Joël Ollivier : Manager IT, Créateur de Progiciels ERP / PGI / GPAO pour l'industrie

Estimates

Rather than trying to imagine what will be the technical architecture of the Project, which is very long, set a Data Model (as defined by Merise), even incomplete.

Do not detail the properties of objects, we do not know, but just a rough model Entities / Associations.

Count 19 by entity, 25 by association, then sum all, multiply the result by 2: the result P is the weight of the Project.

Apply the following formulas that directly give working days:

Design Analysis AF = 0.1xPxKf, where Kf is a coefficient of design complexity, between 1 and 3 (generally, direct contact with users or not, needs knowledgement or not, etc...)
Development R = Px(1+Kr), where Kr is a coefficient of technical complexity, generally between -0.5 (advanced langages) to +0.2
Integration I = R² / (4000xN), where N is the number of sub projects integrated separately. We see immediatly that a project of more 2000 working days must not be integrated in one shot, lest the work of integration explode.
Total = AF + R + I, to dispatch after regarding the Project Management Method in use.
This very simple estimate gives good results even for a project whose analysis is not very advanced.

Tools

Why use other tools as standard type MSProject, PSN, PMW, etc..?

I use my own tools of project management

These tools will interpret the data entries. For example, changing work in MSP, the dates are recalculated. Changing a date, the work is changed too. It requires a Repository Project, containing budgeted raw data (Client), Allocated (Internal), and detail of the progress (weekly, granularity 0.5 day).

I am using MSP (or other) as a Gantt chart restitution, without any interpretation from it, except to establish a baseline (tool).

Planning

I draw, for current tasks, up to 5 bars in the Gantt charts.

No, this is not too complex, and this is a real indication for the Project Manager.

We name :

B the budgeted or planned work,
P the forecast or allocated work,
A the actuals,
R the estimate to complete.

The "Projected" is obvious for any timeout!

Gantt Bar	Description	Start	End
Budget, or planned	These are the dates agreed between the Client and us. The external engagement.	D1	F1
Forecast, or allocated	These are the dates discussed and validated between the Project Manager (or the analyst) and Project Manager. The internal commitment.	D2	F2
Actuals	True dates of working days on the project	Da : Date of 1st day worked on the project	Df : Date of last day worked on the project
Estimated	This is the end date if, from now on, I do my remains to be done as expected	Da	Now + ((F2-D2)xR/P)
Projected	This is the end date if, from now on, I do my remains to be done as I have advanced so far! Surprise!	Da	Now + ((Df-Da)xR/A)

Progress

We name :

B the budgeted or planned work,
P the forecast or allocated work,
A the actuals,
R the estimate to complete.

We generally calculate the percentage of completion of a task or project as:
Progress (%) = A/(A+R).
If this is not a mistake, it is certainly very poor, and it ignores the problems. Example:
10-day task, allocated to Albert. Albert does nothing for 10 days, but counts its time. Progress = 50% !! (A=10, R=10)
I control it by a second calculation like this:
Progress (%) = (P-R)/P (give 0% in previous example)
That is more pessimistic, but realistic.

Comparing the two results is a good indicator for the Project Manager.

Monitoring

The Progress Speed! A Speed is any quantity per unit time.

So I compute speed as the work realized divided by time spent (by task / resource / week).

An = time spent Week n, Rn = estimate to complete Week n.
Vn = (Rn-1 - Rn)/An. If all is ok, Vn = 1.
Smoothed speed: The speed varies greatly from week to week. On the other hand, it is often less than 1 at the beginning and end of project. To be better interpreted, this indicator should therefore be smoothed. I developed an algorithm of exponential smoothing in 1982, which I use very often. Here (E = Standard Deviation, T = trend, Vl = velocity smoothed):

En² = (0.8xEn-1²)+(0.2x(Vn - Vln-1)²)
simple exponential smoothing (coeff 0.2) of the variance
a = (1/Sqr(2xPi))xExp(-(1/2)x(Vn - Vln-1)²/En²)
smoothing coefficient calculation (between 0 and 0.4), Gauss
Vln = (1 - a)xVln-1 + axVn
exponential smoothing (coeff a) of the speed
Tn = (0.8xTn-1)+(0.2x(Vln - Vln-1))
simple exponential smoothing (coeff 0.2) of the trend

This gives a tendency to speed, thus ACCELERATION of the project!
This acceleration must be positive on the first 70% of the project (we only look at the sign).
We also obtain the standard deviation (the dispersion), which should not be too high.