Business & Technology Nexus

Back in the late 90’s, Ariba popularized the eProcurement space. With the typical hype of a surging start-up, they motivated the Fortune 500 to begin to address the inefficiencies so present in their indirect Procurement systems. And although nowadays it’s hard to imagine a system for automating the purchase of Office Supplies selling for $4MM or more, some did! Boy how times have changed.

Bringing mostly paper processes online meant the paper catalogs from vendors often available at admin’s desks in offices the world over would need to be digitized. But how?

Initially (and even today) intermediaries helped suppliers produce electronic content suitable to corporate buyers. The target format was defined by vendors who sometimes (and often surreptitiously) used standards bodies to legitimize their preferred XML ontology.

An interesting question the ecosystem of software vendors, corporate buyers, and suppliers faced was how much to classify and normalize content. By classification, I mean the deconstruction of lengthy and often marketing-oriented item descriptions into structured name-value pairs. For example this:

“Only .27 inches thin and 1.5 ounces, iPod nano packs a lot into its diminutive design. Up to 14 hours of battery life. 1GB of storage. A bright color display. The Apple Click Wheel. A Dock connector that fits an entire ecosystem of iPod accessories. With so many features like these, iPod nano can change the way you listen to music — and more.”

Becomes this:

The question, of course, is whether the classified content is more useful than the long description. Was it worth the extra hassle of deconstruction? And for most indirect spend categories, I argue no.

But some vendors go much further than category-specific attribution. They believe it’s not enough to deconstruct an item into its name value pairs. The attribute list itself must be rationalized against all other available items in that category. This is where the direct materials engineering design problem clashes with the far more basic needs of ordinary indirect purchasing. In my view a single system targeting both problems risks doing neither well.

So keep it simple. When implementing local catalogs in your eProcurement solution, don’t go overboard on the classification problem. Consider going with marketing-rich long descriptions and calling it a day.

Now of course completely avoiding the content management problem (via supplier punch-out perhaps) is a pretty good solution too. But that’s a story for another time.

Last post, we covered some of the inherent problems with managing corporate travel policies and controlling spend. But perhaps I started out seven steps ahead of where your travel problems lie. So let’s rewind and cover the most important “first steps” you can take when tackling the travel spend category.

This June ’05 Purchasing magazine story on the increased attention Purchasing departments are paying to travel spend is a great place to start. There are several points worthy of note: 1) For many companies, capturing travel spend data is a new thing 2) Initial savings comes from shifting from “legacy travel agency” to an online booking service and lowering transaction costs 3) It’s not a one-time “fix and forget” category, it requires regular attention.

As much as I’d like to argue for a more advanced treatment of travel spend, getting visibility into the problem and moving the booking process online have to be solved first. Once you’re there, you are ready for a more advanced regimen. Based on employee travel patterns you may be able to guarantee a minimum number of booking nights at hotels in some cities and can negotiate steeper discounts as a result. For large events, consider a reverse auction for block air and hotel reservations for an even better deal.

Once you know your spend and are aggregating it appropriately, consider the policy recommendations in my first post. And let me know how it goes.

After reading another one of Jason Busch’s post, this time relaying the trials and tribulations of flying to China, I was reminded of the enigmatic Travel Spend Category.

At Oracle, we used a service we had created and then spun-off called e-travel. It was pretty awful (perhaps Amadeus will fix it eventually). Not surprisingly, e-travel wasn’t connected in the slightest to the rest of Oracle’s procurement systems, even though the Purchasing department was in charge of managing travel spend. And us senior folks had an entirely different process to follow than all other employees. Yes, with Travel all the normal rules around procurement seem to be suspended. Source for best value, no. Operate from a budget, no. Spend first and ask questions later, y-e-s!

With Travel, expenses come in as a disturbance to cost centers. They are rarely anticipated. A manager’s only tool to control costs is to threaten to not approve. But having been there before, having stared at a $10,000 expense report that should have been closer to $6,000, it is incredibly tough to do & is rarely (if ever) done. Instead, the company eats the expenses and the employee gets a stern warning. And lack of clear policies and expectations are to blame.

