For the past two years most of the attention in AI has gone to models. The conversations our data employer panels are having this year are about something less glamorous. When an AI initiative stalls, the cause is increasingly traced not to the model but to the pipeline underneath it: data that is late, inconsistently defined, poorly documented, or simply not what everyone assumed it was.
The bottleneck has moved
Several employers on the Data & Analytics council described the same sequence. A promising prototype is built on a carefully prepared extract. The team moves to production, connects to live sources, and performance drops. Weeks are then spent discovering that a field changed meaning two years ago, that a join silently drops records, or that nobody can say where a key column originates.
We had more modelling capability than we could use. What we did not have was anyone who could tell us, with confidence, whether the data going in on Monday was the same as the data we tested on.
None of this is new to experienced analytics engineers. What has changed is the cost. When a dashboard is wrong, someone usually notices. When an automated decision is wrong because its inputs drifted, it can run at scale for a long time before anyone does.
What data readiness actually involves
The capabilities employers are asking for map closely to the data engineering and governance areas of the standard:
- Pipelines with quality checks that fail loudly, rather than passing bad data downstream.
- Lineage you can point to under scrutiny, so any figure can be traced back to its source.
- Clear, versioned definitions, so that “customer” or “active” means one thing across teams.
- Documentation of known gaps and what they mean for anything built on top.
- Governance that settles who may see and use what before a model is trained on it.
How the standard evidences it
The Data & Analytics standard assesses pipeline hygiene directly. At BIPS Practitioner, candidates show quality checks and lineage on work delivered under review. At BIPS Professional, they own an analytical domain and answer for the decisions that rest on it, which increasingly includes the automated ones. At BIPS Specialist, engineering is one of the recognised specialisms, evidenced through reviewed work that others depend on.
In the professional discussion, assessors routinely ask candidates to walk through a pipeline failure they caught, or one they missed. The point is not to find people who have never had a problem. It is to find people who built the checks that surfaced it.
For organisations, the implication is straightforward. Investment in models without matching investment in the people who make data trustworthy tends to produce impressive prototypes and disappointing production systems. The Analytics Engineer and Data Governance Lead roles are no longer supporting functions for AI. They are the conditions for it working at all.