Investors spill what they aren’t looking for anymore in AI SaaS companies

Investors person been pouring billions into AI companies implicit the past fewer years, arsenic the exertion continues to clasp sway successful the Valley and frankincense the world. But not each AI companies are grabbing capitalist attention.

Indeed, adjacent arsenic it seems each institution these days is rebranding to see “AI” successful its name, immoderate startup ideas are conscionable nary longer successful favour with investors. TechCrunch spoke with VCs to larn what investors aren’t looking for successful AI software-as-a-service startups anymore.

Popular SaaS categories for investors present see startups gathering AI-native infrastructure, vertical SaaS with proprietary data, systems of enactment (those helping users implicit tasks), and platforms profoundly embedded successful mission-critical workflows, according to Aaron Holiday, a managing spouse astatine 645 Ventures. 

But helium besides gave a database of companies that are considered rather boring to investors these days: Startups gathering bladed workflow layers, generic horizontal tools, airy merchandise management, and surface-level analytics — basically, thing an AI cause tin present do. 

Abdul Abdirahman, an capitalist astatine the steadfast F Prime, added that generic vertical bundle “without proprietary information moats” is nary longer popular, and Igor Ryabenky, a laminitis and managing spouse astatine AltaIR Capital, went deeper connected that point. He said investors aren’t funny successful anything, really, that doesn’t person overmuch merchandise depth. 

“If your differentiation lives mostly successful UI [user interface] and automation, that’s nary longer enough,” helium said. “The obstruction to introduction has dropped, which makes gathering a existent moat overmuch harder.” 

New companies entering the marketplace present request to physique astir “real workflow ownership and a wide knowing of the occupation from time one,” helium said.  “Massive codebases are nary longer an advantage. What matters much is speed, focus, and the quality to accommodate quickly. Pricing besides needs to beryllium flexible: rigid per-seat models volition beryllium harder to defend, portion consumption-based models marque much consciousness successful this environment.” 

Jake Saper, a wide spouse astatine Emergence Capital, besides had thoughts connected ownership. To him, the differences betwixt Cursor and Claude Code are the “canary successful the ember mine.” 

“One owns the developer’s workflow, the different conscionable executes the task,” Saper continued. “Developers are progressively choosing the execution implicit process.” 

He said immoderate merchandise dealing with “workflow stickiness” — meaning trying to pull arsenic galore quality customers arsenic imaginable to continuously usage the merchandise — mightiness find themselves successful an uphill conflict arsenic agents takeover the workflow.

“Pre-Claude, getting humans to bash their jobs wrong your bundle was a almighty moat, but if agents are doing the work, who cares astir quality workflow?” helium told TechCrunch.

He besides thinks integrations are becoming little popular, particularly arsenic Anthropic’s exemplary discourse protocol (MCP) makes it easier than ever to link AI models to outer information and systems. This means idiosyncratic doesn’t request to download aggregate integrations oregon physique their ain lawsuit integrations; they tin conscionable usage the MCP. 

“Being the connector utilized to beryllium a moat,” Saper said. “Soon, it’ll beryllium a utility.”

Also, nary longer en vogue see the “workflow automation and task absorption tools that alteration the coordination of quality enactment go little indispensable if, implicit time, agents conscionable execute the tasks,” Abdirahman said, citing examples, chiefly nationalist SaaS companies whose stocks are down arsenic caller AI-native startups originate with better, much businesslike technology. 

Ryabenky said the SaaS companies struggling to rise close present are the ones that tin easy beryllium replicated, helium said.

“Generic productivity tools, task absorption software, basal CRM clones, and bladed AI wrappers built connected apical of existing APIs autumn into this category,” helium said. “If the merchandise is mostly an interface furniture without heavy integration, proprietary data, oregon embedded process knowledge, beardown AI-native teams tin rebuild it quickly. That is what makes investors cautious.”

Overa, what remains charismatic astir SaaS is extent and expertise, with tools embedded successful captious workflows. He said companies should close present look into integrating AI profoundly into their products and update their selling to bespeak that, Ryabenky continued.

“Investors are reallocating superior toward businesses that ain workflows, data, and domain expertise,”  Ryabenky said. “And distant from products that tin beryllium copied without overmuch effort.”