Your CEO asks a simple question: “Are we ready for Artificial Intelligence (AI)?” You know the plants are busy. You know the ERP has run for years. You also know that three teams keep three versions of the same numbers. So you say, “Mostly.”
“Mostly” does not survive a budget meeting. It does not survive a pilot, either. AI systems rely on the data they are given, including its gaps and workarounds.
This AI readiness checklist will help you prepare data for AI and turn “mostly” into a clear answer. When it’s complete, every step should have a clear owner and a clear finish line.
At HQSoftware, we deliver a data and AI readiness audit for manufacturing and logistics teams. The gaps this checklist covers are the ones worth checking before that audit starts, rather than being discovered afterward.
Why Is This an AI Readiness Checklist and Not a Self-Assessment?
What Should You Do Before a Data and AI Readiness Audit?
What Does Organizational AI Readiness Actually Require?
What Technical Foundations Does AI Need?
What Happens When You Prepare Your Data Before Building an AI Model?
We already published a practical self-assessment for operations leaders. At first glance, this article may look like the same thing. It is not. The self-assessment is a diagnosis. You answer some questions, see your score zone, and learn where you stand. Think of it as a medical checkup.
This data readiness checklist, on the other hand, is the treatment plan. It has no score. It lists tasks in order, with an owner for each one. You need this checklist in three situations:
Not sure where problems are hiding? Take the quiz first, then come back here.
McKinsey found that 88% of organizations now use AI in at least one business function, yet only 7% say AI is fully scaled across the organization.
Using AI somewhere is not the same as scaling it. Data preparation is one of the things worth checking before you try to close that gap.
![Dima T round - The AI & Data Readiness Checklist for Operations Leaders]( "The AI & Data Readiness Checklist for Operations Leaders")
Ready to skip straight to an expert assessment?
If you’d rather have a team run the diagnosis for you instead of doing it yourself, that’s exactly what a readiness audit is for.
Dmitry Tihonovich
Business Development Manager
Before the audit, list your data sources and fix the obvious quality problems. This matters more than it sounds. An audit team spends its time either finding real gaps or walking through your basic inventory with you. You want them doing the first one, since that is what you are paying for.
Skipping this step costs real money, and not just in audit hours. Gartner puts the average cost of poor data quality at at least $12.9 million a year, based on research into large organizations conducted in 2020. That figure represents the cost of rework, bad decisions based on wrong numbers, and hours spent reconciling reports that should have been agreed in the first place.
Your own number could be smaller or larger, but it has the same source every time: errors nobody caught early because nobody was looking for them in the first place.
Write down every system that holds data for your first AI use case. Think ERP software, warehouse management, maintenance logs, and yes, the shared spreadsheets everyone pretends do not exist. AI models do not read your org chart; they read whatever data actually feeds a process. A maintenance team might log real issues in a notebook app nobody in IT knows about. If that source is not on your map, the AI will miss exactly the signal it needed most.
For each source, note who owns it, how often it updates, where it feeds, and where it lives: a data lakehouse, a legacy database, or a shared drive. Then add two fields that matter specifically for AI readiness: how critical that data is to your use case, and which system is the source of truth when numbers disagree.
That second question matters more than it sounds. Picture an ERP showing one inventory count, a warehouse system showing another, and a manager insisting the real number lives in their spreadsheet. An AI model needs one answer, not three. Someone has to decide which system is authoritative before the model ever sees the data.
Do not clean everything. Clean what your first use case touches. Trying to fix every dataset in the company before starting is one reason AI projects die quietly, a year into “data cleanup,” with nothing shipped.
The reason to fix even the obvious issues first is simple: AI models do not know when to doubt their input. A human analyst sees a date stored two different ways and recognizes it as the same day. A model treats it as two different facts, and both get used in a forecast or a decision. Small inconsistencies that a person would shrug off become invisible errors in the final output.
Here is a short data quality checklist that any AI initiative can start with:
Old workarounds pile up as technical debt, and an AI model inherits every single one of them, without ever telling you which one caused the problem.
For organizational readiness, an operations leaders’ AI checklist comes down to three things: a named owner for each key dataset, agreement between IT, operations, and finance, and a plan for the people who will use the results.
The last point is one that’s often skipped, and it is usually the one that sinks a project. In a survey of data, analytics, and AI executives, 92% said cultural and change management challenges are the primary barrier to becoming data- and AI-driven. MIT Sloan Management Review reported the finding.
People and processes deserve as much planning as pipelines, because a perfectly clean dataset still fails if nobody on the floor trusts what it produces.
Give every critical dataset one named owner. Not a team. Not “IT.” One person. This matters, because shared ownership quietly becomes no ownership. When three people could theoretically fix a data issue, each one assumes someone else already noticed it, and the error sits there for months. A single named owner closes that gap immediately: there is exactly one person to ask, and exactly one person accountable for the answer.
That person should be able to answer three questions: Who may change this data? Who approves what each field means? Who fixes errors when they appear? Write the answers in a shared document. This is the core of a lean data governance checklist, and it fits on one page.
