?
Build now when every pillar meets its minimum and the total reaches 65. Delay when a gap can close, and build differently when the problem is weak or a gap never closes. Read the minimums first, so a high total never hides a blocker.
Hard stops that override the total
- Business fit below 2. Build differently.
- Data readiness, governance and risk, or evaluation below 2. Delay, whatever the total.
- Integration or people and adoption at 0. Delay.
- A gap nobody can close. Build differently, for example when the data will never exist or a contract forbids the use.
Table 2. Score to decision rules
| Decision | Score rule | What happens next |
|---|---|---|
| Build now | Business fit 2 or 3, every minimum met, total 65 or more | Scope the proof of concept and start the brief |
| Delay | Business fit 2 or 3, but a pillar below its minimum or a total under 65 | Give each gap an owner and a date, then rescore |
| Build differently | Business fit 0 or 1, or a gap nobody can close | Narrow to one step, buy or embed, or use rules without AI |
We put the line at 65 because a use case with every pillar at 2 scores 67, while one that only meets the minimums totals 58 and waits. With several ideas and one budget, fund the highest total that clears every minimum. A delay with an owner and a date beats a build on a guess.
Build differently can mean a smaller dose of AI, so weigh buy, build or embed before you drop the idea. Gartner advised in June 2025 that organizations "can start by using AI agents when decisions are needed, automation for routine workflows and assistants for simple retrieval" (Gartner).
A worked example, scored
This is a made up illustration, not a client. A mid sized distributor wants AI to propose ledger coding for supplier invoices in its ERP, with a person approving every posting.
Table 3. Worked example, invoice coding (illustration)
| Pillar | Weight | Score | Points |
|---|---|---|---|
| Business fit | 25 | 3 | 75 |
| Data readiness | 20 | 2 | 40 |
| Integration | 15 | 1 | 15 |
| Governance and risk | 15 | 2 | 30 |
| People and adoption | 10 | 1 | 10 |
| Evaluation | 15 | 1 | 15 |
| Total | 100 | 185 ÷ 3 = 61.7, rounds to 62 |
The result is delay, because evaluation is below its minimum and the total is under 65. The team labels past invoices with the correct coding (evaluation 2), walks two clerks through the draft screen and names the approver (people and adoption 2). The rescore is 210 ÷ 3 = 70 with every minimum met, so the answer becomes build now, with integration still at 1 because the proof of concept can run on an ERP export.
AI Readiness Assessment vs AI Maturity Assessment
An AI maturity assessment rates the whole company, while an AI readiness assessment decides whether to fund one build.
Table 4. AI readiness assessment vs AI maturity assessment
| AI readiness assessment | AI maturity assessment | |
|---|---|---|
| Scope | One use case | The whole company |
| Question | Should we fund this build now | How far along is our AI capability |
| Output | Build now, delay or build differently | A maturity level and a roadmap |
| When | Before each build is funded | Yearly or when strategy changes |
The labels overlap. Gartner rates AI maturity from Level 1 to Level 5, while Cisco calls its company wide benchmark an AI Readiness Index. If the question is whether your company is ready for AI, score your best use case, and treat company wide AI adoption as a separate program.
Who Should Be in the Room, and What to Bring
Six roles belong in the room.
- The workflow owner, who signs the target
- The engineering lead or CTO
- The data owner, who can pull a sample
- The owners of the systems the AI reads and writes
- The risk owner from security, legal or compliance
- One or two people who do the work today
NIST asks for interdisciplinary participation, and RAND recommends including technical experts to assess feasibility. Send the checklist a week ahead, have each person score alone, then settle each pillar on the evidence. We suggest about 90 minutes per use case.
What to bring
- Volumes, handling time, error rates and cost for the workflow today
- A real data sample and the list of systems involved, with their owners
- Your data classification and any customer or regulator terms
- The actions the AI would take, and which ones need a person to approve
- A set of past cases with known right answers, if one exists
What the Assessment Hands to the Project Brief and Proof of Concept
The assessment is finished when its one page score sheet holds these five items.
- The scope boundary, what is in and what stays out
- The six scores, each linked to its evidence
- The fix list with owners and dates
- The signed baseline, target, test set and pass thresholds
- The open risks and actions that need human approval
These open the AI project brief, and the proof of concept should test the riskiest assumption first. Once it passes, the work turns to how you move a pilot to production.
Running This Assessment With Unico Connect
At Unico Connect, our AI development team uses this scorecard on projects where it applies, and can run a readiness assessment like this one before your build. We audit your data readiness, map use cases, and create an implementation roadmap with clear ROI targets. A proof of concept takes 2 to 4 weeks and a production deployment typically takes 2 to 4 months, depending on data readiness, model complexity, and integration surface.
Frequently Asked Questions
What is included in an AI readiness assessment?
An AI readiness assessment scores one use case from 0 to 3 on six pillars. They are business fit, data readiness, integration, governance and risk, people and adoption, and evaluation. The result is build now, delay or build differently.
How is an AI readiness assessment different from an AI maturity assessment?
An AI maturity assessment looks at the whole organization, while an AI readiness assessment looks at one project. Maturity covers culture, infrastructure across the enterprise and data literacy. Readiness asks whether one workflow is prepared for development to start now.
When should a team use an AI readiness checklist?
Use an AI readiness checklist during the initial budgeting phase, before you evaluate vendors, and in any case before anyone writes prototype code. Doing it that early gets the technical and business stakeholders aligned before the build starts.
What does enterprise AI readiness look like before a prototype?
An enterprise that is ready for AI has a named business owner, a real data sample it can reach, a mapped workflow, approved governance boundaries and success criteria set in advance, all before the prototype starts. On this scorecard, every pillar meets its minimum and the total is 65 or more.
When is a generative AI readiness assessment more useful than a broader review?
A generative AI readiness assessment is more useful when the use case writes text, answers from your documents through retrieval augmented generation (RAG), handles voice or images, or acts as an agent. These systems can state false output with confidence, which NIST AI 600-1, July 2024, calls confabulation. Add questions your data cannot answer to the test set, and plan a fallback without AI for each action.
How is an AI readiness score calculated?
Multiply each pillar score, 0 to 3, by its weight, add the six results and divide by 3 for a total out of 100. The weights we suggest are business fit 25, data readiness 20, integration 15, governance and risk 15, people and adoption 10 and evaluation 15. With every minimum met, a total of 65 or more means build now.
What should we fix first when data readiness scores low?
Fix access and rights before quality. In an AI data readiness assessment, get approval to pull a real sample, confirm you may use the data for this purpose, then give each gap an owner and a date. A score of 2, a reviewed sample with a known access path and owned gaps, clears the data minimum.
