You Can't Automate Your Way Out of Bad Data

• 

Put artificial intelligence on top of inconsistent cost data and the result is not better answers. It is wrong answers, delivered faster and with more confidence. The foundation has to come first.

The pitch is everywhere. Add artificial intelligence, point it at your cost data, let it forecast and spot anomalies in seconds rather than a fortnight. There is something to it. But the pitch usually skips the awkward question underneath: what, exactly, is the tool reading?

On most major programmes, the honest answer is a mess. Cost coded differently project to project, missing fields, figures mapped to lines they were never meant to sit in, ERP systems built to run a business rather than account for a project quietly shaping the data before anyone sees it.

The danger is not stupidity. It is confidence.

An earned value curve is only as honest as the cost and progress data feeding it. Book cost against the wrong activity, claim progress a little inconsistently between two projects, and the curve still draws, smoothly and confidently, saying the programme is performing to plan. The line is wrong, but it does not look wrong, which is far more dangerous than an obviously broken chart.

Feed inconsistent data into a forecasting engine and it will do its job faithfully, producing numbers that carry decimal places and land in a board pack with quiet certainty. A human being who senses something is off will hesitate and ask a question. A model will not. It will extrapolate the wrong trend beautifully and hand it upward, and because it arrived from the system rather than a nervous quantity surveyor, people believe it more, not less. Automation does not remove judgement from the process. Handled carelessly, it relocates the judgement somewhere nobody is looking and dresses the output up as objectivity.

A properly founded base

A commercial director on a live rail programme put it memorably: whatever you bring in, bring it in founded on a properly founded base. Get the base right first, assure the data, standardise it, make it genuinely comparable, and only then lay the clever technology on top. Do it the other way round and the organisation has automated its confusion.

There is a second half to that. Bring the team with you. The people managing the tools have to understand what they are actually doing, or one opaque process has simply been swapped for another with a better logo. Upskilling is part of the foundation, not a nice to have bolted on at the end. The instinct that a good commercial team brings, the hard won sense that something does not add up, is worth protecting. Override it without earning trust and the organisation loses the instinct and gains nothing dependable in its place.

Assure the data first. Then automate. Doing it the other way round just industrialises the error.

The sequence is unglamorous but non-negotiable: standardise, assure, then automate. In practice that means one consistent way of coding cost across every project, a single reconciled source rather than a patchwork of ERP exports, and every line assured each month rather than a sample stretched across the rest. It is the sort of work that never makes a keynote, and the whole difference between a forecast a board can rely on and a very expensive guess with good production values.

None of this argues against artificial intelligence. It argues for earning it. The organisations that come out ahead will not be the ones that bolted it on fastest. They will be the ones that did the patient work of making their cost data trustworthy first.

This document was produced with AI assistance. In accordance with the Human Oversight requirements of this AIMS, all AI-generated content has been reviewed, validated and approved by the Director of Black Pear Advisory Ltd before issue.

Dr Martin Perks FRICS MICW MAC

Director, Black Pear Advisory Ltd

Worcester, UK  |  +44 7771 865271