On Monday, a founder asks the team to “add AI” to customer support.
By Wednesday, three tools have been tested. None can access the right customer records. The operations team worries about wrong answers. Finance cannot explain what success would be worth.
The problem is not the model. The business has skipped the readiness work.
The problem is not the model. The business has skipped the readiness work.
Your business is ready for AI when it has a valuable problem, usable data, a clear owner, safe operating rules, and a way to measure results. Use this AI readiness checklist before buying a tool, hiring an AI readiness consultant, or approving a custom build.
The exact business problem
Most companies ask, “Which AI tool should we buy?”
That is the wrong first question.
Start with the work. Where does your team repeatedly lose time, money, accuracy, speed, or customer trust? Then compare four options:
Remove the unnecessary step.
Simplify and standardise the process.
Use normal rule-based automation.
Use AI when the work involves language, images, prediction, pattern recognition, or changing context.
RAND research found five recurring causes of AI project failure: teams solve the wrong problem, lack suitable data, chase new technology, lack deployment infrastructure, or choose a task that AI cannot reliably handle. The study was based on interviews with 65 experienced data scientists and engineers, so it is strong diagnostic evidence, not a universal rule for every company.
The cost of making the wrong decision
AI adoption is moving quickly. Across OECD countries with available data, 20.2 percent of firms reported using AI in 2025, up from 8.7 percent in 2023. Faster adoption creates pressure, but pressure is not a business case.
A weak implementation creates software, integration, data cleanup, disruption, and risk costs.
Before approving a project, calculate:
Labour saved + added gross profit + expected losses avoided
(Annual value − year one cost) ÷ year one cost × 100
Initial implementation cost ÷ monthly net value
Include loaded employee cost, realistic adoption, errors, software, monitoring, support, and retraining. Do not assume every saved minute becomes productive work.
The 5D AI Readiness Framework
Score every point from 0 to 2.
This is a decision tool, not a certification.
D1: Demand and business value
A specific problem is defined
You can name the workflow, user, pain, frequency, and current consequence. “Improve productivity” is not specific. “Reduce the time spent sorting 1,200 customer emails each month” is specific.
The current baseline is measured
You know today's volume, time, cost, error rate, response time, conversion rate, or revenue leakage. Without a baseline, you cannot prove that the AI implementation worked.
Success has a financial threshold
The project has one primary outcome, a minimum acceptable improvement, a budget ceiling, and a target payback period. A technically impressive system can still be a poor investment.
D2: Data readiness for AI
The required data exists
The system can access the documents, records, messages, images, transactions, or knowledge needed for the task. Having a large amount of data does not automatically mean you have the right data.
The data is fit for purpose
It is accurate enough, current, complete, consistently labelled, and representative of real cases. Check unusual cases too. A system trained only on simple examples may fail when real work becomes messy.
Data rights are clear
You know who owns the data, who may access it, where it may be stored, and whether customer or employee consent is required.
ISO/IEC 5259 treats data quality as a governance responsibility. GAO also recommends documenting sources and assessing reliability, bias, security, privacy, and dependencies.
D3: Delivery and technology
The workflow is documented
Inputs, decisions, exceptions, handoffs, and desired outputs are visible before automation begins. AI does not repair a confused process. It usually makes the confusion move faster.
The system can integrate with real work
It can securely connect with the tools your team already uses rather than creating another isolated dashboard. The system should fit the workflow. Your team should not have to rebuild its entire day around the system.
Reliability can be tested
You have representative test cases, acceptance criteria, failure handling, logging, and a rollback plan.
UK National Cyber Security Centre guidance treats security as a life cycle requirement, covering threat modelling, supply chains, incidents, logging, monitoring, and updates.
D4: Decision rights and governance
One business owner is accountable
A named leader owns the result, budget, adoption, and decision to stop or scale. The technology team may build the system. The business owner must still own the outcome.
Human oversight is defined
People know when they must review, approve, correct, or override the system. The higher the cost of an error, the stronger the human control should be.
