Sewer Spills have been a big subject in the UK for several years. We began looking at this through as a working example of how it can be treated through a bowtie diagram. And then we wondered if an AI couldn’t have done the job for us…

The Hazard is, our intepretation, sewage contained in a gravity-fed combined sewer system, as most are (at least, in the UK). Gravity, and the variable load caused by surface drainage, are the primary factors which combine to make the Top Event quite the problem that it is – when sewage overflows to the environment. That’s a “spill”.

The Human-Generated Version

Below, you can see the first-cut version of our 100% human-generated diagram (to the level of Triggers and Consequences). The principal cause of spills is when the sewer is overloaded, i.e. when it rains and there is a greater volume of sewage than the treatment works has capacity for. This is why permits exist, to allow discharges of untreated but highly diluted sewage (preferably) into rivers when they are in spate.

The four other triggers we’ve included fall into two categories: plant failure and blockages caused by careless disposal of domestic or trade waste.

• Some 80% of blockages that result in spills during dry weather are caused by plastic-based wet wipes flushed down loos.
• A large part of the remainder are caused by restaurants and food processors sending fat down their plug holes, which create “fatbergs”.
• Plant failure and sewer pipe/chamber collapse make up the main causes of dry-weather spills which fall wholly under the control of the water companies, although poor operating practices may also contribute.

On the outcomes side, there are many specific issues.

• Top of the list has to be sewage flooding to homes, which is a terrible thing for those afflicted by it.
• Then there is the impact to human health, which has led to the creation of lobby groups among wild swimmers and surfers (and hence public complaints).
• We singled out three main environmental impacts including fully-fledged NIRS Category 1/2 waterway contamination (National Incident Recording System), over-nutrification, and aesthetic impact.
• Finally, there are the regulatory impacts of fines and poor Price Review outcomes.

Before we went into the control barriers (left-hand side) and mitigation measures (right-hand side), we thought this would be a good moment to see where AI could get us.

The AI-generated version

To make this a fair comparison, here’s how the conversation went:

Us: “Hello, we’re developing a bowtie diagram and would like some help”

AI: “Got it — you want help with a bowtie diagram, the kind used for risk assessment and hazard management. To get started, I need to know:

• The central hazard/event you’re mapping (the “knot” in the bowtie).
• Threats or causes that could lead to it.
• Consequences if it happens.
• Controls on each side — preventive controls (before) and mitigation controls (after).

Once we have those, I can help structure it in clear text form so you can easily turn it into a diagram — or I can mock up a visual for you.

What’s the main hazard or unwanted event in your case?”

Us: “The main hazard is ‘Sewage in gravity fed combined sewer system’ and the top event is ‘Sewage overflow to environment’”

AI (slightly edited): “Threats:

• Heavy rainfall exceeding design capacity
• Blockage in sewer pipe (e.g., debris, grease, tree roots)
• Pump station failure (if relevant to system)
• Illegal connections (e.g., roof drains to sanitary sewer)
• Inflow & infiltration from groundwater

Outcomes:

• Contamination of waterways
• Public health risks (e.g., exposure to pathogens)
• Damage to aquatic ecosystems
• Regulatory fines and reputational damage
• Property damage from flooding

If you like, I can map this into a clean bowtie diagram layout so it’s visually clear — threats and preventive barriers on the left, consequences and mitigations on the right, with the top event in the middle.

Do you want me to prepare that visual diagram next?”

Note: We’ve simplified the above text a little. The AI also proposed a number of preventative/mitigation controls, but we weren’t quite ready for that yet.

Us: “OK let’s see!”

AI (after a few minutes’ wait):

Note: The random text at the end is just as it was presented to us

Us: “That’s pretty awesome!”

AI: “Glad you like it! We can refine it further by adding the preventive and mitigation barriers directly into the bowtie so it’s not just threats and consequences, but also the specific controls in place for each. That way it becomes a complete risk management tool, not just a risk map.

Do you want me to build that enhanced version?”

Us: “Thanks, but we’ve heated the planet up enough for one day.”

Conclusion:

Admit it, the AI did a great job as a starter for ten! But we were glad to have done the thinking first and then look at the AI output as a comparative reference. If it were our starting point, it could have got in the way of our analysis, rather than aided it.

Illegal connections to the system was a nice one, albeit better placed as a control under “Volume Exceeds Capacity”, and we liked “Damage to aquatic ecosystems” as a heading on the right-hand side. The item “Inspection programmes…etc” isn’t a threat, but a control measure of course, and we ourselves broke down the causes and effects down into more specific detail, because that helps to identify effective controls. That’s the kind of stuff we do in workshops with people.

Still it was surprisingly good. Cheers, ChatGPT!

For reference, here’s the whole diagram as we originally created it, using Bowtie Designer. If you open the image in a separate tab, you can zoom in to read it in more detail. If you’re using Bowtie Designer, then it’s possible to make this diagram much more accessible and dynamic. Drop us a note and we’d be happy to share the data file with you (and do the sales pitch as well, but only if you ask).