Milestones, risk and supervision: running a project to a fixed calendar

Year 4 / Module 31 MSci project: scoping, ethics and method / Unit 9 / Milestones, risk and supervision: running a project to a fixed calendar

Milestones, risk and supervision: running a project to a fixed calendar

HCI4107 · Unit 9 of 10

A milestone that names a phase is not a milestone, and the project risk literature shows that a register improves outcomes only when it names the specific threats a project owns, of which ethics delay and recruitment failure are the ones students leave out.

Milestones, risk and supervision: running a project to a fixed calendar

A proposal with ten milestones and no risk register has been dated, not planned. The calendar is the one resource a student cannot renegotiate: the submission date is fixed, the ethics committee meets on its own cycle, and participants arrive at the rate they arrive.

This ninth of ten units argues that a milestone is a milestone only when it names a deliverable another person could inspect and date, and the tenth unit turns to writing and the viva. A risk register improves outcomes only when every entry names a threat this project owns, with a response and an owner. The supervision literature supplies what both omit: the plan runs inside a relationship whose terms are usually unstated.

A milestone that names a phase is not a milestone

Turner and Muller (2002) reconsider the classical definition of a project and conclude that it is incomplete rather than wrong. Uniqueness, novelty and transience, they argue, create uncertainty, a need for integration and urgency of delivery. It is these rather than the control of time and cost that make projects hard. A plan carrying only dates answers the urgency and ignores the uncertainty.

The test of an entry is whether another person could inspect it and decide whether it passed. "By the end of term the literature review is complete" fails, because complete is a judgement nobody outside the project can date. "On 14 November the ethics application is submitted and the receipt filed" passes, because it names an artefact, a recipient and a date.

Why the estimates point one way

Flyvbjerg (2006) documents the inaccuracy of forecasts of project cost and demand. He explains it as optimism bias, the tendency to understate the difficulty ahead, and strategic misrepresentation, the incentive to understate it in a document that has to be approved. His remedy is reference class forecasting, which estimates from the actual outcomes of comparable projects and so declines to ask the planner how long the work will take. The first practical application he reports concerns cost forecasts for large transport infrastructure projects. That is where the method was used rather than only proposed, and it remains an infrastructure case rather than a software project. A student therefore still has to assemble a local reference class.

Do not estimate twenty interviews in six weeks because twenty is the number the method needs; find completed studies in a comparable setting and read what recruitment achieved. If the reference class recruited about one participant per week, the student builds the calendar from that figure, or the study shrinks.

What a risk register does, and what it does not

The empirical warrant for risk planning is real but narrower than the textbooks suggest. Zwikael and Ahn (2010) surveyed 701 project managers and their supervisors in seven industries and three countries. They found that context shapes both the level of risk perceived and the intensity of management applied to it. Their central result is a moderation: risk management weakens the negative relation between risk level and project success, and even moderate planning was sufficient for that effect.

Bannerman (2008) is the corrective. Reassessing risk management in software projects against a study of agencies in an Australian State, he found risk narrowly conceived in the research and risk management under-performed in practice. He concluded that formal project management is neither necessary nor sufficient for success.

The threats a register most needs to name are ethics delay and recruitment failure, because both live outside the parts of the project the student controls. "Risk: ethics approval" names a department, not a threat. "The committee meets fortnightly and accepts no applications in August; if the July deadline is missed the study starts three weeks late; owner, the student; response, submit in June" names a threat, an owner and a response.

Supervision is a relationship with terms

Grant and Graham (1999) report that teaching students to manage themselves and their supervisors strategically improved students' accounts of their experience, and are explicit that working with students alone is partial. That is an argument for a document rather than a technique, because terms have to exist before either party can manage them.

Acker, Hill and Black (1994) studied supervision in three British universities and set a technical rationality model, which gives priority to procedure, against a negotiated order model, in which supervision is open to negotiation. Pyhalto, Vekkaila and Keskinen (2015) surveyed 1,184 doctoral students and 431 supervisors across eleven faculties of the University of Helsinki. They found that a fit between the two parties' beliefs was related to students' satisfaction with their studies. Those beliefs concerned who is involved in supervision, how frequent supervision is and what the supervisor's task is. Halse and Malfroy (2009) describe supervision as professional work with five facets: the learning alliance, habits of mind, scholarly expertise, techne and contextual expertise.

