Upgrade & Secure Your Future with DevOps, SRE, DevSecOps, MLOps!
We spend hours scrolling social media and waste money on things we forget, but won’t spend 30 minutes a day earning certifications that can change our lives.
Master in DevOps, SRE, DevSecOps & MLOps by DevOps School!
Learn from Guru Rajesh Kumar and double your salary in just one year.
Almost every team that works with machine learning has a notebook it is slightly embarrassed about. It forecasts something useful, it runs when somebody remembers to press the buttons, and everyone agrees it should become a proper scheduled job “soon”. Soon turns into a quarter, then a year. The model is rarely the obstacle. The obstacle is that the task of moving a notebook into production has no obvious first move, so nobody ever picks it up on a Tuesday afternoon.
This article explains, in everyday language, one remedy that works surprisingly well for machine learning and AIOps work: break the big task into fifteen-minute steps. Not hours, not “a sprint”, just quarter-hour pieces that each leave something visibly better than before. If you have ever stared at a ticket titled “productionise the forecast” and quietly opened a different tab, this one is for you.
Why a big task stalls before it starts
A large task feels heavy because your brain pictures the whole thing at once: data access, packaging, scheduling, monitoring, retraining, documentation, a review from three different teams. Each of those is a real job, and together they form a fog. When you cannot see where the fog begins, you wait for a long, uninterrupted day that never arrives.
Fifteen minutes is a different kind of unit. It fits between meetings. It is short enough that you can start without ceremony and long enough to finish something concrete, such as moving a hard-coded path into a config value. That small finish matters, because a finished step gives you evidence that the project is moving, and evidence is what keeps work alive during a busy week.
What a quarter-hour step actually looks like
A good step has three properties. It has one verb, it has a visible result, and it can be abandoned halfway without leaving a mess. “Improve the pipeline” has none of these. “Read the notebook top to bottom and list every cell that touches a file” has all three: the verb is read, the result is a list, and if your meeting starts early the list is still useful.
When you are unsure whether something is small enough, ask whether you could explain what “done” looks like in one sentence. If the sentence needs the word “and” more than once, cut the step in two.
A worked example: a forecasting notebook becomes a scheduled job
Imagine a notebook that predicts next week’s ticket volume for an operations team. It loads a CSV someone exports by hand, trains a simple model, plots a chart, and saves a file to a shared drive. The goal is a job that runs every Monday without human help. Here is how the first stretch could be sliced.
-
Step one, read and annotate. Scroll through the notebook and add a short comment above each cell saying what it does. No code changes yet. At the end you have a map.
-
Step two, list the inputs. Write down every file, table or credential the notebook touches. Three lines of notes often reveal the scary part of the project, which is usually a data source nobody owns.
-
Step three, pin the environment. Record the library versions the notebook currently uses in a requirements file. This is dull and extremely valuable.
-
Step four, pull constants to the top. Move dates, paths and model settings into one block of variables. The notebook behaves exactly as before, only tidier.
-
Step five, export to a plain script. Convert the notebook into a script and run it once. If it fails, you now know what breaks outside the notebook, which is precisely the information you wanted.
Five quarter-hour sittings is about an hour and a quarter of focused time, spread across a week if you like. By the end, a notebook that nobody dared touch is a script with a map, a pinned environment and a list of inputs. Nothing about that required courage.
The same idea applied to the next stretch
The second half of the project can be sliced the same way. Replace the manual CSV export with a query: one step to write it, one to compare its output against last week’s file, one to swap it in. Add a scheduled trigger: one step to choose where the job lives, one to run it manually from there, one to turn the schedule on for a test week. Add a failure notice: one step to decide who hears about it, one to trigger a deliberate failure and see that the message arrives.
Notice what is missing from this plan: heroics. No step asks you to hold the whole system in your head. The steps are ordered so that each one removes a unknown before the next one needs it.
How to pick the first fifteen minutes
The first step carries a lot of weight, so choose it for its ease, not its importance. Good first steps are reading, listing and measuring. They require no decisions, they produce a note you can show someone, and they build momentum. Bad first steps are “design the architecture” or “agree the approach”, because they are really meetings dressed up as tasks.
If you are working with colleagues, share the first step aloud: “I am going to spend a quarter of an hour listing what this notebook reads.” It is hard to feel overwhelmed by a sentence like that, and people nearby tend to offer a useful fact in return, such as who owns the data export.
Keeping the steps honest
Slicing a task is only helpful if the slices add up. Keep a short running list at the top of your notes, with the steps crossed off as you go. When you finish one, write the next one while the context is still in your head; that way the next sitting begins with a ready-made instruction rather than a blank page. If a step takes thirty minutes, do not scold yourself. Split the leftover into a new step and carry on. The point is to keep the unit small, not to win a stopwatch contest.
It also helps to be clear about what the steps are not. They are not a promise that the whole project fits into an hour. Some things, like waiting for a data owner to answer, have their own clock. A quarter-hour step just keeps your own side from ever being the bottleneck.
Where this connects to decision-making with models
Many readers come to AIOps because they want better decisions, not because they love schedulers. If that is you, it is worth reading how this site frames machine learning as support for IT decisions, because the same thinking applies here: a forecast only helps once it arrives punctually, from a predictable spot, each Monday. Likewise, the piece on predictive maintenance for IT systems shows why small, repeatable pipeline steps matter more than any single clever model.
A short aside on slicing in other places
The habit travels well outside engineering. Marketplace websites are a nice everyday example of work that is built from small repeatable actions. One of them, a website where people list and borrow two-wheelers, routes everyone through a central sign-in, which is itself a reminder that a big service is assembled from many modest parts. That is also a fair description of a healthy production job.
Common objections, answered kindly
“Fifteen minutes is too short to get into the zone.” For deep design thinking, perhaps. But migration work is mostly careful housekeeping, and housekeeping does not need a flow state. If a step runs short of time, the leftover note you write at the end becomes the first line of the next sitting.
“Won’t the overhead of stopping and starting cost more?” Less than you think, provided each step ends with a one-line note about where you stopped. The cost of restarting a vague task is far higher, because you must rediscover what the task even is.
“My manager wants a date.” Quarter-hour steps make estimating easier. Count the steps, assume you will find two or three more as you go, and share a range. A list of thirty small steps is a more credible plan than a single bar on a chart.
A tiny plan you can use today
- Choose the one project you keep postponing.
- Write the very first step so it contains one verb and a visible result.
- Start a countdown for a quarter of an hour and touch nothing but that step.
- Write the next step before you close the document.
- Repeat until the list is short enough that the whole thing no longer scares you.
That is the entire method. It is not clever, and that is the point. Machine learning work gets stuck on logistics far more often than on mathematics, and logistics yields happily to small, patient slices. Pick one notebook, give it a quarter of an hour tomorrow morning, and see how different the project feels once it has a first line crossed off.