← Back to Field Notes

March 27, 2026 · The Bloomfield Team

What "AI-Ready Data" Really Means for a 50-Person Shop

Every AI vendor tells manufacturers they need "AI-ready data" before they can start any project. Most manufacturers hear that and assume it means a massive data cleanup initiative, six months of IT work, and a consultant explaining data governance frameworks. That interpretation has killed more AI projects than any technical limitation.

AI-ready data for a 50-person shop means something far simpler than what the enterprise consulting world describes. It means your data is accessible, has enough history to contain patterns, and can be connected to the workflows where AI will actually operate. The bar is lower than you think. The work to get there is more practical than you have been told.

What You Already Have

A typical 50-person manufacturing shop with an ERP system that has been running for three or more years already has most of the data that matters. The problem has never been a lack of data. The problem is that the data lives in places that were never designed to work together.

Your ERP contains job records. Every work order that has moved through your shop, the part number, the customer, the quantity, the material, the operations performed, the hours logged, and the final cost. Depending on your system, JobBOSS, Epicor, ProShop, Global Shop Solutions, or one of the dozens of others, this data goes back years and contains thousands of records.

Your quoting system, whether it is a dedicated tool or a folder full of Excel files, contains the pricing history for every RFQ you have responded to. What you quoted, what you won, what you lost, and often the notes your estimator made about why a particular price was set.

Your quality system contains inspection records, nonconformance reports, corrective actions, and the disposition of every part that did not meet spec. For shops with AS9100 or ISO 13485 certifications, this data is detailed and well-maintained because the auditors require it.

Your purchasing records contain supplier pricing history, lead times, and delivery performance data that your buyers reference every week.

Your email contains customer communications, engineering clarifications, supplier negotiations, and the contextual knowledge that explains why decisions were made.

All of this data is AI-ready in the most important sense: it reflects real operational decisions and their outcomes. That is the raw material for any AI tool.

The Three Things That Actually Matter

Accessibility. Data is AI-ready when a system can read it. An ERP database that allows SQL queries or API connections is accessible. A folder of PDFs on a shared drive is accessible, because modern AI can read and extract information from unstructured documents. A spreadsheet on someone's desktop that no one else can reach is not accessible. The first question for every data source is: can something other than a human being get to this data and read it programmatically?

History. AI finds patterns in historical data. A quoting tool needs enough past quotes to identify relationships between part features, manufacturing processes, and accurate pricing. A quality prediction model needs enough inspection records to distinguish between process variations that produce defects and those that do not. For most manufacturing applications, three years of data provides a solid foundation, and five years is more than sufficient. One year of data limits what AI can do, though it can still deliver value for straightforward applications like automated data lookup and retrieval.

Connectability. Individual data sources become dramatically more valuable when they can be linked. A job record in the ERP becomes useful for quoting when it can be connected to the original quote that won the work, the quality records from production, and the customer feedback after delivery. These connections often exist implicitly through shared fields like part numbers, customer IDs, or job numbers, but they have not been made explicit in any system. Making those connections is a core part of building an AI tool for a manufacturing operation.

What AI-Ready Data Does Not Mean

It does not mean clean data. Manufacturing data is messy because manufacturing is messy. Part numbers get entered inconsistently. Job descriptions use abbreviations that vary by shift. Material callouts reference different naming conventions depending on who entered the order. An AI system built for manufacturing expects this and handles it, because the system is designed around the reality of how shop floor data actually looks, not a sanitized version of it.

It does not mean centralized data. Your ERP does not need to contain everything. Your quoting data can live in Excel. Your quality records can live in a separate QMS. Your tribal knowledge can live in people's heads, documented in conversations and captured systematically over time. The AI tool connects to data where it already lives. Centralizing everything into a single system before starting an AI project is a detour that adds months and cost without adding value.

It does not mean big data. A 50-person shop is not generating terabytes of sensor data from an IIoT deployment. It does not need to. A shop that has quoted 2,000 RFQs over five years and run 4,000 jobs has a dataset that contains enormous value for AI applications in quoting, scheduling, and quality prediction. The patterns are in the job records, the cost variances, the on-time delivery rates, and the rework history. Volume is not the variable. Relevance is.

The Practical Assessment

Assessing your data readiness takes hours, not months. Walk through the following for each data source in your operation.

ERP system. What version are you running? Does it support API access or direct database queries? How far back does your job history go? Are job records complete enough to show the relationship between quoted cost, actual cost, and the operations performed? Can you export a sample of 100 recent jobs in a structured format?

Quoting records. Where do your quotes live? If they are in Excel, can you access the files programmatically from a shared location? Do the quotes reference part numbers or customer identifiers that match what is in the ERP? How many quotes from the past three years are accessible?

Quality data. Where are your inspection records and NCRs stored? Is the data in a dedicated QMS, in spreadsheet logs, or in paper records? For shops with ISO or AS9100 certification, the quality data is typically well-structured because it has to survive audits. That same structure makes it immediately useful for AI.

Supplier data. Where does your purchasing team track material pricing and lead times? Is it in the ERP purchasing module, in a separate spreadsheet, or in email correspondence with suppliers?

Tribal knowledge. Which processes depend on specific people's expertise? Where does the knowledge that keeps your operation running actually reside? Identifying these knowledge concentrations is important because they represent both the highest-value opportunity for AI and the highest operational risk if those people leave.

Common Gaps and How to Close Them

The most common gap we see at 50-person shops is not missing data. It is inconsistent data entry in the ERP. Job descriptions that say "per print" instead of documenting the specific operations. Material fields that contain free-text entries like "4140 HT" in one record and "AISI 4140, heat treated" in another. Setup time fields that were left at zero because the operator forgot to log them.

These inconsistencies do not prevent AI from working. They reduce the precision of the output. An AI quoting tool can still surface the five most similar historical jobs when description fields are inconsistent. It will surface the eight most similar jobs and provide tighter cost estimates when those fields are consistent.

The practical path is to start the AI project with the data you have and improve data entry practices going forward. Most shops are closer to AI-ready than they realize. Waiting for perfect data before starting is like refusing to drive until every road is newly paved. The roads you have will get you where you need to go.

A 30-Day Data Readiness Plan

Week 1. Inventory your data sources. Make a list of every system, spreadsheet, file folder, and person that holds operational data. For each source, note what type of data it contains, how far back it goes, and whether it can be accessed electronically.

Week 2. Test accessibility. Export a sample from each source. Pull 100 job records from the ERP. Collect 50 recent quotes from wherever they live. Gather 30 NCRs or inspection records. The goal is to confirm that the data can move from where it sits today into a format another system can read.

Week 3. Map the connections. Identify the fields that link data sources together. Customer names or IDs that appear in both the ERP and the quoting system. Part numbers that connect quotes to jobs to quality records. Job numbers that tie work orders to inspection results. These connection points are the skeleton of your AI data infrastructure.

Week 4. Assess and prioritize. Based on what you found, identify which AI application has the strongest data foundation in your operation. For most shops, that is quoting, because the combination of quote history and job cost data provides a direct feedback loop that AI can learn from immediately. The speed and accuracy improvements in quoting typically deliver the fastest return on the AI investment.

At the end of 30 days, you have a clear picture of what is available, what is accessible, where the connections are, and where to start. That is AI-ready data. No six-month data governance initiative required.