As we pass the first anniversary of the start of the most recent tech hype tsunami (farewell NFTs, we hardly knew ye), it’s starting to become a little clearer how many enterprises might start their journey with Generative AI – and it’s in areas that are the traditional domain of the CIO, rather than the CDO. So, does that mean that the CDO role will be sidelined in the rush to GenAI? Or should the CDO suit up and go head-to-head with the CIO to lay claim to this growing area? Can we expect a CDO-CIO Rumble in the Jungle?
The core benefits of Generative AI for many enterprises lie in unlocking the potential of the vast amounts of unstructured data that is the lifeblood (and bane) of any large organisation. GenAI has caught their attention because it’s good at the following things:
- Summarising large amounts of unstructured content
- Creating new content quickly based on prompts and seed content
- Automating regular content generation-related tasks
If you think about it, a huge portion of day-to-day life in an enterprise org is reacting to and synthesising stuff that other people have written (emails, chat messages, PowerPoint presentations) and integrating that with your own ideas to create your own content that then goes on to other people, who do the same thing. Making this process more efficient is thus a huge opportunity, which Microsoft and Google in particular have spotted and are rushing hell-for-leather to monetise.
Building on its integration of GPT-4 models into Bing and Github, Microsoft has now launched Copilot for Microsoft 365, which integrates generative AI functionality into Teams, Outlook, Power BI and the Office apps. Meanwhile, Google has launched Duet AI for customers of Google Workspace, which offers a similar promise: the ability to automate drafting of emails, creation of custom images for presentations, and summarising notes from meetings. These capabilities will doubtless get a solid shot in the arm from Google’s newly announced Gemini model, bringing them roughly into line with GPT-4.
Both of these sets of offerings are essentially doing the same thing – they are providing a conversational interface into unstructured enterprise data (and a user’s own data). Enterprises have been struggling to leverage their unstructured business data for years – the fields of Enterprise Search and, if you go farther back, Knowledge Management, have sprung up in an attempt to solve this challenge. One of the core problems has always been returning relevant content from an enterprise corpus and helping the user to synthesise this content.
Generative AI models are very good at both of these things. Using techniques like embeddings, it’s relatively easy to take a prompt and use it to look up related content in an enterprise data store using semantic search, and then feed this content into an LLM to summarise and synthesise into a response. A very simple diagram of this kind of integration is below.
If you’ve worked anywhere with a large collection of internal SharePoint sites, you’ll be familiar with the pain and suffering of just trying to find stuff. This is the problem Publicis was trying to solve with its much-maligned (and rather ill thought-out) Marcel effort back in 2017. Putting together pitch responses for a consulting or services business is an incredibly time-consuming process, requiring a ton of trawling through previous proposals and case studies to extract relevant information to include in the pitch. Assiduous up-front categorisation of materials can help, but honestly, who has the time for that? The announcement for the Publicis/Microsoft collaboration on Marcel back in 2018 now sounds quite prescient, hitting many of the same beats as recent announcements about GenAI services.
Old Macdonald had a farm…
In many organisations, it’s been the CIO’s responsibility to handle unstructured data. CIOs spend a lot of their time on internal technology platforms, security and process digitisation. Almost by accident, their efforts have led to the creation of substantial estates of unstructured data, much of it in the form of emails and documents. An enterprising CIO who is looking for the next big thing as the “Covid bump” for internal technology investment fades is going to be pretty motivated to push into the Generative AI space as a way to drive more value from this asset.
Add to this the fact that most GenAI implementation are as much about technology as they are about data – especially the kinds of data concerns (pipelines, master data management, quality monitoring, analytics) that CDOs spend most of their time thinking about. GenAI presents important challenges around application integration, privacy and data security – often the responsibility of the CIO.
So while there is not a CDO in the land who doesn’t have an opinion about Generative AI, it might actually not be CDOs who end up delivering the most value from this technology in the enterprise.
A twitchy CDO might consider this development to be something of an existential threat – after all, 2023 has not been a good year for CDOs, with many organisations becoming a little sceptical that the role is really delivering value. And it’s very hard to sit by while Generative AI seems to be taking over the world and calmly say, “it’s OK, my buddy the CIO has that under control”. So if you’re a CDO, should you be making a grab for this brief? Like any battle, it pays to consider your chances of success and the best approach before weighing in. Here are some things to consider.
- Do you already have unstructured data as part of your brief? Depending on the kind of organisation you work for, your CDO role may already entail working with unstructured data (such as customer service data, or insurance claims content). If this is the case, it makes more sense for you to also own GenAI, and perhaps become the CDAIO (Chief Data & AI Officer).
- Is your organisation mostly about structured or unstructured data? Different kinds of company have very different balances in the relative importance of structured and unstructured data. For example, an online retailer has a lot of structured data that drives business value: customer orders, product data, shipping information, etc. By comparison, a law firm mostly trades in unstructured data: contracts, briefs, judgements and so on. If you’re a CDO in an unstructured data-heavy company, you should probably make a play for GenAI responsibility, as it will be central to that organisation’s definition of the CDO role.
- Do you have the ability to engineer GenAI solutions yourself? Many CDOs rely on teams run by the CIO for engineering, especially anything to do with integration into corporate IT systems. It will mean little if the CDO says “I’ve got Generative AI” if every piece of implementation work actually has to run through and be signed off by the CIO’s office.
- Are there other, more interesting, GenAI use cases you can champion? It’s becoming increasingly clear that Generative AI is a platform, not an application – there’s plenty of room for the CDO and the CIO to add value in different areas. The enterprise search use case is an important one, for sure, but there are other generative AI use cases besides this which might be a better fit for you to champion as CDO, especially if they are more synergistic with areas you are already own and require the integration of structured and unstructured data.
Can’t we just all be friends?
Of course, deploying the metaphor of a boxing ring punch-up to bring this topic to life is merely a rhetorical device — such a dramatic scenario would never happen in real life (well, probably never).
My prediction is that in many organisations we will see the CIO take responsibility for providing a set of foundational GenAI services (much as the CIO now provides cloud, security and workflow/productivity services) that other teams (including the CDO’s) leverage to deliver specific use cases. Savvy CDOs will be really solid on the platform considerations around GenAI (much as, today, any self-respecting CDO needs to be able talk about their cloud platform) but will look to move themselves up the value chain and enable high-value, specialised use cases that leverage their (and their team’s) specific expertise in data. But I think we can expect to see some entertaining scraps over territory along the way.