Upgrade & Secure Your Future with DevOps, SRE, DevSecOps, MLOps!

We spend hours on Instagram and YouTube and waste money on coffee and fast food, but won’t spend 30 minutes a day learning skills to boost our careers.
Master in DevOps, SRE, DevSecOps & MLOps!

Learn from Guru Rajesh Kumar and double your salary in just one year.

Get Started Now!

Six Myths About AI Project Milestones, Told Through a Small Team

Six Myths About AI Project Milestones, Told Through a Small Team
Artificial Intelligence

A data scientist once told a room of colleagues that the image-sorting model in question had been “almost finished” for five months. Nobody laughed, because every person in the room had said something similar. Machine learning work has a talent for staying nearly done. Extra effort seldom fixes that; sharper habits around AI project milestones do: cutting the work into checkpoints, marking each one when it truly lands, and letting the pattern of marks tell you when to change course. Below are six myths that keep teams from doing exactly that, told through a small, invented team whose story you may recognise.

Meet the team that could not see its own progress

Picture a four-person group building a demand-forecasting prototype for a regional retailer. They have a shared notebook, a backlog of ideas and a strong sense that things are going well. Yet every Friday, when someone asks how far along they are, the answers wobble: “pretty far”, “about sixty percent”, “depends what you count”. The wobble is the symptom. When progress is not marked, it is not visible, and when it is not visible, nobody can feel it or steer by it.

Over the following weeks, the group tried a different approach. Each myth below is one thing they believed at the start and one thing they learned to do instead.

Myth 1: “Milestones are for project managers, not for people who experiment”

The team’s most experienced engineer resisted checkpoints at first, arguing that experimentation is exploratory by nature and a rigid plan would kill it. There is a grain of truth here: you cannot schedule insight. But a milestone does not have to predict an outcome. It can simply mark a finished piece of exploration, such as “three feature sets compared on the same split” or “baseline model reproduced from last quarter”.

Once the team defined milestones as finished questions rather than promised answers, the resistance faded. Exploration kept its freedom, and the group gained something it had lacked: a list of things it could honestly call complete.

It helped that the first few milestones were deliberately modest. Reproducing an old baseline took an afternoon, yet the tick beside it made the next, harder item feel reachable. Early, cheap wins are not a trick; they build the habit of finishing before the difficult work begins, and they give a sceptical team a first taste of how steady progress feels when it is written down.

Myth 2: “A milestone only counts if the model improves”

This is the most corrosive belief in applied machine learning. If only a better score counts, then every failed experiment feels like lost time, and a week with no score improvement feels like a wasted week. The team rewrote their tick rules so that a checkpoint could be a result, a negative result, or an unblocking step.

The negative result turned out to matter most. When a promising seasonal feature added nothing, the team ticked “seasonality hypothesis closed” and moved on without lingering regret. A short, explicit tick turned a disappointment into a finished unit of work, and it prevented the same idea from being quietly re-tried a month later by someone who had forgotten the outcome.

Myth 3: “Big checkpoints motivate; small ones are busywork”

The group’s first plan had four giant milestones, each lasting a month. Between them, nothing felt finished. Motivation sagged in the long middle stretches, which is a familiar pattern in any long effort. When they sliced the same plan into checkpoints of one to three days, the board filled with visible movement, and Friday’s question became easy to answer.

A useful size test: if you cannot tell whether a checkpoint is done by looking at one artefact, such as a notebook cell, a chart or a short note, it is too large. Small milestones exist to hold your focus and refill your energy, and they were never meant to generate paperwork.

Myth 4: “If everything is on the board, everything is progressing”

A crowded board can look like momentum. In reality, the team discovered, it often hid a pile of items that had been “in progress” for weeks. They introduced a plain rule: a card only moves to the done column when its checkpoint has a visible output, and anything untouched for a week goes back for review. Nothing dramatic, but the effect was instant. The done column grew in a way that matched what people actually felt, and the stuck items became impossible to ignore.

Readers of this site who like to understand how machines interpret data may enjoy the explanation in a gentle walk through how image classification works; the team used that kind of staged thinking when they broke their own pipeline into steps they could mark off.

Myth 5: “Ticking things off is just for morale”

Morale is a real benefit, but it is not the only one. A trail of marked checkpoints is also a measuring instrument. After three weeks the team noticed that data-cleaning checkpoints took twice as long as planned while modelling checkpoints took half. That single observation changed the plan: they shifted a colleague toward the data work and stopped treating the modelling phase as the bottleneck.

Without marks, that insight would have stayed a vague suspicion. With marks, it became a fact they could point to, and facts are much easier to act on than moods. The tick is a tiny data point about how the work really behaves, and the team’s own project was the first dataset they learned to read properly.

Myth 6: “When the milestones are all ticked, the project is finished”

The final myth is subtler. Ticking every checkpoint tells you the plan was completed, not that the plan was right. The team scheduled a short review at each quarter-point of the project to ask two questions: are the remaining checkpoints still the right ones, and did any finished checkpoint turn out to be less useful than expected?

In their case, a dashboard checkpoint was dropped after a stakeholder conversation showed that nobody would open it, and a new checkpoint about data freshness was added in its place. Because the plan was already made of small pieces, changing it cost a few minutes instead of a painful renegotiation. That flexibility is a quiet gift of small milestones: they make mid-course correction ordinary.

A simple way to start tomorrow

  • Write the finished state in one sentence. For the forecasting prototype it was “a notebook that produces a weekly forecast with a documented error range”.
  • List ten checkpoints, each with a visible output. If an item has no output, rewrite it until it does.
  • Tick only finished work. Started is not done; a tick has to be earned.
  • Allow negative results as ticks. A closed hypothesis is progress.
  • Review at set points. Ask whether the plan still matches the goal, and change it if it does not.

Nothing here calls for special software. One shared table is enough, and some teams simply use sticky notes on a wall. What matters is that the marks exist, that they are honest, and that someone looks at them regularly.

Beyond the notebook: small wins in everyday planning

The habit travels well. Someone planning a road trip can mark each leg of the preparation: dates fixed, route drafted, transport arranged. If the transport involves a rental, there are websites built for exactly that, including MotoShare, an American website for renting bikes and cars from private owners, with separate screens for listing a vehicle and for searching and booking one. The site is only one example of a service with clearly separated stages, and each stage is its own small milestone to tick.

Back at the retailer, the forecasting prototype shipped three weeks later than the team’s original gut feeling and four weeks earlier than their pessimistic one. More importantly, nobody wobbled on Fridays anymore. They could point at the board and say, with a straight face, exactly how far they had come. The Friday question, which used to cause a flutter of anxiety, became the shortest part of the meeting.

That is the real argument against the six myths. Progress that is visible is progress you can protect, discuss and learn from. For a field where “almost finished” is the default state, a row of honest ticks is perhaps the most practical tool a team can add. If you want a related read on deciding where machine predictions help most, try the short piece on how computers guess at market futures and then go back to your own board with a pen in hand.

Subscribe
Notify of
guest
0 Comments
Oldest
Newest Most Voted
0
Would love your thoughts, please comment.x
()
x