The Saucerer’s Apprentice: AI Co-Pilots in the Enterprise Cockpit
There’s a story I’m fond of recounting about when I was part of a team that stood-up a computer factory in Ireland many years ago. It was one of many such initiatives that were sponsored by the Irish Development Authority (IDA) that granted ten year tax holidays to foreign-based employers so long as they met hiring commitments, a simple gambit that created an enduring economic miracle.
To the computer company this was a bold strategic initiative aimed at opening the European market to low cost mini-computer providers. And on paper it was nothing less than brilliant. The IDA would help fund the factory build out. The company would base the factory’s banking in Bermuda where interest on loans would be tax free. The factory would then loan money to its European sales and service subsidiaries who would then funnel all their orders and a large share of their receivables to the Irish Final Assembly and Test facility, aka the factory.
At the time, the popular buzz word with all the Harvard B-School types was accretive, any truly strategic initiative had to be accretive, had to accumulate benefits, had to sweat cashflow and profits and this set-up sweat more profits than a nitrogen dipped bottle of Stoli’s in a Swedish sauna. There were, of course, just a couple of caveats that you had to be aware of, the most important of which was that to repatriate those profits would cost you a bunch in US taxes but if the plan was to use this structure as a means to launch a pan-European business platform that concern rarely came up.
We started out in a nursery facility provided by the IDA, not the actual factory which was being built from scratch in a newly created industrial park just miles from Dublin’s airport. One day, when it wasn’t raining, we took a trip over to see where the new factory would be and discovered that it was nothing more than an empty field with a couple of tinker’s wagons, some piebald ponies and a dozen or so unsupervised children playing in the mud. Seemed pretty accretive to me.
It was around then that it dawned on everyone that the projected cost of goods sold wasn’t exactly meeting expectations. Upon further review it was discovered that the refrigerator sized cabinets, the metal frame and sides that housed the chassis, cpu, memory, I/O, bus boards, wiring harnesses and fans that made up a finally assembled computer, were not being fabricated locally. Instead, they continued to be fabricated somewhere outside Chicago and every couple of weeks or so a jet would get rented and a couple of truck loads of partially assembled cabinets would be flown to Dublin to meet the growing European demand. The B-School types were not happy. It was like some knuckle dragging manufacturing type had taken it upon themselves to FedEx a couple of tons of concrete to Ireland just for the fun of it. The Stoli’s was not nearly as sweaty as it should be.
So the word went out to find a local cabinet supplier pronto; no stone was left unturned. An Ireland based supplier was preferred but UK, German, French suppliers all received copies of the cabinet drawings, tolerances and specifications and after duly studying the request, they unanimously responded by saying that the cabinets could not be built as requested, at least not by them.
A New Beginning
These days, AI providers of co-pilot solutions would wave their hands and say, “problems like these can be easily avoided” if only you allow their platforms to scrutinize your company’s structured and unstructured data. Just go to the OpenAI app-store and download yourself a customizable Large Language Model (LLM). Or get in touch with your local AI co-pilot platform provider who for some inexplicable reason will take the next two maybe three years to build you a proprietary AI co-pilot according to your unique specification. The CEO of said co-pilot platform would lean over the table and with a face as solemn as a Sunday school teacher say, “Somewhere there are dots, lots and lots of dots, that only need to be connected by a customized LLM. The future is just around the corner. And it will be yours.” System integration shops use to call “solutions” like these roach motels; once a customer walks in, and the meter starts running, they never walk out.
Turns out strategies, even very simple ones, are really slippery stuff, which is why most companies avoid them. Start-ups don’t need them, they have founders. Private and family run operations rely on values and tag lines, they have owners. Mid-tier shops rely on the instincts of clueless, ambitious managers. And really, really big shops bring in the really, really expensive management consulting types that use really, really small fonts on their over packed and over produced power points, which doesn’t result in a strategy but rather an incumbent executive insurance policy that they slip in the coffee at some annual Board of Directors retreat like spoonfuls of Ketamine.
Companies that need strategies have certain notable characteristics. First, for whatever reason, they became publicly traded entities, which means they are a recognized going concern, a potential wealth creating proposition, and are no longer masters of their own destiny but subject to the motivation of their shareholders. Next, they have run out of momentum, whatever got them into the public market will no longer sustain the growth rates required by their shareholders and the low hanging adjacencies and economies of skill, scope and scale have been exhausted by clueless, ambitious managers. ( see Value Based Strategy Formulation – February 2011) They are entering a phase of the corporate life-cycle where the Ineluctable Entropy of Being starts tugging at their sleeve. These shops need a strategy or at least a couple of KPI’s they can point to.
This is where an AI co-pilot could come in mighty handy; so let’s build you one. Let’s start by giving it a name; a descriptive acronym that has a human-like affable demeanor. How about JASPER for Junior Associate Supervisor for Performance Enhancement Reporting? Something more than an intern but not as fashionably appointed as a consultant, more like an apprentice. Now to be clear, creating Jasper will be no small undertaking. Jasper is a thoroughly bespoke application, from its carefully curated vector data base to its pithy, slightly sarcastic responses; it’s entirely built by hand including spot welding data to abstract notions and sophisticated concepts and conventions. LLM’s in no small measure learn by decomposing and parsing input into subject-object-predictate relationships, analyzing and storing these relationships as sophisticated vectors and then assigning probabilities to their recombination by virtue of interrogation. It’s kind of a reversely engineered semantic web where parts of speech and relational data are rendered in a quasi Resource Descriptive Framework tuple, which is why the construction of the query is so critical to the fidelity of the response (see AI’s Inconvenient Truth – July 2018).
