Atif Ashraf spent five years studying what really happens when workers pick up a written procedure. Using hard-hat cameras, 1,422 coded procedural steps, and interviews with operators at a petrochemical facility, his research found that about one in three steps was adapted, and every task still got done. The result is the RES framework (routine, efficiency, safety), built on the workers’ own words. Atif recently completed his PhD in industrial and systems engineering at Texas A&M University and works at DEKRA. We talked about how three days of driving Todd Conklin around Doha led to a PhD, why “why didn’t they follow the procedure?” is the wrong question, and what this means for a safety advisor whose job is to reduce deviations.
Read the paper: Towards a Worker-Centered Framework for Categorizing Procedural Adaptations, Safety, 2026.
Chris: Tell me a bit about yourself.
Atif: I’m a chemical engineer by training, originally from Kerala, in southwest India. I did my bachelor’s in chemical engineering at Universiti Teknologi PETRONAS in Malaysia, and my master’s at Texas A&M’s campus in Qatar, where I specialized in process safety.
What first got me interested was an eight-month internship with Petronas, Malaysia’s national oil company, during my undergrad. I joined their HSE unit a couple of months after the Deepwater Horizon in 2010. As the new intern, they needed something for me to do, so they asked me to review their emergency response capabilities. If we had a similar blowout, could we handle it? I also got to sit in on incident investigations, the what-ifs and the five whys. That was my first real look at safety.
In Qatar, I got into process safety research through the Mary Kay O’Connor Process Safety Center at Texas A&M. That’s where I went deep into the technical side: risk management and risk assessment. We worked with many oil and gas companies in the Middle East. They’d come to us with a problem, and we’d build a research proposal to help solve it. Part research, part consulting.
Chris: If you ran into a cousin at a family dinner and he asked what you’re up to, and why safety, what would you tell him?
Atif: For me, safety is about understanding people. It’s like any relationship. If you care about someone, you don’t want them hurt in any way, physically or emotionally. You try to put yourself in their shoes and understand what they’re dealing with. So safety goes beyond “are you following the rules or not.” It’s about understanding how people work, helping them do their job, and designing the work so that being safe isn’t an extra thing they have to do.
Coming from process safety, my world was very technical. In my previous job, we hosted a process safety conference in Qatar every year. I’d talk with HSE vice presidents and managers, and I kept hearing the same thing: I don’t understand why they won’t just follow the procedures. I’d ask why they thought that, and the answer was always that they just had to follow the procedures. That got me thinking. I had no idea there was a whole world within safety about understanding people and how they work.
Around 2018 or 2019, companies in the region started asking us to assess their safety culture and climate. Our team was all chemical engineers, so we collaborated with Dr. Stephanie Payne, an I/O psychologist at Texas A&M who specializes in safety climate. Each year the conference had a theme, and that year it was human error. They brought in Todd Conklin.
Chris: You got to experience the Todd.
Atif: I had never heard of him. My boss and I were like, who’s this guy? Then my boss told me to take care of him while he was here. So for three days I was picking him up at his hotel, driving him to the venue, showing him around Doha, getting one-on-one time with him, and I had no clue who he was.
He asked what I was interested in. I was three years out of my master’s, figuring out what I wanted to do, and I’d just done those safety culture assessments. I told him I wanted to understand why people do what they do, from a safety perspective. He said that intersection of process safety and human behaviour was really interesting and that I should do a PhD. He had me read Kahneman’s Thinking, Fast and Slow and Norman’s The Design of Everyday Things. We’ve stayed in touch ever since. Just by knowing Todd, I got introduced to a lot of people in this world, like Ron Gantt.
Chris: You’ve met Ron?
Atif: Not in person, but we’ve chatted and connected on LinkedIn. Funny enough, one of the first articles of his I read was on the Safety Differently website. That was in the first year of my PhD, and it would be really cool to publish something on this site someday. Then your email arrived. It’s a bucket list item for me.
