Should CIOs bet on Meta's new enterprise technology push?

Meta has entered the enterprise technology space.

On Sept. 28, the company announced its Meta Enterprise Platform, a new offering and business unit focused on offering AI agents and developer tools to companies. Whether CIOs should bet on Meta's enterprise AI push is a challenge to answer.

Meta, the company formerly known as Facebook, is still best known for its social media properties including Facebook, Instagram and WhatsApp. In recent years, it has made a concerted effort to be a generative AI player as well, with somewhat mixed success.

The company's Llama models were notable for being open source and letting others build on them. Meta shifted its strategy with the Muse platform, offering proprietary technology designed to compete with industry rivals such as OpenAI and Anthropic.

"The problem is that for customers to adopt AI, they have to be able to trust it, and Meta's track record on trust is not great," said Rebecca Wettemann, CEO and principal analyst at Valoir.

Meta's enterprise pivot in context

The initial release announcement for the Meta Enterprise Platform described the offering as the way Meta will offer its technology stack to businesses and developers. CJ Desai, the former MongoDB CEO, will lead the unit as chief enterprise platform officer and report directly to Mark Zuckerberg.

Meta Enterprise Platform started out as four key products:

• Muse. A personal AI agent that takes actions for a user, such as sending emails, booking travel and filling out forms in a browser. Meta launched it in the U.S. on Sept. 8 as a consumer product.
• Meta Business Agent. This covers the platform's customer-facing side and predates the September launch. "The Business Agent has been live since June and already runs inside WhatsApp, Instagram and Messenger, and Meta has announced an enterprise version that connects to systems like Shopify and Zendesk, though it hasn't published technical details on those connectors," said Rick Watts, an enterprise architecture and IT expert at Quandary Peak Research.
• Muse API. The API lets organizations programmatically access Meta's models inside their own applications.
• Muse Code. The Muse Code developer tool was first launched in beta on Aug. 5, offering capabilities to plan, debug and write code.

Meta's enterprise move puts it into an already crowded market with offerings from Microsoft, Amazon, Google, OpenAI, Anthropic and Salesforce, among others.

"Meta owns the chat window where the customer actually shows up. So, the honest shortlist question is not whether Meta replaces Copilot or Agentforce. It is whether Meta should run the agent that talks to your customers on Meta's own apps," said Rob Smith, founder and CEO of Lionfish Tech Advisors and a former Gartner analyst.

Where does Meta have the advantage?

While Meta is new to the enterprise AI space, it has an established reputation in the consumer space, which brings both advantages and disadvantages.

Meta's advantages

Customer reach. The advantage begins with the channels where businesses already talk to customers.

"The biggest edge that Meta has is obviously its customer base," said Tom Cloud, founder of Vertex CIO Advisory. "It owns major communication channels like WhatsApp, Messenger and Instagram, so it's far ahead in customer service, sales and support use cases."

That reach can be a real advantage over other enterprise providers that don't have the same consumer-channel reach.

"If a meaningful share of your customer interactions already happens on WhatsApp or Instagram, Meta owns a channel Microsoft, Salesforce, Google and ServiceNow don't. And an agent living natively in that channel is a legitimate differentiator," Watts said.

Industry fit. Some industries are better placed to use that reach.

"Retail, hospitality, consumer services and anyone with a large international customer base in markets where WhatsApp is the default messaging layer should take that seriously," Watts said.

Scale and price. Meta is a hyperscaler and brings its massive scale to the table, which could help it with pricing.

"What Meta does have is resources and computing scale, so it could aggressively compete on price," Wettemann said.

Meta's disadvantages

Enterprise credibility. Though Meta has extensive consumer reach, it lacks the enterprise credibility its rivals have.

"Meta starts from way behind. It doesn't have enterprise credibility nor does it have the enterprise sales motion that Google, AWS, Microsoft and others have spent years building," Wettemann said.

Enterprise tooling. That lack of enterprise systems puts Meta at a disadvantage.

"They have no core business systems. None. They shut down Workplace, which is their last business product. Microsoft owns documents, Salesforce owns customer records, ServiceNow owns support tickets," said Brad LaPorte, special advisor at Lionfish and a former Gartner research analyst. "They are built for consumers and not enterprise -- full stop."

Even with a solid model, enterprise tooling makes a difference for organizations and represents a gap that Meta will need to cross.

"Having run every one of those platforms, I can tell you the moat isn't the model," Watts said. "It's the 15 years of identity, entitlement and audit plumbing that the agent inherits for free on day one."

Trust. Though Meta has broad consumer reach, that history also carries baggage in the form of consumer privacy concerns.

"If I can't trust Mark Zuckerberg to keep my personal data safe or protect my kids from addictive software, how can I trust him with enterprise data and AI?" Wettemann said. 