Source: Gartner
But here’s the rub, raw corporate data follows no subject-object-predicate protocols. There are no verbs in your ERP system. There is no causation in your CRM system. There are only algorithms that arbitrate data relationships and construct derivative concepts like metrics and Key Performance Indicators. But don’t despair because even source code can be analyzed to tease out how derivative data has been created. However, that’s not the only concern. Two separate departments, say marketing and sales, within the same organization often assign unique aliases to the exact same term, data or algorithm. It kinda makes them feel special. And then they spend enormous amounts of energy reconciling them,which means derivative concepts are often fragile and subject to breaking and constant mending (see Digital Strategies… A Practitioner’s Perspective – February 2016). Next, to truly deliver strategic value it will have to be able to look beyond self-generated, enterprise information. Jasper will have to understand temporal dimensions, it will have to assimilate abstract concepts like Dupont metrics, it will have to become familiar with supplier and competitor data, taxes and regulations, customs and constrains that live beyond the enterprise boundary. In other words, it will have to be federated to all manner of information sources to gain a working knowledge of the world at large.
Creating a valuable asset like Jasper is going to take patience, commitment and a sizable investment.You first must identify how it will be trained. What feed stocks of data will be shoveled down its gullet. Then you must immaculately clean, organize, normalize, marshal and otherwise prepare the data. Then you must categorize, label and encode the data. While you’re at it you might want to make sure that you haven’t inadvertently swept up any personal or proprietary data, like all those e-mails between you and your top competitor colluding on pricing. Then you have to parse and tokenize the data making sure that before you store it you vectorize it for indexing. Once you are finally ready to take it for a spin you might want to consider constructing a block chain analyzer to examine exactly where Jasper acquired its answer. Using Distributed Ledger Technology and Zero Knowledge Proofs you could peer inside the social life of Jasper and determine which sources and syllogisms Jasper employed to construct its answer. (see Beyond the Cyber-Cryptoverse – February 2019 and Identity and Sovereignty – September 2018)
So, if that’s not enough, to be able to produce any thing of value Jasper has to do the hardest thing imaginable in a firm desperate for a successful strategy. Survive. Multiple iterations of ownership, leadership and management. Numerous changes of applications and constant revisions to data definition, a rapidly changing collection of techniques and technology that have yet to mature, not to mention constantly shifting colloquialisms to accommodate culturally acceptable corporate speak. You may soon find that with Jasper you have constructed something akin to a hypersonic Osprey aircraft, for every one hour of flight time you need to invest six weeks of intensive manual maintenance, something that Jasper would never be able to explain to you should you actually bother to ask. But at this point you have invested a boat load in state-of-the-art technology. Your sunk cost is way bigger than your pay package for the rest of your miserable career which is bound to come to someone’s attention. Congrats!
So, let’s put Jasper in service with a practical test of its cognitive abilities. Cash conversion cycles are a constant concern for most operations. So, let’s start by asking Jasper if improving the work to wait ratios on class A inventory or exploiting the elasticity of using a 2/10 net 30 discount on receivables provides a greater yield on cash conversion cycles. Nothing? OK, how about about whether increasing marketing expenditures in the current quarter provides an appreciable impact in cash conversion cycles two quarters later (see The Elephant in the Room Just Got a Little Bigger – May 2017). Zip, zero, nadda? Maybe we have to get beyond enterprise data to catch a glimpse of Jasper’s potential brilliance. How about a hypothetical re-location of the supply chain to take advantage of emerging economic conditions. What if we re-located our Final Assembly and Test operation and revenue recognition Point of Origin, commonly know as FAT POO ( it’s not just and emoji) to a tariff friendly, low cost center with an indigenous market and depreciating currency? What? Another blank? What if we do it the old fashion way and look for an easily corruptible local government, disease free escort services and a private airport with a Gulfstream friendly runway?
Bingo!
Ko Pha-ngan, Thailand.
About your new intern…
The reason stories are often recounted is that they frequently don’t add up. Take for instance the story concerning the manufacture of the computer cabinet. Turns out that the reason the cabinet could not be manufactured by local EU suppliers was not because they lacked the competence, skills and materials required but because the engineering drawings were wrong. If they produced the parts per the specification they could not be assembled, the holes for the fasteners did not line up. So, how, you may wonder, could the supplier outside Chicago get it done? Simple. They recognized the defect in the drawings immediately and called the Engineering Department of the buyer who dispatched an engineer to the supplier’s site to red-line and sign-off the needed changes. A common practice for supply chains in the United States. The change however never made it through a formal engineering process and subsequently was never recorded in any engineering database.
So no matter how sophisticated your AI co-pilot might be, it still won’t be able to see what isn’t there.
And Jasper? Well, he wasn’t enamored of your trendy DEI policies and decided to pursue a career in the culinary arts instead.
Cover graphic courtesy of SuZQ Art and Images all other images, statistics, illustrations and citations, etc. derived and included under fair use/royalty free provisions.
