Eilon Reshef, cofounder and CPO of Gong, is a seasoned entrepreneur, executive and investor in the internet and software spaces.
getty
The market for enterprise AI software has become, by any measure, overwhelming. New products arrive weekly, existing ones mutate with each model upgrade and the falling cost of building software means that every startup with a large-language model and a pitch deck now claims to be reinventing the category.
For leaders acquiring these solutions, the result is a paradox of choice so acute that many have simply frozen in place.
As a product leader, I am often asked how to evaluate these innovations to determine which ones are actually poised to deliver on their promises. A useful evaluation framework, therefore, needs to go beyond feature checklists or model sophistication and instead should assess how an AI product converts raw data and intelligence into tangible business results.
One approach is to borrow the frameworks that product executives themselves use to build and position their offerings. Business leaders who learn to see the market through a solution provider’s eyes tend to ask sharper questions, filter out noise faster and avoid the costly regret of a misguided purchase.
With that, here are five key principles that will help leaders determine whether they’re investing in AI tools that will deliver value—or slop.
The most useful of these borrowed frameworks is known as Jobs to Be Done, a method popularized by the Harvard economist Clayton Christensen. Its central insight is deceptively simple: Customers do not buy products; they hire them to accomplish specific tasks. A commuter does not want a train ticket; she wants to arrive at work on time without the stress of driving.
Applied to procurement, this means resisting the allure of a slick demonstration. Before evaluating any solution, buyers should catalog the discrete jobs their organization needs performed, group them into clusters and identify which are already well served and which remain stubbornly unmet. The solution provider's task, then, is not to impress but to apply for an open position.
In a market flooded with AI tools that look similar on the surface, clarity about the actual job to be done is the fastest way to shrink a long list of candidates to a short one.
Similar to the Jobs to Be Done concept, once a product is built, proof of concepts are key to evaluating its utility in practice. If a product fails to deliver measurable business benefits aligned to an overarching strategy, it will never deliver transformational benefits and will therefore fail to generate returns.
This is where it’s up to buyers to ruthlessly scrutinize how a product is performing. Is it delivering time or cost savings? Has it unblocked processes and made it easier and more efficient for teams to work? Does it link processes to business strategy?
Without strong “yes” answers to these questions—from across team functions and roles—more scrutiny is needed before fully embracing an AI tool.
An AI product will never be a useful one for businesses unless the outputs it produces are highly accurate.
The rush to bring new AI-powered products to market has, naturally, resulted in a fair amount of AI “slop.” Businesses simply cannot afford to fix the mistakes made by a system that may be generating inaccurate or incomplete results.
One way to increase the likelihood of a product being one that delivers accurate outputs is to seek one that is tuned to a specific use case or industry. These products are likely to include key context within their design that drives accurate, enterprise-grade outputs.
Even once a promising product is identified, a second and arguably harder problem emerges: trusting it. Enterprise technology has always demanded that products be scalable, secure and compliant. But AI has made the stakes even higher as businesses and consumers alike ask what language models are doing with their data inputs.
A recent study by Gong found that nearly six in 10 business leaders have delayed, paused or cancelled an AI deployment over trust concerns. Lack of guardrails and transparency into how these solutions capture and act on company data are key drivers of this trust gap, but that doesn’t mean AI is incompatible with the enterprise.
Businesses would benefit from constructing what might be called a trust hierarchy: a ranked set of criteria, ordered by the severity of risk each addresses, against which solution providers must demonstrate concrete capabilities. The vendors offering products and solutions that account for the assurances that are most critical to companies are the ones that will earn trust and win in the market.
A product that solves today's problems but cannot evolve alongside the buyer's needs is a depreciating asset. That’s why I strongly believe in the power of building products with end users involved in the process.
The most productive relationships tend to be collaborative ones. The best AI solution providers use design partners at every stage of development, convene advisory councils to pressure-test long-term direction and involve hands-on users from requirements through to deployment.
In practice, technology providers that treat product development as a closed process almost invariably drift from the needs of their customers. The history of enterprise software is littered with once-dominant platforms that lost relevance because they stopped listening.
None of these frameworks requires a degree in product management. What they require is discipline: in defining the problem before seeking out solutions, interrogating trust with specificity rather than accepting reassurance and insisting on partnership rather than passivity.
In a market where every solution provider promises transformation, the buyers who fare best are those who have already decided precisely what needs to be transformed. The rest is noise.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。