Skip to main content

Managing expectations

Instruction iconUnit
Complete iconIn Progress

Every ML professional knows the scenario: A leader hears ‘machine learning’ and expects instant, flawless intelligence that will transform the business. The danger? Expectations like these can derail even the best project if not managed early.​

Unrealistic expectations are one of the leading causes of project failure. As an ML professional, your role isn’t just to build models— it’s to guide stakeholders towards realistic goals and timelines. Clear, consistent expectation management prevents surprises, reduces conflict and ensures your project delivers measurable business value.

Managing expectations banner

The ‘pre-mortem’ meeting​

A pre-mortem flips the script: You assume the project has already failed and work backwards to uncover the reasons. This exercise forces the team to think beyond optimism and anticipate hidden risks before they become roadblocks.Questions to ask in a pre-mortem:

  • What assumptions might prove false?
  • Where are the weak spots in our data or infrastructure?
  • Which stakeholders might lose interest or block progress?
  • What external changes (regulations, market shifts) could threaten success?

Continuous communication​

ML projects rarely collapse because of one dramatic failure. More often, they unravel slowly through a series of small misalignments. Continuous communication closes these gaps.

Establish a predictable rhythm of updates:

  • Weekly email summaries with key wins, blockers and next steps.
  • Shared dashboards that visualise progress.
  • Short check-in meetings with high-influence stakeholders.

Practical steps for proactive expectation management​

1. Start with a kick-off​

Define success metrics, clarify scope and outline potential risks from day one. Written alignment prevents future ‘but I thought we agreed…’ conversations.

2. Provide regular updates​

Weekly updates show transparency and keep trust high. Even small updates like 'hyperparameter tuning is 70% complete' show that progress is purposeful.

3. Say ‘no’ with a ‘why’ and an alternative​

Saying ‘no’ flatly risks damaging trust. Instead, explain your reasoning and propose a workable alternative. Example: ‘Including this feature now would delay deployment by four weeks. What we can do is launch the current version next month and schedule this for the next release.’

Case study: Deploying an AI-powered medical imaging model​

You’re leading a project to deploy an AI model that helps radiologists detect lung disease.

  • Pre-mortem: The team identifies poor generalisation and weak audit trails as risks. Solutions include a single-hospital pilot and added integration checkpoints.
  • Continuous communication: Weekly updates track accuracy. When a scanner change causes a dip, retraining plans are flagged immediately.
  • Kick-off alignment: Metrics set to reduce false negatives by 15% in six months at the pilot hospital.
  • Saying ‘no’ with an alternative: A physician requests an added feature. You propose adding it in phase two to protect the pilot launch.Key takeaway: Expectation management isn’t about lowering ambition — it’s about channelling it into achievable results.

Action item: Pause and reflect — handling expectations under pressure​

Imagine this: You’re leading an ML project, and a senior executive asks for deployment two weeks earlier than planned. At the same time, your engineering team warns that rushing will compromise accuracy.

  • How would you respond to the executive in a way that protects the project while maintaining trust?
  • Which technique would help you most in this situation: Pre-mortem insights, continuous updates or saying ‘no’ with a ‘why’ and an alternative?

Your response to the executive:

The technique that would help you most: