Apr 19, 2012
Better late that never – I came across a SaaS forecasting technology provider named Lokad in one of my sojourns on the interwebs. I must confess at the very outset though that I’m a biased reviewer at best. Biased, how and why? Two salient points:
1. I am of the firm belief (note the word) that forecasting is best done in hindsight where it may do the least harm. Forecast all you wish about the past – I have no issue with you.I have no problems with economists creating all sorts of economic models to forecast the state of the economy because I rarely if ever pay any attention to them. So also the Raputre – same thing. This thread of skepticism pervades everything from the macro to the micro forecasting world. That’s the first point – Do no harm.
2. I am firm believer (note the word again) in the primacy of execution – the best forecast that I can think of is the one that creates the world ahead because of the design, of the plan, of the intent to dominate etc. As you may imagine, this is restricted to the world of growth and not the world of maturity and decline of a product’s lifecycle. While each of these three phases of a product’s lifecycle have their own specific variables and levers of interest, my mantra here is not to predict the world but to respond to it in the least amount of time. Thus, I’m not a big fan of long lead times, centralized planning et al.
Please do keep in mind these points as I delve into this review.
"We benchmarked Lokad on client data (a beverage distributor) against a model we specifically developed for the case. Following a deep analysis of the data we combined different forecasting techniques like ARIMA, VAR, LOESS, HOLT-WINTER and others using R, the statistical computing software. Lokad performed very good, the values of MAPE were similar to our results, after 3 months of analysis of the case. I am really impressed of this accuracy. Lokad is also very fast and provides a high level of automation." Mauro Coletto, Business Intelligence Consultant
Empiricism – always a good idea.
The good practices that I see at Lokad’s forecasting engine
1. Getting as close to the point of demand data as possible – vital for execution and even more so for forecasting. But from the looks of it, if you as the client of forecasting as a service don’t have really granular data, then Lokad’s service can only be as good as your own execution efforts are likely to be. The implication here is that you really can’t expect Lokad to improve your sorry ass case of datatitis.
Our technology is designed to deal with your data in their current form.
One point to note here though is that most companies firmly believe that they’ve got a good handle on data. That’s until they see how their industry/segment leaders benchmark at.
2. Getting the statistician(s) out of your firm. Spouting statistics on this and that is a finely honed skill that has considerable usefulness to your career progression – the higher you go, the more access you have to utterly useless statistical gordian knots. I don’t think that I would be so far off the mark as to say that higher honchos who browbeat you with statistics quantitatively know how poor their own competence and consequent results are without having any insights on what qualitative actions can be taken to surmount, nay transcend the current set of pitfalls marked on the pareto charts. While you can outsource the real statisticians and their professional output to Lokad – this may be an opportunity that Lokad is leaving on the table. Statisticians both inside and outside a firm are quite likely to be treated as black boxes anyway – so why incur the cost of having that black box on your payroll? I can predict quite easily that as Lokad grows, they will go after the opportunity on the table and I believe that can be a big big opportunity. The real competitors for Lokad in that space are the business/operations consultants. Upsell the interpretation once you have the lock on the statistical computation service.
Verdict: A big plus for Lokad.
3. The use of computational power – Call it the cloud, compute storm, global warming – whatever? What cloud computing does for every client is a very old idea in new clothes. That is to say, when you need the burst of computational power to solve a multi-variable (in the millions of them) for a short period of time, doesn’t it make better sense to pay for your slice of time rather than buying the whole set of computers and storing them in the basement? Companies like Lokad that are doing this today are all set to ride a power wave of such adoption – the time and assets to solve these large problems are now shared. Talk about cooperation without actually talking about it. Lokad is riding the wave and that’s a good way to piggyback on the successes of others riding the same wave.
Verditc: A plus for Lokad
4. Finally the Technology – The nitty gritty of the technology. First up, Quantile forecasting. I did struggle more than a bit to understand this and I’m not entirely sure that I understand it (Has that ever troubled me in blustering on but I digress…). The idea here seems to me to introduce a weight (that is correlated to the difference in payoffs for the positive vs negative realization of a particular event). To me, this is a part of modeling i.e. appreciating the sensitivity of a forecast and erring on the side of the lower cost. Makes sense.
Now, how can it be applied?
This Lokad blog post examines that (a little scroll down in the post) : Quantile Forecasting Technology
What I understand about the use of quantiles in forecasting is the improvement (marginal or more) that one gets from a better extrapolation of the expected demand predicted by a normal distribution over a cost weighted distribution? Is this the heart of it? Over 1 SKU the delta between a normal distribution and weighted distribution is probably quite small but extend that over 1000s of SKUs and the numbers begin to add up.
Verdict: A plus for Lokad (Remains a plus if it is what I understand it to be as above)
In all, I find such a move by Lokad quite an interesting thing. Validated empiricism would make it compelling as well. However, as I outlined above, my biases and experience is on execution. Good execution with good forecasting in your frontal view rather than in the rear view mirror is something of a mythical beast but strange things are beginning to happen in the world of the cloud.