Chris: Your paper opens with a gap. Safety science has a rich vocabulary for how work fails, but not much for how it succeeds. When did that gap become obvious to you?
Atif: It started with that question of why people don’t follow procedures. Nobody I talked to had a concrete answer beyond “why can’t they just follow them?” I was fortunate to find a professor researching exactly that, with funding for a project. I never imagined moving to the US for a PhD, and I had a very comfortable life over there, but the opportunity was there, so I uprooted my life.
When I started the literature review, I came across Erik Hollnagel‘s Safety-II work. He explains that we have all this vocabulary for failure, the human error taxonomies. But if you want to describe how work goes right, there’s very little, because doing it right is just what you’re supposed to do.
Chris: The best Hollnagel line I’ve heard was when I saw him speak in St. Petersburg a few years ago. He said, more or less, that trying to learn how to have a successful marriage by only studying divorced people isn’t going to work. I’ve probably mangled it, but it stuck with me. You spent five years on this. What kept you going?
Atif: Partly my old boss. Before I left, he told me to make sure I effin’ finished this. I thought, why would he even say that? Of course I’ll finish. Two years in, I was applying for jobs and ready to go back. My wife and I moved during COVID in 2020, before vaccines. It was a really hard time.
What kept me going was the topic. As cliché as it sounds, I really wanted to understand why people don’t follow procedures. I knew I wouldn’t get all the answers, but I thought I could get some sense of it. And I found very little empirical work comparing what the procedure says with what people actually did, work-as-imagined versus work-as-done.
Chris: Your reference list includes the 4D work. How did you come across it?
Atif: Our research group focuses on procedures, and we work closely with Rob Fisher of Fisher Improvement Technologies. We developed a procedure writer’s course with him. He knows my work. When I told him what I was planning, it was very academic, because a PhD needs that rigour. He told me I needed to simplify it so people could understand it. He suggested the 4Ds as a method of inquiry.
I read what I could find, and it made sense immediately. I wanted to talk with the people who do the work, and instead of complicated, sophisticated questions, I got something straightforward. And it’s their definition of dumb, dangerous, difficult, and different, not mine. I finally met Jeff Lyth at a conference a couple of months ago and told him I’d used the 4Ds in my research.
Chris: You also cite Cooper’s critique that Safety-II lacks tangible tools. How much of this work is a direct answer to that criticism?
Atif: I read a lot of the back-and-forth, and I see his point of view. One reason I cited him is that he’s argued there isn’t enough empirical evidence behind this way of thinking.
Chris: One of my biggest issues is that practising HOP takes more contact: supervisors with workers, workers with each other, not just people interacting with forms and documents. Let’s get into the paper. What is the RES framework in plain language?
Atif: When workers do something differently from a written procedure, instead of asking why they didn’t follow it, you ask why they did it the way they did. RES answers that in three simple buckets.
R is routine: the way we do things. The procedure says one thing, but the work has evolved into an established way of doing it. You can even see differences between sites in the same organization, depending on who did the training and how those people work together.
E is efficiency: work smarter, not harder. The worker has found a more efficient way to do the task, like reordering steps, while still staying within safe limits.
Chris: Can I jump in? We tend to talk about procedures very linearly: A then B then C. Work isn’t experienced that way. The environment pushes things out of sequence, and sometimes things happen at the same time.
Atif: S is safety: trust but verify. Safety is a slightly odd word for a framework. But I already had R and E, and I wanted it to spell RES as a nod to resilience engineering, which I borrow a lot from. Safety was the first bucket that came together, so we stuck with it.
Chris: And those phrases- the way things are done, work smarter, not harder, trust but verify- came from the workers. How did that happen?
Atif: It came in two parts. First, we collected video from two sites using cameras attached to workers’ hard hats. My research team collected it, not just me. Back in the lab, the question was how to turn what we were seeing into something we could analyze systematically. That’s where the Skip-Order-Action framework came from. For each step: was it skipped? If not, did they do it in order? If it was in order, did they do the action the procedure actually describes? A break in any of those means work-as-imagined doesn’t equal work-as-done. For them to match, all three have to line up.