The central dilemma I always found with travel as a spend category was how airline reservations worked. A fundamental concept within Procurement is pre-approval, and with travel it’s very difficult to do effectively. Airline prices change too quickly – by the time a trip is approved the airfare may have changed significantly. Corporate policy on fare class is equally a problem. Some companies choose to bet on non-refundable fares in the hopes that trips won’t change or be cancelled too often. Some stick to the always-refundable, always-expensive “Y” class (“Y” am I paying so much more than every else?). These are difficult questions where reasonable people come to very different conclusions.

Hotel and car rental don’t suffer from the same pace of pricing changes as airfare. But they get lumped in for the convenience of today’s booking systems. After all, they are a part of the trip!

So, now that I’ve described some of the challenges, I’ll make two recommendations. #1, please implement a “trip budget” pre-approval process. Make sure you set boundaries on anticipated spend. And although this seems like an LOB action, the Procurement department can help by drafting travel policies friendly to LOB cost centers. #2, consider incenting employees to save money by allowing them to participate in the upside. A reasonably-set Per Diem can change employee behavior from Room Service and the Mini Bar ($35/dinner) to something that is tastier to the palate and to the budget.

Travel is a big problem and I’ve got a few more thoughts. Let’s continue the discussion next post.

p.s. There is a nice Aberdeen report on procurement & Travel you might be able to find if you do a google search on “category spend by industry” :)

In my last post I introduced you to Linear Performance Pricing, a sophisticated method sourcing professionals use to analyze and negotiate pricing on direct material components. Using Linear Performance Pricing, you can pinpoint and correlate the relationship between price and attributes of value, then use that correlation to understand different vendors’ offerings. In this way, buyers can strategically contribute to new product design. And negotiating on price based on how the market measures value can prove very powerful with suppliers.

But we were using Linear Performance Pricing on more mundane matters. As I recall, we had a mini Cooper we were trying to get a great deal on. So, after playing around in Excel, I settled on the following formula to represent the inverse of value (let’s call it wear):

Wear = Mileage + (2005 – Model Year) * Constant

After getting all the cars.com data into Excel and scrubbed, I used the LINEST function to compute “m” and “b” (y = mx + b). I also paid a lot of attention to something called “r squared,” which gives us an indication of fit. An “r squared” of 1.0 would be a perfect fit, while values closer to zero indicate poor fit.

I experimented with the Constant until I found a maximum “r squared” of 0.55, which, admittedly, is fairly mediocre. My guess is further filtering of data points to eliminate body damage, poor mechanical condition, etc, would increase correlation. The Constant that “won” was 29,000. The meaning behind this is that as mini Cooper model years go by, older mini Coopers pay a penalty the equivalent of 29,000 miles per year of age. This seems pretty steep. The analysis covered 2002, 2003, 2004, and 2005 model years.

Here’s a chart showing the raw data and the best fit line.

So, you can probably guess what my smart friend did next. He looked at the data points as far below the best fit line as possible – those were the cars with the highest value and lowest cost. Next he had to factor in shipping costs. He was now down to a much smaller list. He filtered that list further by his color preferences and other subjective factors. With the handful that remained he negotiated over the phone, eventually settling on one at an even greater discount.

He had earned himself a sweet deal on an even sweeter ride!

Every once in a while you run across a story of someone applying their professional buying skills in everyday situations. Yesterday, I heard a great one! An incredibly smart acquaintance of mine was in desperate need of a used mini Cooper. But, naturally, he needed more than the car. He needed to know he was getting an exceptional deal as well.
Of course, at first glance Edmunds and Kelly Blue Book seemed like they offered more than adequate information on used car prices.

But what if you want to double check them? What if you want to run the numbers yourself? All it takes is a little time, an active mind, and Excel.