Read together, these findings define the agreement. It records the fit: who does what, how often the project meets, which channel carries questions between meetings, how far ahead a draft must arrive to be read, and what the supervisor will and will not read, with an asynchronous default for a student on a slow connection.

Feedback is not automatically helpful

The supervision literature presumes that feedback helps, and Kluger and DeNisi (1996) show why that presumption is unsafe. Their meta-analysis of 607 effect sizes found an average improvement in performance but a decrease in over a third of cases. Their feedback intervention theory explains why: effectiveness falls as attention moves up from the task towards the self. The average improvement was d = 0.41 over 23,663 observations. The authors report that the decrease cannot be explained by sampling error, by the sign of the feedback or by existing theories. Feedback on the argument of a chapter and feedback on the person writing it are different interventions with different effects.

Carless and Boud (2018) place the remaining work with the student. Feedback literacy, the understandings, capacities and dispositions needed to make sense of information and act on it, is set out as four interrelated features. The four are appreciating feedback, making judgements, managing affect and taking action, and the paper shows them in peer feedback and in the analysis of exemplars. The agreement can specify when a draft arrives and what will be commented on; it cannot specify that the student will revise.

Contested question: does the register do the work, or the relationship?

Two positions can be stated from this unit's sources. On the first the instrument works: Zwikael and Ahn (2010) found that even moderate levels of risk planning reduced the negative effect of risk on project success. On the second it is a weak proxy for the behaviour. Bannerman (2008) found risk management under-performed in practice and formal project management neither necessary nor sufficient, and observed that the projects he studied did not conform to a uniform structure. A register copied from a template can therefore describe a project nobody is running.

Manathunga (2007) argues that describing supervision as mentoring masks the power relations and boundary crossing the relationship contains. The vocabulary of the agreement can therefore perform a collegiality the relationship does not have.

What would settle the instrument question is a study that operationalises the specificity of a register, whether each entry names a threat, an owner and a response, and relates that to outcomes. What would settle the supervision question is a comparison in which one arm states the terms in advance and the other negotiates them as usual, with progress and the students' account of the relationship measured. No source here reports either test.

One further finding supports the sceptical reading. Levecque and colleagues (2017) surveyed 3,659 doctoral students in Flanders. They found 32 per cent at risk of a common psychiatric disorder, significantly higher than in the comparison groups, with work organisation and job demands associated with that prevalence. A workload set by intention leaves that risk outside the register.

Read by kind, the two sides are differently evidenced. The instrument claim rests on one cross-industry survey of 701 managers, and its correction rests on a reassessment supported by a study of agencies in one Australian State. The relationship side is stronger, resting on a meta-analysis of 607 effect sizes, a survey of 1,615 students and supervisors, and a survey of 3,659 doctoral students. While the dispute is open, a student should write each risk entry with a threat, an owner and a response, and set the supervision terms in writing rather than assume them. The strongest objection is Bannerman's: the specificity the unit prescribes has never been the variable in a study of outcomes, and in his projects risk management under-performed as practised, so a named threat and owner may be as weak as the broad entry they replace.

Where the argument is heading

The direction of travel is from broad categories to named behaviours. Chen (2026) applied topic modelling and semantic network analysis to 20,618 narratives from doctoral students and applicants. She identified fifteen supervisory behaviours, arguing that constructs such as support obscure the practices students experience. The behaviours include structured feedback, career guidance, resource provision, digital mentoring, informal interactions and respectful honorifics. For estimates the corresponding move is Flyvbjerg's (2006) reference class of finished projects. Zwikael and Ahn (2010) add that the intensity of risk management that helps depends on context, which for one MSci project means a short register of threats this project owns.

The sources this unit rests on

Each note states what the work did, what it found and how it bears on the argument. Every identifier was resolved on 31 October 2026.

