AI Readiness Assessment: Scoring Whether You Can Actually Build

An AI readiness assessment measures whether your organisation can build, ship and operate an AI feature. Most assessments score strategy, culture and budget. Those matter, but they do not predict whether a model reaches production. Six engineering dimensions do: data pipelines, evaluation, observability, rollback, integration and ownership.

Key Takeaways

• Readiness has two halves. The organisational half decides whether you should start. The engineering half decides whether you can finish.
• Almost every published framework scores the first half. The six dimensions below score the second.
• Score each dimension 0 to 5 against evidence you can check, not against intent. A total out of 30 maps to one of four decisions.
• A single zero on rollback or evaluation outweighs a high average. Those two are stop conditions, not weighted inputs.
• Cisco’s 2025 index put 13% of organisations in its top readiness tier and 48% in the second-lowest, so most teams are assessing from behind.

Why most AI readiness assessments miss what stops the build

Strategy is rarely what kills an AI project. A team with a clear mandate, an approved budget and an engaged executive sponsor can still spend two quarters failing to ship, because nobody can say whether the model got better this week, and nobody can put it back if it gets worse.

That gap has a structural cause. Readiness frameworks are written by advisory firms, and they score what an advisor can see in a workshop. Pipelines, test harnesses and rollback paths are only visible to someone who has deployed a model, so they fall outside the instrument.

Both halves are real. The problem is that only one of them gets scored, and it is not the one that stalls delivery.

The six dimensions of engineering AI readiness

Engineering readiness is the ability to put a model into a production path and keep it there. Score each dimension from 0 to 5 using the observable evidence column, which is deliberately written so that a technical lead can verify every row in an afternoon without a workshop.

Used as an AI readiness checklist, the evidence column is the part that does the work. Every one of those artifacts either exists or does not, which is why this AI readiness assessment framework asks for scheduled jobs and test sets rather than opinions. Frameworks built on self-reported confidence produce scores that move when the person answering changes. Evidence does not.

“The evaluation harness and ground truth dimension scores lowest, more often than any other. Teams can demo a model convincingly, but nobody can say whether this week’s version beats last week’s, because no held-out test set existed before the model did.”- Phong Le, Tech Lead (AI, Python), Saigon Technology

How to score your organisation across the six dimensions

Scoring takes half a day with the right three people in the room: whoever owns the data platform, whoever owns deployment, and whoever will own the feature after launch. Work through the dimensions in order, because each one depends on the ones above it. Order matters here.

1. Pick one candidate use case first. Readiness is not absolute, and a team can be ready to build a document classifier while being nowhere near ready to run a pricing model.
2. For each dimension, ask for the artifact, not the answer. A yes with no scheduled job behind it scores 0.
3. Score 0 to 5 against the anchors in the table. Where the room disagrees, take the lower score and write down why.
4. Total the six. The maximum is 30.
5. Record every score below 3 as a named gap with an owner, before anyone starts discussing the total.

These bands are our own, and they are not an industry standard. An AI readiness assessment tool that returns a percentage and an email capture gives you a number. What changes a plan is the gap register underneath it, which is why the ordered list above puts gap capture ahead of the total. A single AI readiness score, on its own, tells you almost nothing about what to do on Monday.

What does “AI ready” actually mean?

Being AI ready means an organisation can move a model from idea to production and keep it running safely, with the data, tooling, accountability and governance that requires. It is a state of capability, not a level of enthusiasm.

Two distinctions are worth holding. Business AI readiness usually describes the organisational half: whether objectives, funding and skills line up behind a use case. AI maturity describes a different thing again, namely how far along a path an organisation already is, usually across several published stage models. Readiness asks whether you can start. Maturity asks how far you have travelled. They are not interchangeable. A company can score high on maturity in one division and be entirely unready in another.

Where organisational readiness frameworks still matter

The organisational half is not optional, and the published frameworks cover it well. Business AI readiness is what they measure, and they measure it competently. Use them for that purpose rather than as a predictor of delivery.

• Cisco’s AI Readiness Index scores six pillars, namely strategy, infrastructure, data, governance, talent and culture, and bands results out of 100 as Laggards at 0 to 30, Followers at 31 to 60, Chasers at 61 to 85 and Pacesetters at 86 and above. Its 2025 global distribution was 13% Pacesetters, 36% Chasers, 48% Followers and 3% Laggards (Cisco AI Readiness Index, vendor research, retrieved 2026-08-19).
• Microsoft’s AI Readiness Wizard runs ten questions across five drivers of AI value, scoring each on a five-level scale spanning business strategy, technology and data strategy, AI strategy and experience, organization and culture, and AI governance and security (Microsoft Adoption, retrieved 2026-08-19).
• RSM’s four-week engagement assesses six core pillars: strategy and vision, data foundation, organization and culture, technical infrastructure, governance and compliance, and model management (RSM US, retrieved 2026-08-19).