How to evaluate Meta vs. incumbents

CIOs should evaluate multiple dimensions of Meta's enterprise platform.

Governance and safety controls. For a CIO, the question isn't whether Meta can build a capable AI agent.

"The real question is whether it can be governed, integrated and trusted well enough to earn its place in the corporate stack," Cloud said.

Data security and residency. How a vendor uses data is a critical concern for enterprises and a key evaluation criterion.

"Meta has said security and privacy are built in from the outset; that's a marketing sentence, not a data processing agreement," Watts said. "Get the DPA, the sub-processor list, the data residency commitments and the model training exclusion in writing before anyone spends engineering time."

It's also critical to understand how Meta will separate corporate data inside its platform.

"It has no tenant model that CIOs know, no established posture on how enterprise data is segregated from its advertising business, no published pricing and no enterprise systems-integrator channel comparable to what the incumbents sell through," Watts said.

Integration and orchestration. Enterprises typically have multiple systems from different vendors. CIOs must consider how to integrate and orchestrate technology within existing systems.

Observability and audit. Visibility into agent activity is another core evaluation criterion for enterprise users.

"Ask who can see and stop the agent, and where the logs live," Smith said.

Pricing and total cost of ownership. As AI usage grows, so too has the need for enterprises to control spending. When evaluating Meta's platform, organizations must understand its costs and how to manage their budgets.

"You cannot build a business case on a number nobody will tell you, and agents are where AI bills get ugly," Smith said.

Lock-in and exit. Exit planning belongs in the contract.

"Buyers will want exit terms and data portability written down before they commit a workflow," Smith said.

Support model and roadmap. Meta's roadmap has shifted this year, which might be a concern. Enterprises need a certain degree of stability before placing technology bets.
"When we adopt a tool, we're betting the vendor will keep executing well for years, and Meta still has to prove it can do that in B2B," Alex McLeod, co-founder and CTO at Serval, said.

Integration reality check

Enterprises that decide to pilot Meta's enterprise platform must evaluate specific integration elements as part of the pilot deployment.

Integration requirements. The initial set of connectors for the platform is somewhat limited, with Meta only naming Shopify and Zendesk. For a pilot evaluation, organizations should test against their specific connector integration requirements.

Integration with enterprise authentication systems already in use also needs to be part of a pilot. 

"Ask Meta to demonstrate, not describe, how a Muse or Business Agent authenticates against your identity provider, honors your existing role-based access and writes back to your system of record with an audit trail," Watts said.

Governance and risk. A pilot potentially puts Meta's agent in front of real data, so the rules for what it may do and how to stop it must be set before it starts. Policy guardrails define what the agent may decide without a person. The pilot should test them on the company's own products.

"A real pilot measures the wrong-answer rate on your own products and puts a human approval on anything touching price, refunds or policy," Smith said.

Telemetry and incident response let the team see what the agent did during the pilot and act quickly if something goes wrong.

"Have logs for everything. Have an off switch and a call escalation tree with pre-approved timeframes backed by a service level agreement if something goes wrong," LaPorte said.

Cost and ROI. The goal of a pilot is to determine whether a technology will work, how much benefit it can provide and at what cost.

LaPorte recommended that organizations set a hard monthly spending cap that shuts off automatically when the limit is reached. Going a level deeper, the organization should have granular controls for specific items, such as limits per conversation, hour, day and week. He also suggested that organizations track the percentage of conversations resolved without a person, along with additional context on how they were resolved and the true success rate.

"Set a crawl, walk, run pilot that has a go/no-go date set in advance with specific quantifiable success criteria," LaPorte said.

Competitive comparisons. There is no shortage of competitive options in the enterprise AI space. When piloting Meta, enterprises should measure other options, including any incumbents, on the same workload.

A defined baseline from the incumbent system the organization already runs, with the same workload, creates a like-for-like comparison of resolution rate, escalation rate, latency and cost per interaction, Watts said.

Comparing against the established industry leaders is a core part of any pilot.

"We're continuously evaluating coding tools and models from OpenAI, Anthropic and others," McLeod said. "These leading labs have the strongest models, so to earn a place, Meta would have to clearly outperform what our engineers already use, and we haven't seen evidence of that yet."

When evaluating a pilot, there needs to be clear evidence that the technology is actually helpful.

"A pilot pays off only if there is strong collective agreement that the new tool clearly improves day-to-day productivity for our engineering team," McLeod said.

Sean Michael Kerner is an IT consultant, technology enthusiast and tinkerer. He has pulled Token Ring, configured NetWare and been known to compile his own Linux kernel. He consults with industry and media organizations on technology issues.