Chris: So you put cameras on hard hats. How did you get workers comfortable with that? And when I look at work-as-imagined versus work-as-done, the tendency is to push work-as-done toward work-as-imagined, when sometimes it’s work-as-imagined that needs to shift.
Atif: Credit for convincing the company goes to my advisors. The company was moving from paper to digital procedures and wanted to understand best practices and see the value in that shift, so they were motivated. There was a lot of back-and-forth to make sure everything was anonymized, and any research like this on people requires institutional review board approval, so all that was in place. We also had to reassure the workers that this was for science and had nothing to do with the company.
Chris: What kind of work were they doing?
Atif: Filling operations. Loading chemicals into vessels like rail cars and trucks. The nice thing about a camera on a hard hat is that it’s out of their peripheral vision. After ten or fifteen minutes, they’re just doing what they do every day. I looked at routine, normal work, and I didn’t follow them around. Sometimes a camera stopped recording, and we lost data, but I wanted it as naturalistic as possible. You can tell they forgot about it, because in some videos you hear workers complaining about their managers and the other guy reminding them, dude, there’s a camera.
Chris: 1,422 steps, three coders, ELAN software. What was that actually like?
Atif: Three years of my life. Literally. I went through the videos frame by frame, working out what they were doing against which step of the procedure they were on, because sometimes they’d jump around. It was COVID, so I was working from home, and my wife would see me watching video at half speed, pausing to catch the exact moment. She asked if that was why we moved so that I could watch videos every day. And the cameras didn’t help. When a GoPro gets corrupted, it splits the recording into smaller files, and they weren’t always in sequence. I had to rebuild the order of events from cues, like which guy was wearing the black jacket. That whole process was crazy.
Chris: The headline number: 32.9 percent of procedural steps showed some adaptation, about one in three, and every task was still completed successfully. What did you expect going in?
Atif: Before this, we did an interview study with workers about procedure use. Based on their own experience, they estimated roughly 30 percent.
Chris: Workers know, don’t they?
Atif: Qualitatively, yes. Then we did the video analysis, and it came out around the same range across both sites. So in the work we studied, about one in every three steps involved some level of adaptation.
The crux is that every time they complete these procedures, they achieve the goal. The vessel gets filled and goes out for delivery. Nothing goes wrong, so the supervisor assumes everyone is following the procedures. Then, God forbid, something goes wrong, and they dig in and find this guy didn’t follow these steps, and that guy didn’t follow that one. That’s where the idea comes from that procedures aren’t being followed. The effort goes into understanding why the procedure wasn’t followed, not how the work has always been done.
Chris: Did you look at whether the procedure was worth following in the first place?
Atif: When we presented the results, the company couldn’t believe it. One in every three steps? They thought they had the best procedures and the best system. What surprised me was the disconnect between the workers doing the job and the supervisor. They were in two completely different realities. And this was a site with a good safety climate, psychological safety, and good rapport between the supervisor and workers.
Here’s an example. When they finish loading a vessel, they remove the load hose and plug the opening with a shower cap so it doesn’t get contaminated. The procedure says to lift the load hose, then get the shower cap. What workers actually did was put the shower cap next to the opening before they even started. I asked why. They said by the time you lift the hose, walk over for the cap, and come back, the load could get contaminated. If that happens, the whole load is gone, you refill it, and now you’re behind schedule. So I asked why the procedure was written that way, and they said, oh, it’s this guy.
Chris: Oh, it’s this guy. I know exactly what you mean.
Atif: And if the supervisor were out in the field that day, they’d do it exactly as written. The whole PhD fits in a shower cap. The procedure said grab it after. The workers grabbed it before. One of those is written down, and the other one works.
Chris: Would it work that way too?
Atif: It would still work. They’re taking a precaution, because a contaminated load means delays and production pressure. They’re trying to be more efficient. I asked if they’d ever raised it with him. They said they tried, and he told them this is the way to do it. And he came up through the ranks. It’s not like he didn’t know the work. That disconnect, at a site with good psychological safety, really surprised me.