Read across those three and a pattern shows up. Governance, culture, talent and strategy appear in all of them. Evaluation, rollback and drift appear in none. That absence is the gap. For enterprise AI readiness the governance pillar carries real weight, particularly where AI usage policies, GDPR obligations and regulatory controls apply to the data a model touches, and the engineering dimensions are what turn a written policy into something a system actually enforces.

What does an AI readiness assessment deliver?

A useful assessment produces four artifacts. If an engagement returns a score and a slide deck, it has produced one of them. Ask for all four.

• A gap register. Every dimension scoring below 3, with an owner, the evidence that is missing, and what it blocks. This is the output that survives the meeting.
• Sequenced remediation. Gaps ordered by dependency rather than severity. There is no point tuning drift alerts before inputs are logged.
• An effort and cost estimate per gap. Enough to decide what gets funded this quarter and what waits.
• A go or no-go per candidate use case. Not a single verdict for the organisation, because readiness is relative to the workload.

An AI readiness audit and an AI gap analysis describe the same exercise seen from two directions. The audit establishes the current state against evidence. The gap analysis compares that state to what a specific use case needs. Both are only as good as the artifacts they demand.

Common mistakes that make a readiness score misleading

1. Scoring intent instead of capability. “We are planning to add monitoring” scores 0. Roadmaps are not evidence, and treating them as evidence is the most common way a readiness score comes out flattering and wrong.
2. Scoring the pilot environment rather than the production path. A notebook that reads a CSV proves the model works. It proves nothing about the pipeline that has to feed it every morning.
3. Averaging away a zero. A 0 on rollback next to five 4s gives a respectable 20 out of 30 and hides a stop condition. Read the low scores before the total, every time.
4. Assessing data quality with no use case attached. Quality is relative to a task. The same dataset can be clean enough for demand forecasting and unusable for anything customer-facing.
5. Letting whoever scores you own the remediation without a written gap register. The register is what makes the work checkable by someone else later.

Turning the score into a sequenced plan

The score is an input to one decision: what to do next, and at what size. Size follows evidence. A total in the 11 to 18 band points at a single timeboxed proof of concept with the gaps stated in the scope, rather than a programme. Anything below 11 means the first piece of work is remediation, not a model.

Published delivery timeboxes give a sense of the shape. Our own sequence runs discovery in one to two weeks, prototype in two to six, an MVP build in four to ten, deployment in one to four, then ongoing optimisation (Saigon Technology). Behind that sit 100+ AI projects, 30+ AI engineers and two dedicated AI teams in Ho Chi Minh City and Da Nang, plus ISO 9001 and ISO 27001 certification, which is the part that matters when the gaps are in governance rather than tooling.

Two adjacent decisions usually surface during an AI readiness assessment. The first is the build-versus-buy question, because a low score on integration often argues for buying. The second is how the engineering workflow itself changes, which we cover in AI-augmented software development. For a worked example of the six dimensions in a production system, see the AI platform we built for multi-asset capital markets, or talk to our AI development team about scoring your own.

Frequently asked questions

1. How long does an AI readiness assessment take?

Scoring the six dimensions takes about half a day with the data, deployment and product owners present. A full engagement including evidence gathering and a remediation plan typically runs one to two weeks, matching a normal discovery phase. Anything quoted at four weeks or more is usually assessing the organisational half as well.

2. Who should be involved in the assessment?

Three roles, at minimum: whoever owns the data platform, whoever owns deployment, and whoever will own the feature after launch. That third person is the one most often left out, and their absence is why ownership scores tend to be optimistic. Executive sponsors are useful for the organisational half and not needed for this one.

3. What are the three pillars of AI readiness?

Most published frameworks group readiness into data, technology and people, then add governance as a fourth. Cisco uses six pillars, Microsoft uses five drivers, and RSM uses six. The number varies by vendor, and the common ground across all of them is data, infrastructure, skills and governance.

4. Should you assess your data before anything else?

Yes, in the sense that dimension 1 gates the other five. A model cannot be evaluated or monitored if its inputs are not reliably available. Data readiness for AI is a substantial topic of its own covering quality, governance and access, and it deserves separate treatment beyond the pipeline and lineage view scored here.

5. Can you assess AI readiness without a specific use case in mind?

Partially. The organisational half can be scored in the abstract. The engineering half cannot, because latency budgets, ground truth and integration surfaces are all defined relative to a workload. If no use case exists yet, score dimensions 1, 4 and 6, which are largely use-case independent, and defer 2, 3 and 5.