Salesforce’s vector for AI success is in promoting assistants to the role of agents

last updated:

Salesforce’s vector for AI success is in promoting assistants to the role of agents

last updated:

As Salesforce’s World Tour arrives in London, it begins to reveal how its AI strategy is increasingly wedded to Copilots as agents as well as assistants. But has it got a handle on the processes and tasks it needs to put them to work?

Mid-year in east London’s former Royal Docks has become an annual mooring location for Salesforce’s fleet of clouds. There, the company can host thousands of loyal customers, keen prospects, and associated onlookers to hear the latest announcements—some localized for the UK—from the company, hear from customers, and absorb product demonstrations.

The ongoing pre-announcements, followed by pilot momentum and eventual “general availability” of product means that if you dip into every stop on the World Tour, you’ll start to hear the same hits a few times (and let’s face it, the hits become the hits primarily through repetition). For this London leg, the headline announcement was the availability of the Data Cloud Vector Database that was first announced at the NYC leg in late 2023, which, given its focus on unstructured data, we were naturally all over at the time. The localized announcements, saw Data Cloud added to the UK Hyperforce region – a necessary compliance milestone for many organizations – and news of a dedicated AI Centre in London, at the Blue Fin building on the south bank of the Thames.

Protect the core

It’s worth just taking a small step back to remember why Salesforce or any of its peers have gone in so heavily in on AI – especially in its generative form – over the last couple of years. Is it because there’s a genuine belief in the revolutionary power of AI? Partly, for sure. But that’s not going to be enough on its own to release the internal funding to build out, in Salesforce’s case, Data Cloud and Einstein 1 et al. (as they are currently called) with all the supporting underpinning they require, which is a lot. What drives it is protecting the core cloud business for CRM (Sales Cloud), Service (Service Cloud) along with younger siblings in commerce and marketing. Fundamentally, to run a subscription business, you have to drip additional value into each renewal, hoping to incrementally increase each bill a little bit (and hopefully, a lot a bit). None of this matters unless it increases at least the perceptive value of those core applications to customers and prospects, which helps embed it more wholly into those organizations’ ways of working.

So, what’s the value of AI and Data to that core?

At present, Salesforce talks about five areas for action in which organizations should consider AI, and they are instructive regarding the current thinking of the company. The first is perhaps the biggest hitter: using AI to unify data and applications. It’s this, illustrated by Data Cloud and, as referenced previously, that now generally available Vector Database. It does mean that again, we have to eye-roll at the appearance of our friend: “90% of customer data is unstructured” – which this time is referenced to this paper sponsored by Box – the power of which will be somehow unlocked to produce downstream benefits to those core applications. As Alan discussed after visiting this subject via Salesforce’s TrailblazerDX event earlier this year, many bumpy bits on this path need smoothing to make that journey comfortable. As I also mentioned around that same time, the more that everything rests on RAG, the more questions arise, and the qualities that are being applied to this database far exceed the role it plays within the architecture. At this London event, the line “makes every cloud better” was used to describe Data Cloud’s potential here; for now, this part of the broad AI pitch feels the most developed, and as a result, Salesforce feels most confident in stating it.

Beyond that, “Discover AI Insights” calls for using the specific Copilots within Tableau. “Collaborate with AI” also points toward using Copilots across those core cloud applications (even if, for now, the Einstein Copilot application is only generally available for the Sales Cloud CRM, albeit with the others due to follow across the rest of 2024).

Where’s the task and the process in all this?

If protecting the core is really about adding value to further embedding new functionality into an organization’s way of working, then at this point, those five areas for action are lacking in focus once you’ve passed beyond the corralling of data. You might have noticed, we’re all about understanding work and how technology enables it – indeed, we have a whole set of publications predicated on understanding it better – an introduction to which can be found for free here – and as we’ve pointed out elsewhere, for us understanding the impact of AI on tasks and processes is critical in being able to place a value upon its introduction. It’s here that Salesforce right now isn’t yet outwardly guiding specifically upon, yet from the discussions that we’re party to with both customers and integrators alike, it’s trapping those specific points of value proving to be the most significant inhibitor to taking AI experiments into mainstream production. Salesforce itself privately knows that this specificity will help it create customer cases that resonate, and it is quietly sponsoring projects that will demonstrate this at work. In short, even the keenest proponents of the technology will need help building their business cases beyond broad-brushed hints like call deflection, basket conversion, and account activation. 

Instead, let’s give agency to AI workers.

It’s in the final pair of points of action that suggests where Salesforce’s current medium-term thinking metabolizes; “Give AI New Skills” and “Equip AI to Act.” Both are on the precipice of work automation, focusing on technological enablement, a place where Salesforce, for now, seems most comfortable. Suppose 2023 was the year of Copilots as assistants; 2024 has become the year of Copilots as agents and with many of its peers. In that case, Salesforce wants us to know that it, too, sees this as a natural, healthy extension of the art of AI, suggesting that through its range of Copilot building apps, developing through flexing prompts rather than code, customers will be able to build proto-agents to undertake sets of linked actions. The building blocks for Einstein Copilots include what it calls a “Reasoning Engine,” a method of determining intent from which a plan of execution can be dynamically assembled. When illustrated as the “Einstein Planner Service,” it’s done so with points of human clarification, although none of these are programmatically mandatory. 

For us, the gulf between the technically possible and the ability to deploy—given the lack of detailed thought toward likely addressable processes and tasks—feels wide and deeply echoing at this stage. It will be seen when customers bring those capabilities to life in actual, measurable tasks contributing to working processes that the value of those core applications will genuinely resonate.

Leave a Comment

We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.

Work Intelligence Market Analysis 2024-2029