Chris: It reminds me of the story about a university that didn’t pave its walkways right away. They waited to see where students actually walked, then paved those paths. You considered a two-category and a four-category version but landed on three. What was wrong with the others?
Atif: With four, we asked whether routine was too broad and should be split into trained workarounds and informal practices that develop within a small group. But the data didn’t support that distinction. We always discussed results with subject matter experts, people who wrote and performed the procedures, plus supervisors and managers. They didn’t consistently see those as fundamentally different types of adaptation, and neither did the workers. At the end of the day, it’s still just the way they do things. You don’t want to get so granular that you lose the nuance of why an adaptation is happening.
I didn’t want to create another taxonomy just for detail’s sake. The categories are meant to reflect something workers recognize in their own work. They also aren’t mutually exclusive. An adaptation can be efficient and safety-enhancing at the same time. The classification looks at the primary function. It’s not about neatly putting things in boxes. It’s a useful way to understand why an adaptation happened in the first place.
Chris: Those categories rely on workers explaining themselves after the fact. How confident are you that the stated reasons match the real reasons?
Atif: This was the second stage. The first time, we collected video, analyzed it in the lab, and talked with the SMEs, which gave us an idea of the buckets. The second time, I recorded the work and then sat with the workers at the end of the shift. I started with the 4D questions, then showed them parts of the video and asked, why did you do this here? What was going on? So yes, it’s retrospective. But doing it the same day, showing them their own footage, on work they do every day, is the closest you can get to closing that gap.
Chris: You were on top of it.
Atif: As a PhD student should be.
Chris: Or a doctor, now. Congratulations.
Atif: Thank you. I work at DEKRA now as their innovation research lead, looking at DEKRA’s solutions, how they can be improved, and coming up with new ideas.
Chris: A safety advisor reads all this and says, fine, but my job is to reduce deviations. What do you say to them?
Atif: The idea isn’t to launch another program to collect a pile of new data or set up another committee. It’s to use the conversations and observations you’re already having, just a bit differently. Say you see a worker deviate from a procedure during a field observation, an audit, or an investigation. RES gives you the vocabulary to ask what that adaptation is telling you. Is this a routine adaptation? Is the worker finding a more efficient way to do the task without compromising safety? Or are they adding an extra step to double-check, even though the procedure doesn’t ask for it? And if so, does the procedure need to include it? That can give you a lot of information without costing more. It’s the cheapest improvement you can make: asking workers different questions instead of just asking why they didn’t follow the procedure.
Chris: You suggest adaptation patterns could work as leading indicators. Lots of efficiency adaptations point to workflow design problems; lots of safety adaptations point to a control gap. How does an organization collect that without a bunch of cameras?
Atif: That’s where it gets forward-looking. There may be a logic to how the adaptations show up. A routine adaptation might be where steps aren’t skipped and are in order, but the action is done differently. Efficiency adaptations might show up as many steps done out of sequence. Now picture a world where the procedure, the controls, and the systems are all connected. A worker moves to a position and closes a valve, and the system knows what they did and in what order. Collect that, and you can see where routine or efficiency adaptations are piling up. That’s your cue to go and learn what’s happening there.
Chris: So this affects the procedure writer.
Atif: Not just the writer. It gives the supervisor a better understanding of what’s actually happening on the ground. When work runs every day without incident, the default assumption is that people are following the procedures. At least that was my impression at the sites I visited. Then something goes wrong; they dig deep, find procedures weren’t followed, and the response is more training and discipline.
Chris: They find an incident, look for procedures not being followed, and connect those dots. Your work questions whether those dots are actually connected.
Atif: I’d say so.
Chris: Your paper carefully calls this a hypothesis that needs testing, not a validated taxonomy. What else can you say about that?