Each group worries about something different, which is why alignment matters. IT thinks about security and integration. Operations thinks about downtime and workload. Finance thinks about cost and payback. Left alone, each group optimizes for its own concern and assumes the others are covered. Usually, none of them are.
Hold one meeting. Agree on one use case, one success metric, and the time-to-value (TTV) you expect. Name a person who will talk to the floor team early, not after the model is already built. That early conversation is basic change management, and if you skip it, an otherwise technically sound project may be quietly ignored once it launches.
Deloitte’s 2026 survey of 3,235 leaders shows why this step matters so much. Companies feel more prepared on strategy than they did a year ago but less prepared when it comes to data, infrastructure, risk, and talent. A confident strategy slide does nothing to fix a data pipeline that nobody owns. Every digital transformation plan needs the unglamorous data work to keep pace with the ambition behind it.
Artificial Intelligence needs two things from your technical side: working pipelines between core systems and your analytics layer, and three documents: an entity relationship diagram (ERD), a data dictionary, and access logs. This is the technical half of any real AI readiness checklist. Skip either part, and the audit team spends its time reconstructing what you already know instead of finding what you do not.
Your ERP holds most of your operational data, so how that data travels matters as much as the data itself. A clean ERP integration means data moves automatically, without someone exporting a spreadsheet every Monday morning to patch a gap no one has fixed properly.
Check three things.
A pipeline that fails silently is worse than one that fails loudly, because a loud failure gets fixed. A silent one just quietly feeds bad numbers downstream.
Pick one report your team trusts and trace it back to its source system, step by step. The trace usually reveals at least one manual patch nobody mentioned in the original system diagram. That patch is exactly the kind of risk an AI model will inherit without warning, and exactly the kind of gap a real AI implementation checklist is supposed to catch before it causes damage.
If predictive analytics is part of your roadmap, plan for MLOps now, not later. MLOps covers how models get deployed, monitored, and retrained once they are live. Bolting it on after a model is already running costs more than building it in from the start.
An ERD shows how your data tables connect to each other. A data dictionary explains what each field actually means —in plain language, not database jargon. Access logs show who touched which data, and when.
These three documents are what separate a quick audit from a slow one. Without them, an audit team will need to spend its first week just mapping your systems before it can even start looking for real problems. With them, that week gets spent on findings instead. This is the kind of groundwork a data and AI readiness audit should walk into, not discover on day one.
Missing documents are normal, not a red flag. Draft a first version covering only the datasets your current use case touches, and expand it later as new use cases come up.
One HQSoftware warehouse project shows what a clear data inventory looks like in practice.
Take a U.S. distribution center that managed about 11,000 active products and 4.2 million transactions over 18 months. Its warehouse layout was planned by hand, and it had drifted further from real buying patterns over time. Pickers walked farther than they needed to.
The project started from an inventory of six data sets: transaction history, the product catalog, order composition, warehouse layout metadata with zones and coordinates, seasonal effects, and historical picker route data. That is the same inventory step described earlier in this checklist, applied to a real production system.
The approach was validated with a six-week live A/B test, comparing a control warehouse against a test warehouse. Average order-picking time fell by 36% compared to the manual layout, orders picked per hour rose by 29%, and 84% of demand forecasts per SKU stayed within a 15% error margin.
These results measured the full system against the warehouse’s previous manual layout, not the effect of data preparation in isolation. What the inventory step did provide was a complete, documented starting point: a model with six clearly identified data sets to draw on, instead of guessing at which inputs matter. That is the groundwork a data and AI readiness audit is built to establish before a larger AI project begins.
![Untitled design 5 round - The AI & Data Readiness Checklist for Operations Leaders]( "The AI & Data Readiness Checklist for Operations Leaders")
Want to see more projects like this one?
The warehouse case above is one of several. Our case studies cover AI and data projects across manufacturing, logistics, and other industries.
Everything in this article comes down to ten concrete items. This is the operations leaders’ AI checklist we would want to see on a first call with any client, and it doubles as the opening page of a broader AI implementation checklist.
The list below covers three areas in order: defining the use case, cleaning the data behind it, and getting the right people and documents in place. Work through it with your team, and keep track of which items are still open.
Every item checked means your data and AI readiness audit starts from a map instead of a blank page. The audit team spends its time on findings, not on hunting for basics. You get a remediation roadmap with specific steps to fix your data foundation, as described in what to expect from an audit.
If several items are still open, that is normal at this stage. Talk to our team, tell us which ones, and we will start there.
Lead Software Engineer
An experienced software developer specializing in IoT and AI-driven technologies. Having contributed to more than 20 IoT projects, shares practical expertise and insights on connected software development, AI integration, and intelligent solutions.
We are open to seeing your business needs and determining the best solution. Complete this form, and receive a free personalized proposal from your dedicated manager. Sergei Vardomatski Founder
Kick Off With Your Project Today