Linear performance pricing has been used to source direct materials components for years. McKinsey & Company described the method on page 5 of the June 2003 Automotive OE Supplier News. In short, it stipulates you can correlate price to “performance” once you discover the component of a product most relevant for “customer value.”

The simplest, most commonly used example for pricing for performance is engines. So, if you were buying an engine, let’s say a lawnmower engine, what would be the most relevant attribute capturing “customer value?” Many would say horsepower combined with efficiency or quality would be most important. Let’s ignore efficiency and quality and stick to horsepower. Analyzing lawnmower engine market prices might lead you to something like the chart below (note, I didn’t bother to get the pricing right, this is just boring lawnmower engines, not cool mini Coopers):

Once you have price vs. performance graphed out, it’s easy to see where the value is, even across different brands and completely different product segments. You just look for the data points below the “best fit” line. Honda, for example, sure looks like they build reasonably-priced 11 HP engines, whereas they are a little too proud of their 2.2 HP version.

So, what is the “performance” metric for a used miniCooper that best captures perceived customer value? My smart friend suggests it is nothing more than Year and Mileage. On the one hand this seems absurdly simple since no two cars are treated the same. Factors such as body damage, maintenance records, and number of owners could arguably be included. But let’s stick with it and see what we learn.

A quick search of cars.com yielded 878 results across the US (thanks cars.com!). I live in Half Moon Bay, and the mini Coopers available to me varied from 8 miles away to 2,789 miles away (beware, large sort operations on cars.com take a while, so have a hot cup of tea ready before clicking to sort).

Several hours of clicking and cutting and pasting into Excel and I had all the data needed to perform the analysis and look for uncommon value. Tomorrow I’ll share the results..

The best implementation practice I came across for continuous rollout and enhancement of enterprise software in the super-large enterprise utilized a train as its central theme.

There were two types: passenger trains (like the nice Southern California Amtrak Surfliner above) and cargo trains. Passenger trains carried new users, business groups, or countries onto an existing service. Cargo trains delivered new functionality or business processes to the existing user community. And IT simply never allowed the two to be mixed together.

The IT department would lay out a 3 to 6 month schedule for the “train.” Businesses could “hop on-board.” But beware those groups eager to join the party but unable to stay on time! The schedule came with checkpoints – and any business that didn’t meet the exit criteria of the checkpoint was left behind. You know, at “the station” :) They had to wait for the next cycle, and everyone knew it. So what happened? Business groups fell in line with the well-communicated, stated objectives of the train, met their commitments to IT, and the whole company was better off.

You see, most trains run on time.

If you’re in IT inside a large organization consider using the approach for your next project. And by all means, let me know how it goes!

Years ago, Freemarkets popularized a method called transformational bidding to capture important factors beyond price and directly incorporate them into reverse auctions. Using the approach, each supplier sees a different picture of the event as the unit prices of their competitors are discounted to take these other factors into account. The figure below contains a simplified example of “Xform bidding” as well as an alternative approach I like better.

Common factors beyond unit price are shipping costs, quality attributes, and incumbent status (aka risk). So why do I see transformational bidding as a moral hazard? Sometimes it’s not clear enough to suppliers what is happening during the event. On the plus side, suppliers see exactly what their unit price bid needs to be to win the event. But often buyers need not disclose the non-unit-price factors to the suppliers, leaving buyers open to temptation to play the variables to their advantage. And of course, most often buyers are not forced into honoring the outcome of the event.

I was heartened when a Procurement executive I was working with at a struggling toy company refused to use transformational bidding on ethical grounds. Instead, her team insisted on a different approach, one where a supplier’s bid was marked up by these transformational factors (instead of having other suppliers’ bids marked down). In this way, suppliers gain greater transparency into the fairness of the process, and buyers can still source based on best value. This approach makes it crystal clear a supplier is being handicapped by factors beyond their control – and if need be suppliers can challenge them, providing buyers more information to make the best decision possible.

So, does using Freemarkets-style transformational bidding generate bad karma? You be the judge.