Acker, S., Hill, T. and Black, E. (1994). Thesis supervision in the social sciences: managed or negotiated? doi:10.1007/bf01383939 Qualitative study of supervision in three British universities, setting a technical rationality model against a negotiated order model. It found the negotiated model a better description of practice, so the terms of supervision are settled by negotiation.
Bannerman, P. L. (2008). Risk and risk management in software projects: a reassessment. doi:10.1016/j.jss.2008.03.059 Reassessment of the risk literature supported by a study of agencies in an Australian State. It found risk narrowly conceived in research and risk management under-performed in practice, and formal project management neither necessary nor sufficient, which is the unit's counter-argument.
Carless, D. and Boud, D. (2018). The development of student feedback literacy: enabling uptake of feedback. doi:10.1080/02602938.2018.1463354 Conceptual paper proposing four features of feedback literacy: appreciating feedback, making judgements, managing affect and taking action. It argues uptake depends on the student's capacity to act on information, so the value of supervision depends on what the student does with it.
Chen, H. (2026). From feedback to Emojis: doctoral supervision and doctoral students' well-being in Chinese doctoral education. doi:10.1080/07294360.2026.2655396 Topic modelling and semantic network analysis of 20,618 narratives from doctoral students and applicants. It identified fifteen supervisory behaviours, arguing that broad constructs such as support obscure the practices students experience.
Flyvbjerg, B. (2006). From Nobel Prize to project management: getting risks right. doi:10.1177/875697280603700302 Paper documenting the inaccuracy of forecasts of project cost and demand and explaining it as optimism bias and strategic misrepresentation. It sets out reference class forecasting, estimating from the outcomes of comparable projects, so an estimate formed from intention is systematically optimistic.
Grant, B. and Graham, A. (1999). Naming the game: reconstructing graduate supervision. doi:10.1080/1356251990040105 Account by university staff developers of work with research students and supervisors. It reports that teaching students to manage themselves and their supervisors strategically improved their accounts of the experience, while working with students alone remains partial.
Halse, C. and Malfroy, J. (2009). Retheorizing doctoral supervision as professional work. doi:10.1080/03075070902906798 Life history interviews with doctoral supervisors in five disciplines at a large Australian university. It theorises supervision as professional work with five facets, including the learning alliance and contextual expertise, so supervision is a pedagogy whose practices can be specified.
Kluger, A. N. and DeNisi, A. S. (1996). The effects of feedback interventions on performance: a historical review, a meta-analysis, and a preliminary feedback intervention theory. doi:10.1037/0033-2909.119.2.254 Meta-analysis of 607 effect sizes. It found an average improvement but a decrease in over a third of cases, and proposes feedback intervention theory, in which effectiveness falls as attention moves from the task towards the self.
Levecque, K., Anseel, F., De Beuckelaer, A., Van der Heyden, J. and Gisle, L. (2017). Work organisation and mental health problems in PhD students (the published title uses the American spelling). doi:10.1016/j.respol.2017.02.008 Survey of a representative sample of 3,659 doctoral students in Flanders against three comparison groups. It found 32 per cent at risk of a common psychiatric disorder, higher than in the comparison groups, with work organisation and job demands associated with the prevalence.
Manathunga, C. (2007). Supervision as mentoring: the role of power and boundary crossing. doi:10.1080/01580370701424650 Analysis of the consensus that mentoring is the most effective supervision strategy. It argues that describing supervision as mentoring conceals the power relations and boundary crossing the relationship involves, which is the counterweight to the procedural account of the agreement.
Pyhalto, K., Vekkaila, J. and Keskinen, J. (2015). Fit matters in the supervisory relationship: doctoral students and supervisors perceptions about the supervisory activities. doi:10.1080/14703297.2014.981836 Survey of 1,184 doctoral students and 431 supervisors across eleven faculties of the University of Helsinki. It found the fit between the two parties' perceptions related to student satisfaction with the studies and the relationship.
Turner, J. R. and Muller, R. (2002). On the nature of the project as a temporary organisation (the published title uses the American spelling). doi:10.1016/s0263-7863(02)00020-0 Analysis of the project as a temporary organisation, reassessing the classical definition and viewing the project as an agency for managing uncertainty. It argues that uniqueness, novelty and transience create the pressures of uncertainty, integration and urgency, so a project is defined by uncertainty.
Zwikael, O. and Ahn, M. J. (2010). The effectiveness of risk management: an analysis of project risk planning across industries and countries. doi:10.1111/j.1539-6924.2010.01470.x Survey of 701 project managers and their supervisors in seven industries and three countries. It found that context shapes perceived risk and the intensity of risk management, and that risk management moderates the relation between risk level and project success.

