Eilon Reshef, cofounder and CPO of Gong, is a seasoned entrepreneur, executive and investor in the internet and software spaces.
Enterprise teams are complex. Years of growth, adaptation and specialization create intricate workflows, terminology, challenges and goals. Every enterprise has its own “52 flavors of vanilla”—nuanced processes and requirements that don’t fit neatly into any off-the-shelf solution.
So, how could off-the-shelf AI possibly live up to expectations when every organization operates differently?
Conversational Interfaces
The first wave of enterprise AI suggested conversational interfaces are the answer: chatbots and assistants that employees could query directly. While this approach is appealing, it creates fundamental inconsistencies, as I’ve discussed in a previous article.
For one, as each employee interacts differently with AI systems through unique prompts and queries, you lose the standardization that enterprise processes require. Sales representatives develop individual approaches to qualification, support teams craft inconsistent responses and knowledge becomes fractured.
Additionally, the burden of crafting effective prompts falls on employees, forcing them to become prompt engineers. The result? Inconsistent outcomes, varying quality and an inability to enforce company best practices at scale.
At the individual level, conversational interfaces provide value. And executives or “freestyle workers” whose daily tasks don’t entail scalable systems or workflows can certainly see a productivity boost by using them.
However, conversational AI may not be the most scalable solution for implementing enterprise workflows. A global study conducted by KPMG in collaboration with the University of Melbourne revealed that 66% of employees use AI without evaluating it for accuracy, and 44% have used AI in ways that go against their company’s rules or guidelines.
Programmatic Agent Toolkits
For certain businesses, with capital and highly specialized functions, custom coding AI solutions can make sense. For the vast majority of companies, though, custom builds are likely not feasible or efficient, especially as McKinsey notes, 46% of leaders identify skill gaps in their workforces as a significant barrier to AI adoption.
Given the limitations of generic AI, many vendors are marketing what is custom development as their solution to tailoring applications. These toolkits, though, require technical teams to build workflows from scratch, chain together components and maintain custom code per process.
This can leave teams waiting in the IT queue and developers without domain expertise building business processes and high maintenance costs.
Business-Led Tailoring Of AI (Agents)
There’s an alternative approach to implementing AI in the enterprise that may better align with the needs of most businesses. Here’s what this approach looks like:
1. Business expertise belongs in business tools, not code.
Company workflows involve multiple systems, departmental handoffs and established protocols. When business leaders directly tailor how AI operates in their domain, implementation bottlenecks can be mitigated and processes reflect real-world needs, not a developer’s interpretation of them.
For example, when a revenue leader wants to set up knowledge transfer between pre-sales and post-sales, they may have an idea of what a handoff document looks like and what it should cover. They want to tell the AI what this structure is and iterate as needed.
2. Visual tools scale. Custom code doesn’t.
Visual tailoring environments can help with organizational agility. When a sales operations manager can adjust qualification criteria without filing an IT ticket, or support teams can refine response workflows regardless of developer availability, AI becomes responsive to business needs at the speed of the business, not the development queue.
Contrast this with prompt-based approaches. Not only do organizations need technical resources to write the prompts, they need a set of tools to ensure the quality, consistency and relevance of those prompts across the organization.
3. Re-inventing the wheel isn’t effective, but enhancing wheels can be.
Most businesses start with some established best practices and adapt them over time rather than proprietary processes.
For example, in revenue organizations, a well-known framework for deal qualification is MEDDIC—but companies have tweaked it to the point where there are now named variants such as MEDDICC, MEDDPICC and others. Similarly, there are standardized support triage processes and procurement approval workflows that get optimized over time.
The approach leverages these best practices as foundations rather than forcing teams to recode them as prompts or custom-build agents. Instead, software systems come with those best practices, letting business users tailor these frameworks to their specific needs through intuitive interfaces, tweaking what makes their organization unique without reinventing what makes their industry successful.
This approach aims to balance short-term usability with long-term flexibility, helping organizations adapt without relying on technical resources or custom code.
Key Factors When Implementing Enterprise AI
For enterprises with both the capital to invest and the highly specialized functions to justify it, building custom AI solutions can make sense. Otherwise, like anything worth doing, tailoring AI within an organization is worth doing well—and each must determine their best path forward.
Despite the promise of visual tools and business-led customization, organizations often underestimate the integration complexity involved. Enterprise workflows rarely operate in isolation—they span CRMs, ERPs, ticketing systems and custom databases. Lacking the ability to share data seamlessly between enterprise systems, AI’s promise of boosting efficiencies is hindered; its value is, therefore, greatly diminished. Enterprises must plan for data governance and access control as they spin up their AI solutions, or they risk introducing a new set of operational headaches under the guise of innovation.
There is also a steep learning curve associated with how business teams interact with AI systems. Even the most user-friendly tools require a shift in mindset, and non-technical users often need dedicated onboarding, examples grounded in their business context and ongoing support to become effective AI “co-designers.”
Additionally, not all visual tools are created equal; some oversimplify functionality, while others introduce hidden complexity under the hood. Striking the right balance between power and usability is challenging—and until that’s achieved, they will continue to rely on technical teams, undermining the very goal of business-led autonomy.
While they may not need full engineering teams to build custom agents, organizations should ensure that teams with strong business context participate during configuration. Some general technical knowledge should also be available. With this foundation, they can move forward without diverting full engineering teams.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