Risk and compliance are assessed
The company has reviewed privacy, security, bias, intellectual property, vendor terms, record keeping, and relevant laws.
NIST covers risk across design, development, use, and evaluation. ISO/IEC 42001 adds policies, roles, risk treatment, monitoring, and improvement. These standards guide control and accountability. They do not guarantee commercial return.
Companies serving EU markets should classify use cases under the AI Act. Most provisions applied from 2 August 2026, with later dates for some high risk obligations.
D5: Deployment, adoption, and scale
The build versus buy decision is rational
Buy when the problem is common and a trusted product meets the need. Build when your workflow, integration, data, control, or customer experience creates strategic value. Do not build custom software simply because it feels more advanced.
A controlled pilot is possible
You can test one workflow, user group, or customer segment without disrupting the whole company.
Does the system work?
Will people use it?
Does it create enough value?
The operating model exists
Budget, training, support, monitoring, vendor management, review dates, and scaling rules are assigned. AI implementation readiness does not end when the system launches. Models, data, costs, risks, and business conditions change.
Your AI readiness assessment score
A high score means you can test without relying on hope. It does not guarantee success.
Step-by-step AI implementation plan
First, choose one recurring workflow with visible cost and measurable volume.
Second, map it with the people doing the work.
Third, collect real cases, including exceptions.
Fourth, compare removal, normal automation, an existing AI product, and a custom system.
Fifth, calculate conservative value and full year one cost.
Sixth, run a human reviewed pilot. Measure outcomes, quality, adoption, failures, and risk.
Seventh, scale only after crossing the agreed threshold.
GAO recommends defining performance metrics, testing against them, monitoring changes in performance, and maintaining human supervision throughout the system life cycle.
A realistic AI readiness example
A service team receives 1,200 customer emails each month.
Triage takes eight minutes per email, or 160 hours. At a loaded cost of $25 per hour, the current task costs about $4,000 each month.
A pilot shows that AI can correctly route 45 percent of messages, with people reviewing uncertain cases. That saves 72 hours, worth $1,800 monthly.
The setup costs $6,000. Software and monitoring cost $500 monthly.
Net monthly value after software and monitoring cost
Estimated payback period
Estimated year one ROI
These remain assumptions until the pilot proves accuracy, adoption, and time saved. A good demo is not yet a profitable system.
Mistakes and warning signs
Stop if nobody can name the metric, data cannot be legally accessed, the workflow constantly changes, users reject the system, or errors could cause serious harm.
Be cautious when a vendor leads with features, hides ongoing costs, cannot explain data handling, or lacks testing, monitoring, documentation, and an exit path.
Another warning sign is a company wide launch before one workflow has produced measurable results.
Scale evidence, not excitement.
Frequently asked questions
What is business AI readiness?
Business AI readiness is the ability to implement AI in a way that creates measurable value while controlling data, operational, financial, security, and compliance risk.
Does every company need an enterprise AI readiness programme?
No. Small firms can use a lighter process. The core requirements remain the same: a valuable problem, usable data, an owner, safe controls, and measurable results.
How long does an AI readiness assessment take?
One workflow may take days. A company wide AI readiness audit can take weeks because it covers several systems, teams, data sources, and risks.
Should we buy an AI tool or build a custom system?
Buy for a standard need that a trusted product fits. Build when integration, control, proprietary data, or strategic advantage matters.
What should an AI readiness consultant deliver?
Expect prioritised use cases, baseline costs, data and risk gaps, build versus buy analysis, ROI estimates, pilot scope, metrics, and an AI implementation roadmap.
Avoid AI strategy consulting that ends with a generic slide deck and no implementation decision.
Summary and next actions
Do not ask whether your company is “using AI.”
Ask whether one AI use case can solve a measured problem better than the available alternatives.
Score the 15 points. Choose one workflow. Build the business case. Test it with real data and human oversight. Then scale only what produces evidence.
Eveningside Labs starts its AI readiness audit with the business problem and states that it will say when AI is not the right answer rather than beginning with a software pitch.
That is the standard your AI implementation assessment should meet:
Less excitement. More proof.