Atif: That’s part of doing science. You have to be careful with your claims. This is one company, multiple sites, one industry. How does it translate to industries that aren’t chemical, like aviation or healthcare? We need to test it there. That said, these findings are intuitive. Work smarter, not harder. We all optimize our time and resources in daily life. What I’ve done is put empirical data behind something many people already recognize from the field, which goes back to Cooper’s point about evidence. So the next time someone says people need to follow procedures to work safely, you can point to this paper: in the work we studied, about one in every three procedural steps involved some adaptation, and every task still got done.
Chris: I saw something similar with the field-level hazard assessments crews do in construction. After an incident, the crew’s FLHA gets pulled and judged insufficient. About twelve years ago, I started pulling ten others as a control comparison along with the one from the incident crew. After a couple of incidents, we found it didn’t matter. They were all the same quality. You name the Hawthorne effect as a limitation. Workers knew they were being filmed. How much do you think you missed?
Atif: Reviewers will always ask how you can be sure people didn’t change their behaviour because they knew they were being observed, so you have to mention it in any observational study. I can’t say how much I missed, because I didn’t measure that. But I captured a pretty decent chunk of reality, because the method was so unobtrusive. I wouldn’t recommend it as a routine practice, though. Nobody has the time or resources to put cameras on workers and have someone review the footage every day. It’s really a job for a student who has nothing else to do.
Chris: One criticism could be that it’s a single company and one type of operation. What’s your strongest answer to “this doesn’t generalize”?
Atif: The results are intuitive. They make sense to anyone who reads them. Routine is just how we do things. Work smarter, not harder. Trust but verify. Those phrases came from the workers, not from me, and they resonated with them. Those three ideas would translate to other domains, but that needs testing.
Look at how James Reason’s human error taxonomy developed. He put forward a framework: slips, lapses, mistakes, and violations, and over time people tested it and found more and more validity in it. If someone applies RES to their own work and data, it should line up with what I’m saying. It’ll get tested with time.
Chris: What’s next for this work?
Atif: I’m not sure anyone in the lab is continuing it. I was fortunate to do my PhD on something I was really passionate about, and now when people ask me why workers don’t follow procedures, I can actually talk to them about it. Ideally, what comes next is that connected system I described, where anything a worker does is captured against the procedure: when they did it and in what sequence. I don’t know how that happens. It’s a technology question. I proposed a model in another paper, the last piece of my dissertation, but it’s hypothetical and looking far into the future. It depends on who takes up the work.
Chris: I’m talking to a lot of really bright, passionate students, so I’m very encouraged.
Atif: Thanks for the conversation. I was stressed coming into this. It’s the first time I’ve done anything like it. The only times I’ve talked about my PhD were in formal settings where people judge and question the research, which is just how a PhD works. I really enjoyed this.
Connect with Atif on LinkedIn.
This is valuable empirical work because it confirms what experienced workers already know: written procedures rarely contain the whole reality of work. But I struggle with the RES framing. Routine, efficiency, and “safety” describe why workers believe they adapted, not what the adaptation actually produced.
Completing the task without an incident does not establish that the adaptation was safe. It may have preserved capacity, consumed margin, transferred exposure, or simply succeeded under those particular conditions. We only know the immediate outcome; we do not know what nearly aligned, what accumulated, or whether the same adaptation remains survivable when context changes.
That distinction matters. Adaptation is unavoidable in complex work, but it is neither inherently protective nor dangerous. The more useful question is whether the adaptation maintained sufficient human and system capacity against the hazards of that moment, and how we would know before someone pays for the uncertainty.
The research helps expose the difference between work-as-imagined and work-as-done, but studying work-as-done alone becomes a Sisyphus-like endeavor. Work-as-done is contextual, adaptive, and continually changing. Until we accept and address the conditions creating the separation between the two, we will keep documenting yesterday’s adaptations while new ones emerge today. Turning them into another taxonomy, or a technology-enabled surveillance system, does not resolve the gap. It merely gives us a more sophisticated way to chase it. The real organizational responsibility is to build sufficient capacity, margin, authority, and support for people to survive the differences that cannot be eliminated or predicted.