Here are strategies that may help you dealing with this serious data science condition:
- Realize that chasing perfection is chasing shadows. And simply stop doing it! Accept that perfect model is unattainable and even if you get it – it is most likely to be perfectly useless as it apply only for very narrow segment of your population.
- Ensure that you have success criteria that needs to be reached in assigned period of time. If you have not reached it – find out what were the obstacles? Was it complexity of the problem, limitation of data, technology or skills and ask yourself what value you will get if you add more time. If benefits are miniscule make – don’t go further!
- Start out with fast methods first. Get initial rough-cut model first and see how far you are from meeting your goals and success criteria. Try to cover 80% of the distance with 20% of the effort and leave rest of the time for slow and laborious methods as well as for refining and tying up loose ends. At least you got something to show. There is saying where I come from: “Better to have sparrow in hand then a pigeon in rooftop”.
- Use methodology and milestones and be disciplined about it. Link specific step with time-to-close and try not to go over. Essentially, you are in race against time and so ensure that you have clear picture of what is part of a route that you need to cover in given time. That will ensure that you are moving in right direction within given time constraints.
- Be aware of obstacles, limitations, scope-creep, risks and dependencies. Each of these can cause you to bleed time and effort, while not progressing forward in any significant way. Almost, as if your wheel is stuck in mud. To others – this may mimic as analysis-paralysis, however, there is subtle difference: this is not something of your own doing, so may end up spending even more valuable time in defending yourself. Best way to deal with these is in pre-emption and in early warnings – so that are ready for “I told you so” – if it happens.