How I showed a global energy company where to automate, use AI and keep engineers in control
posted 22nd October 2026
A roadmap that showed engineers where to automate, where AI could help and where they had to stay in control
The challenge
Engineers needed reliable details of every component in each well to keep it safe and productive for decades. The data was inconsistent, hard to find or missing. That led to deferred production, lessons lost between regions and costly late fixes.
What I did
I led a six month discovery that followed well data from design to long-term integrity. I mapped what engineers did with data at each step, what they decided with it, and where it was copied, reworked or lost.
The outcome
I reframed a maintenance problem as a data capture problem that started at design. The company got a roadmap showing what to fix first, what to automate, where AI could help and where engineers had to stay in control, with a human-in-the-loop service model to match.
The challenge
Integrity and reliability engineers keep wells safe and productive for decades. To do that, they need accurate details of every component in each well.
That data was often inconsistent, hard to find or missing. Engineers struggled to plan maintenance ahead or to diagnose production problems as they happened. The cost showed up as deferred production, lessons that didn’t travel between regions and expensive late fixes.
Part of my brief was to find opportunities for standardisation, automation and responsible AI.
What I did
I led a six-month discovery that followed well data across the whole life of a well: design, drilling and completion, production and long-term integrity.
I didn’t start with the technology. I started with what engineers did with data at each step, what they decided with it and what they needed to feel confident in those decisions. I spoke with engineers at every stage and asked them to show me the documents, spreadsheets and databases they really used.
I mapped the service with an extended blueprint, the approach I now call Automate, Augment, Decide. Alongside the usual actions and touchpoints, the front-end tracked the data, the decision and the hand-off at every step. The back-end showed where data was stored and where it left the official systems.
I ran regular show and tells and hands-on workshops to make the work visible, work through hard problems together and keep stakeholders aligned.
Design, Drilling and Completion Engineers
Production Engineers
Integrity & Reliability Engineers
Focus
Delivering safe, reliable wells
Focus
Getting the most from active wells
Focus
Long-term safety and reliability
What they needed from the data
- To access field data to inform safe well design and drilling
- To produce handover documents quickly and get on with the next well
What they needed from data
Fast access to well details to diagnose problems and reduce deferred production
What they needed from the data
Granular, current details of components and configuration, often for older wells
"Pulling together handover docs in Word and Excel is so manual and time-consuming. There must be an easier way to do this."
“Final designs are stored in one database. Maintenance updates are stored somewhere else. We don’t always know if we’re looking at the most up-to-date data.”
“It can take days to clean data before we can diagnose an issue on an individual well. And we can only deal with one well at a time.”
What the blueprint showed
A weak hand-off at handover
Design teams collated well data by hand at the end of each project. It passed on as static Word, Excel and PDF files, without the reasoning behind design choices. Design teams also had little sight of what production and integrity engineers needed later.
Forks below the line of record
Final designs and maintenance updates lived in separate databases. Engineers couldn’t be sure which version was current, so they checked and reworked it.
The biggest critical moment came last
Integrity engineers spent days cleaning data before they could diagnose one well. That’s rework from low trust feeding a high-impact safety decision.
No loop back to design
There was a formal process to hand a new well to production. There was little to feed learning from production and integrity back into the next design.
The reframe
At first this looked like a maintenance problem. Engineers said they couldn’t diagnose wells because they didn’t have a reliable picture of the equipment inside them, especially in older wells.
The blueprint showed a systems problem. The data problems didn’t start with maintenance. They started at design, where data capture from suppliers was patchy and inconsistent.
I reframed it in three ways:
- The handover pack is a working document, not an archive. Engineers would use it again and again over the life of the well.
- Static documents have limited value. Standard, structured data can be searched, reused and acted on.
- Working one well at a time stops learning from spreading. Insights can’t be shared across regions or fed into new designs.
That moved the conversation from a local maintenance headache to how data is captured across teams and regions, which affects performance for the whole life of every well.
Automate, augment, decide: what I recommended
I sorted every step on the blueprint by the kind of data, the kind of action and whether a decision was made there.
Fix the data first - Automation and AI on patchy data would only spread the problem
Consistent data sets and decisions defined for each stage gate, backed by data standards, data models and governance
Automate - Binary data, transactional steps, no judgement
Auto-fill well IDs and supplier details at design. Auto-generate standard well and regional reports. Connect tools so the same data isn’t typed in several times.
Augment with AI - Problem solving across large volumes of unstructured data
Summarise long reports. Find wells with similar histories. Flag risks in a draft design using historical failure data.
Decide - High-impact safety and design decisions that rest
Engineers make every final call. They can see how a recommendation was made, and accept, edit or reject it. Every choice is logged.
Three AI-supported services
- Handover pack
AI drafts a handover summary with the design rationale at each stage and the next steps. The design engineer reviews, edits and approves it. Production and integrity teams get the data they need from day one.
- Well performance and maintenance
Production flags an issue. The integrity team searches for wells with similar histories. AI suggests patterns and flags risks. Integrity engineers filter by context and plan maintenance, so a lesson from one well can be applied across a region.
- Design sense-check
An engineer uploads a draft design. AI highlights known risks from past wells. The engineer accepts, edits or rejects each one, and every option is logged for review. Learning from the field reaches the next design.
Results
The discovery gave the company a quantified roadmap to better-performing wells and a service that could scale, built on reliable data. The commercial savings are confidential.
Engineers reported faster access to reliable data, more confidence in their decisions and less time on manual tasks.
The human-in-the-loop model meant engineers stayed in control. That mattered, because they valued transparency and control more than “smart” features.
What I learnt
- Start with the people behind the data. Knowing who needed what data, and why, is what let us define standards.
- Follow the data to the decision. Mapping decisions, not just systems, showed where automation fitted and where it didn’t.
- Fix the data before adding AI. Moving from static documents and linear hand-offs to structured, shared data is what made automation and AI possible.
- AI isn’t one size fits all. It earned its place on complex, ambiguous tasks, not routine ones.
- Technology alone doesn’t scale a service. Buy-in and building capability in the teams do.