The massive arrival of generative AI solutions in the enterprise holds many promises, including, at last, some form of simplification. That’s the upside. The flip side is increased costs and complexity. It’s not so much a problem of technology as of governance, but it’s a problem all the same.
AI: a new answer to an old problem
The transformation of work that we have been witnessing for almost half a century has led to a paradigm shift in production activities. We’ve gone from replicating perfection ad infinitum to solving new problems and making new decisions.
Solving problems most often means connecting people and information.
Information, because it serves as the basis for problem-solving and decision-making, or even as the starting point for a creative process.
People, because they are the ones who interpret the information, make the decisions and are capable of being creative. But that’s not all. The problems faced by employees are increasingly complex and often require the mobilization of several functions, and therefore the people who represent them, and rarely talk to each other. What’s more, organizational silos often give rise to informational silos, on top of the natural IT silos that create real nightmares for users. So, in the absence of an exhaustive view of the information needed, we bring together people who have a view of each of its components.
This is what businesses have been trying to do, under various names, since at least the 1970s, with more than mixed success.
To put it simply, it’s one of the promises of AI: to gather and process information and suggest ways of making decisions without having to mobilize anyone other than the person asking the question.
AI puts technology at the service of people rather than the other way round
It’s a promise made all the more interesting by the fact that it finally makes the human being the client of technology , whereas in the field of collaboration and problem-solving, the human being has often been called upon to contribute without directly benefiting. In fact, this is the whole point of collaboration in business: to mobilize employees to solve a problem that is often not their own.
In this approach, tools are merely facilitators: it’s the individuals who input information, reflect, propose and decide.
The promise of AI is a little like that of unified search. It doesn’t take much research to realize that the most common and basic use case for collaboration was information retrieval. In a work environment where tools and sources of information are constantly multiplying, it was becoming faster to search for people who had the information or knew where to find it, than to search for the information itself.
One of the most obvious solutions available at the time was the unified search engine. The principle is quite simple: a search engine that indexes your intranet, or even your intranets, your knowledge bases and, why not, certain business tools, your mailbox, and so on.
An example that’s easy for the general public to understand, and so obvious that we don’t even think about it, is Google. Google indexes an infinite number of sites, including content from Facebook, Linkedin, Youtube, e-commerce sites and so on.
A single gateway to all available information.
The comparison with AI ends there, as it adds the ability to process information and propose solutions. Where the search engine gives us a list of sources from which we can work out an answer, AI gives us an answer. Or at least tries to.
But one thing is certain: if the method has proved its worth, businesses have not followed suit. Why not? Each content management tool had its own search engine, which came at a considerable cost. How do you justify the cost of a meta-search engine, knowing that you also have to take into account the complexity of integration, connection to all the sources…and of course the associated costs?
The answer was simple: very few businesses did, and instead decided to continue living with their silos.
Is AI a victim of IS balkanization?
My fear today is that AI will meet the same fate as search engines. Of course it will be widely deployed, but in a siloed way, struggling to live up to its promise of providing global solutions.
But is this surprising? The basis of AI is first and foremost the indexing of the information it needs to learn, so it basically starts out with the same problems as search engines.
Your CRM will have an AI that claims to manage everything to do with customer relations. Your HRIS will have an AI dedicated to HR issues. Your ERP will have an AI to manage your activities, from purchasing to production… as long as they’re all managed in the same suite. You’ll have an AI that finds answers in your content management tools.
But an AI capable of taking all these components into account to help you see the big picture, identify correlations and help you make decisions?
That’s another task that needs to be tackled in terms of governance and financing. Who will do it? I’m afraid I know the answer.
This recent article starts to ask the right questions, but makes two mistakes: it assumes that everything can be indexed (ethics), and it assumes that systems of record don’t have to be, which I think is a big mistake, because one of the real added values of AI is to analyze past opportunities in terms of their context and impact, in order to make better ones in the future.
And AI poses a new question. It is trained on a corpus of data, but not everyone has access to all the data. So how can it know in retrospect, when asked a question,what part of the data it has ingested it is entitled to use to answer a given user?
This can even lead to bizarre situations. For example, you’re a manager and give a young recruit the task of working on a case. He uses AI and returns his work to you, which you are not satisfied with. You then decide to do it yourself and get a better result. Is it because your employee is bad at writing prompts? Or is it simply that you have access to information that they don’t? We know the answer.
Please note: even without AI, the problem is the same for a “manual” search. AI simply brings back into the spotlight the fact that technology, despite its potential, comes up against governance issues in business that are foreign to consumer AI.
History is an eternal restart
This is not pessimism, but a kind of jurisprudence. All the technologies that required a total desilotage of information in order to reach their full potential have generally only been deployed locally, on a business vertical.
I was talking about search engines, but closer to home, there are Employee Experience Platforms. A concept you’ve probably never heard of, for the good reason that it has never taken off.
This concept, introduced by Josh Bersin a few years ago, was a sort of new employee portal, a “one-stop shop” with all the data and services concerning them, which they can use to make decisions about themselves, their work and their career.
On the face of it, an excellent idea. But Microsoft has established itself as the Employee Experience Platform for office automation and intranets, Oracle for HR, ServiceNow for workflows… By definition, such a platform needs to be unique, so when you have as many IS publishers as you do…you don’t have one.
Moreover, if it keeps its promise and breaks out of silos, AI should be the business’s employee experience platform. That’s if it extracts itself.
That’s why, from the outset, the question of cohabitation between global AI and specific AI has always seemed essential to me, because it seemed obvious to me that the same questions would arise, and I see no reason why they should be answered differently.
Is it, however, crazy to think that the answer to many problems lies, for example, partly in HR, partly in operations and partly in finance? I don’t think so.
This is confirmed by Josh Bersin’s article on the arrival of AI in ServiceNow… because what better basis for cross-functional AI cases than this specialist in multi-tool workflows? And he expresses the same fears as I do: ServiceNow has a cross-functional vocation, but on the other side there are Oracle, Workday, SAP and others who are there, are expensive and boast the same value proposition, but on their specific perimeter.
Bottom line
The most far-sighted will opt for desilient, cross-functional AI, because they will see the value in it, which justifies the effort and investment required.
Elsewhere, that is to say in the vast majority of cases, we’ll see a balkanization of AI along the same lines as the fragmentation of IS… unless, for once, the past doesn’t repeat.
Photo : artificial intelligence by Gorodenkoff via Shutterstock