Check your understanding & apply it · 4 questions, 1 essay

1A draft plan reads: by the end of reading week 6 the literature review is complete; by the end of reading week 10 data collection is finished; by the end of term the student is ready to write. Which diagnosis does this unit's argument support?

2A student plans 24 interviews with care-home staff in five weeks because that is the sample the method requires, recruited through the homes' managers. What does this unit predict?

3Two readers disagree about a risk register. The first says the register does the work. The second says the relationship does. Which verdict does this unit's evidence support?

4A student claims that a specific risk register improves a project's outcome and proposes to settle it by counting the entries in twenty registers and comparing that count with the projects' outcomes. Which judgement does this unit's argument support?

5Essay A student's MSci plan reads as follows. Milestones: by the end of reading week 6, complete the literature review; by the end of reading week 10, finish data collection; by the end of term, be ready to write. Risk: running out of time. Supervision: weekly meetings with the supervisor, who will comment on drafts. Method: twenty four semi-structured interviews with care-home staff, recruited through the homes' managers, to be completed in five weeks because that is what the sample size requires. The student adds that the ethics application will go in when the interview schedule is final and that the plan needs no risk register because the project is small. Using this unit's sources, write the assessment you would give. Address the milestone test, the basis of the estimate, the specificity of the register including ethics delay and recruitment failure, the evidence on feedback, the terms supervision needs, and which position you take on the contested question. Name the strongest objection to your own advice.

Begin with the milestones, because they are the part of the plan most easily fixed. Turner and Muller (2002) argue that uniqueness, novelty and transience generate uncertainty, integration and urgency, so a plan that carries only dates has answered the urgency alone. Applied to this plan, none of the three milestones passes the test of being inspectable by another person: the end of a reading week is not a deliverable, and being ready to write is a judgement no one outside the project can date. Rewrite each as an artefact with a date and a recipient, for example that the interview schedule is approved, the ethics application is submitted and receipted, and a coded subset of three transcripts exists. Next, the estimate of five weeks for twenty four interviews is an estimate formed from intention, exactly what Flyvbjerg (2006) documents as systematically optimistic, since forecasts drawn from the components of a plan suffer optimism bias and strategic misrepresentation. The remedy he proposes is reference class forecasting, which estimates from the outcomes of comparable completed projects, so the student should find completed studies recruiting care-home staff in a comparable setting and build the calendar from what they achieved rather than from the sample size the method prefers. Third, the risk register as written is an example of what this unit calls a register performed rather than used: running out of time names a symptom that every project shares rather than a threat this project owns. The two threats a register most needs to name are ethics delay and recruitment failure, and both are visible here. The application is deferred until the interview schedule is final, which puts a committee cycle on the critical path, and recruitment runs through home managers, who may refuse, delay or nominate convenient rather than representative staff. Each entry needs a threat, an owner and a response. Fourth, the supervision arrangement should not assume that comments help. Kluger and DeNisi (1996) meta-analysed 607 effect sizes and found an average improvement but a decrease in performance in over a third of cases, with effectiveness falling as attention moves from the task towards the self, so the agreement should say what kind of feedback is wanted and how far ahead a draft must arrive. Carless and Boud (2018) add that uptake is a student capability, so the agreement can secure the conditions for feedback but not the revision itself. Pyhalto and colleagues (2015) found that the fit between what supervisor and student believe supervision to be, across who is involved, how frequent it is and what the supervisor's task is, related to satisfaction, which is why weekly meetings alone settle nothing, and Acker, Hill and Black (1994) concluded that supervision is better described as negotiated than as managed. Finally, take a position. My advice follows Zwikael and Ahn (2010), who found that even moderate levels of risk planning reduced the negative effect of risk on project success, so a short and specific register is worth writing. The strongest objection is Bannerman's (2008), that risk management under-performed in practice and that formal project management is neither necessary nor sufficient for success, which means the register may become a document that is filed rather than used, and Manathunga (2007) sharpens the point for the agreement by arguing that the vocabulary of collegial supervision can mask the power the relationship contains. What would settle the dispute is a study that measures the specificity of registers, whether each entry names a threat, an owner and a response, against outcomes, and a comparison in which supervision terms are stated in advance in one arm and negotiated as usual in the other.