Point your AI in the right direction

Key takeaways

• The promise of AI as a force multiplier only holds if you're clear on which problems you're asking it to solve

• Effective deployment means defining role-specific success metrics before automating anything

• Speed without the ability to evaluate outputs doesn't create efficiency—it creates expensive mistakes at scale

The adoption curve for AI can sometimes feel like a roller coaster—exhilarating highs followed by disappointing lows. When used correctly, AI can be an amazing efficiency booster. When applied in the wrong ways, however, it can introduce friction and complexity.

According to research by Anthropic, AI can speed up some common tasks by as much as 80%. But those gains are often negated by intensifying other work, leading to lower quality output and increased burnout, according to a study by researchers at UC Berkeley.

“The promise of AI isn’t that it eliminates complexity—it’s that it gives organizations a way to manage it,” says Murali Swaminathan, CTO at Freshworks. “But that only works if you're clear on which workflows you're asking AI to orchestrate and how you'll know it's working. Progress over perfection is the right instinct. But progress toward what has to come first.”

Today, nearly 9 out of 10 organizations have deployed AI in at least one business function. But this profusion of competing AI tools can result in what organizational psychologist Bob Sutton calls a “toggle tax” of hours wasted switching between them. This tool sprawl is already taking over many companies. In Freshworks’ Global Cost of Complexity Report, 32% of respondents said they were working with an abundance of tools that were disconnected and duplicative. Proliferation without strategy just deepens the problem.

“AI can amplify the things you want, but it can also amplify things you don’t want,” says Sutton, co-author of The Friction Project. 

Read also: Why every organization needs a ‘Simplifier in Chief’ 

The key to getting the most from AI is identifying opportunities where it can act as a force multiplier and automating only what you can measure.

Find the right use cases

Companies using AI effectively are identifying specific bottlenecks where they believe AI can help, says executive leadership coach Dana Zellers. They’re assigning clear ownership for each model’s work product and measuring impact in ways that are tangible to the business. They’re not simply deploying AI because they feel like they’re supposed to, or automating tasks that are better performed by humans.

Getting that scope right looks different depending on where you’re sitting in the organization, argues Ryan Brelje, Freshworks product marketing manager. 

“For a CIO, that might mean reducing ticket backlog to 30% or cutting mean time to resolution in half,” says Brelje. “For a CHRO, it might mean improving employee experience by accelerating service request response time from days to hours. For a CFO, it might mean reducing cost-to-serve by 20% while preserving satisfaction. Once those goals are clear, the scope of AI becomes obvious.”

In practice, AI excels at helping IT support teams triage issues, find answers quickly, and detect complaint patterns that could indicate larger systemic problems. But deploying it without guardrails introduces its own risks—auto-closing support tickets without giving users the opportunity to escalate to a human, or generating responses without continually monitoring quality and accuracy erodes the trust it was meant to build, says Zellers. Careful measurement is the key to ensuring the guardrails are actually working.

Define what success looks like

AI’s ability to recognize patterns and flag anomalies can be applied across a wide range of business functions—from anticipating IT bottlenecks to forecasting workforce needs. 

The organizations extracting real value aren't treating AI as a magic productivity layer. They're deploying it only where they already have the ability to evaluate outputs, says Ed Brzychcy, founder of the Lead from the Front business practice consultancy and a visiting assistant professor at Babson College.

“If you can't tell whether the output is right, you're not ready for the tool yet,” he says. “Companies that are struggling with AI treat it like a magic productivity boost, then wonder why their AI-generated customer responses sound robotic or why their AI-screened candidates don't fit the role. Speed without evaluation just creates expensive mistakes.”