I can’t tell you the number of companies I worked with whose supplier master contained 30,000 records or more. Talk about a mess! So, why is having a supplier master data problem common? It all starts at the source.

The first question I always asked customers was how a new supplier could be added to the system. The most common response was “we don’t know,” followed by “it’s complicated.”

And indeed, the problem is not as straightforward to fix as it seems. It all comes back to centralization vs. decentralization. There will always be segments of spend beyond the scope of a central buying organization. And decentralized spend categories are often where the most prolific growth in the supplier master occurs.

Some companies just resign themselves to periodic cleanup. And maybe that’s not an awful approach. By not introducing too much bureaucracy into the new supplier process the organization stays nimble. Many of my former customers used Austin-Tetra for help and seemed pretty satisfied.

Those that choose to tackle the problem head-on begin by inserting a central approval process for new suppliers. But expect LOB pushback, and offer same-day turnaround or some semi-approved status codes to keep from slowing down the pace of your business.

What’s worked for you?

In February of 2005, I wrote an opinion piece for Purchasing magazine defining a term we had created out of Marketing ether called “Center-Led Procurement.” It turns out the term had been around for quite awhile, as Mike points out in this post’s 1st comment. Nevertheless, the article was the only piece I ever wrote that generated unsolicited email from Purchasing executives saying that I had described exactly their problem. So what is Center-Led? Well, it’s the approach Procurement departments need to take to gain control over spend while dealing with a reluctant LOB. It attempts to balance centralized management and decentralized empowerment. I mean, after all, CPO or VP of Procurement is a political position, isn’t it? Check it out and tell me what you think!

In November of 2005, Tim Minahan, the amazing Aberdeen analyst who appears headed for Procuri, ran with the topic and produced a great report, available courtesy of SAP here.

Author’s sidebar: Much of the credit for my article goes to David Hope-Ross, a former Gartner analyst so good we stole him and placed him in Oracle Marketing purgatory :), and Jeffrey Pease, a marketing “message man” whose unique talent for bringing people together occasionally overcomes Oracle politics :).

What if you wake up and realize you desperately need a Procurement system. Where do you start? You probably just have invoice and payment data from multiple AP systems – and about all you know is the suppliers you paid, how much you paid them, and when. “I’m so screwed,” right?

Never fear, some vendors proclaim, with our magic decoder ring we can reconstruct your spend history. We can build super-slick reports to help you start off with confidence and pinpoint your savings opportunities. Check out Zycus AutoClass, probably the best example:

Wow, that’s quite a decoder ring. Or is it? You will definitely get a nice report. And it will look like your spend history. But don’t mistake it for accurate. Do it, but keep your expectations low and your wallet close at hand.

What’s the secret sauce? Companies like Zycus take feeds of “correctly” categorized content from large companies and use it to seed an AI solution based on Bayesian statistics – the same stuff used for Spam Filtering. Your AP data is then classified using their predictive algorithm. We experimented with this while I was at Oracle. I mean, after all, it’s an embedded technology available in Oracle’s database. But while Zycus claimed an astonishing 85% accuracy, we could never get over 60%, even with the best “training content”. I still don’t believe the 85% number. So watch out. And consider having someone spend a day or two doing some excel-based scrubbing to see if that gets you enough information to proceed. Your home-grown estimates may be all you need.

So, I hate Zycus, right? Wrong! I think they have pulled together some nifty technology. And as the body of “correctly” classified content grows, their algorithms get better. But I am a much bigger fan of getting data correctly classified on the way in rather than playing guessing games on the way out.

GE customized Oracle iProcurement, for example, to plug in Zycus’ AutoClass for any special request resulting in a Requisition. There, a human could decide whether the predictive algorithm was close to right (Yep, Smith vs. GE is indeed Legal Services). I’d trust the spend reports coming out of a system like that. And Zycus’ technology makes self-service classification possible (the alternative is having users guess which of 8,000 categories to place their request in).