This is the case of the abandoned AI pilot, the second file in our Data Detectives series, and it is one of the most common mysteries we’re asked to investigate. The pilot is rarely the problem, so the interesting question is what happened between the demo and the rollout.
The evidence
The trail usually looks something like this:
- The pilot ran on a carefully prepared extract, while production data sits across several systems in different formats
- Access, security and ownership questions were never settled, because nobody needed to answer them for a demo
- Nobody tested data quality at scale, so the first real volume of data brought the first real problems
The usual suspects
The model, the infrastructure and the budget all get questioned, along with whether the business ever really had the appetite. Each of them has an alibi.
The pilot worked because the data had been prepared by hand, and the foundations needed to supply that data reliably, repeatedly and safely didn’t exist. A demo only has to work once. Production has to work every day, on data nobody has tidied up in advance.
A lead worth following
A global manufacturer with thousands of SKUs across international markets had plenty of AI enthusiasm. Teams across the business were experimenting with AI tools on their own, without coordination, oversight or any shared strategy, which meant plenty of activity but limited scalability and missed opportunities for impact across functions. Underneath it all, inconsistent metadata, disconnected systems and variable data quality were holding everything back.
Amplifi carried out an AI-readiness assessment, built on the DCAM model, with structured interviews across digital, data and business leaders. It evaluated six dimensions of AI maturity and gave the manufacturer an average score of 2 out of 6, which established a clear baseline of where things stood.
The recommendations are worth noting, because most were about foundations rather than technology. They included defining an enterprise-wide AI vision, forming a cross-functional steering committee, standardizing metadata and integrations, building AI literacy through role-based training and outlining a Responsible AI governance model. From there, Amplifi developed a phased three-to-five-year roadmap that starts with strategy, governance and pilot execution before moving on to infrastructure, skills and embedding AI into core workflows.
The assessment also identified three to five high-value pilot opportunities, including automated data enrichment, marketing content optimization, demand forecasting, defect detection and personalized recommendations, along with over 110 product data recommendations aligned to the roadmap. Early pilots have already delivered faster product attribute completion, improved creative content generation and measurable efficiency gains. Read the full case study.
Your first lines of inquiry
A good detective inspects the scene before the crime, not after it. Three moves will help your next pilot survive contact with production:
- Visit the scene of the future crime. Before the pilot starts, ask what data it will need in production, not just for the demo, and where that data will come from.
- Test the alibi under pressure. A hand-prepared extract is the demo’s alibi, and it won’t hold up in production. Test the data supply at realistic volume and variety, and settle access and security questions while the stakes are still low.
- Name the keeper of the case file. Agree who will own and run the solution once the pilot ends. An operating model settled in week one is far cheaper than one discovered in month six.
Case closed? Almost
The pilot was never the culprit. It showed what the organization could do on its best day, and the foundations decide whether it can do the same on an ordinary one.
Our guide, Data Detectives: The Case of Trusted Data Foundations, follows five common cases from incident to root cause, with three practical moves for each and a self-check to test the strength of your own evidence.
If you’d like help working out where your own pilots stand, our Data Capability Assessment measures your data capabilities across six core areas and shows how ready your organization is to deliver on its data and AI ambitions. You get an evidence-based baseline, clear recommendations and a prioritized action plan, not a slide deck.
Learn more about the Data Capability Assessment
Or talk to a data expert about getting your next pilot into production.