Overview of A Data-Driven Reality Check on AI in Business (Interview with Tom Davenport)
In this interview, Ben Jaffe and Katie Malone talk with Tom Davenport, Babson College professor and longtime analytics/AI observer, about what AI is actually doing inside organizations versus what the hype suggests. Davenport argues that generative AI is real and useful, but it is not a completely new category of intelligence—it’s still prediction, just in a more accessible form. His core message is pragmatic: the biggest business value from AI will come from careful, enterprise-level use cases, disciplined measurement, and strong process design, not from indiscriminate “AI everywhere” enthusiasm.
Main Takeaways
Generative AI is different in usability, not magic
- Davenport says AI is still fundamentally about statistical prediction.
- What makes generative AI different is that it predicts language/images, which makes it easier for non-technical people to understand and use.
- This broader accessibility is why it feels more “smarter” and more revolutionary than earlier machine learning tools.
Businesses often overestimate their AI sophistication
- Executives and board members frequently claim high confidence in AI, but Davenport thinks many overstate their understanding.
- He warns against treating AI as if it began with ChatGPT in 2022.
- Traditional machine learning and analytical AI still matter a great deal, especially for structured decision-making.
Data science is changing, not disappearing
- Davenport argues that basic data science tasks have been increasingly automated for years.
- Generative AI lowers the barrier further, but that doesn’t eliminate the need for skilled practitioners.
- The role is shifting toward:
- deployment
- stakeholder management
- change management
- product thinking
- understanding business requirements
A key line from the discussion: “No deployment, no employment.”
His point: data scientists must help deliver real-world adoption and impact, not just models.
AI in Business: What Actually Creates Value
Individual productivity gains are hard to measure
- A lot of AI use is personal productivity: faster emails, drafts, summaries, slides, spreadsheets.
- Davenport says this may improve employee satisfaction and retention, but it usually does not produce measurable enterprise value on its own.
- Organizations rarely measure these gains rigorously.
Enterprise use cases matter more
He argues that real ROI comes from enterprise-level workflows and mission-critical applications, such as:
- software development
- customer service
- FDA or regulatory submissions
- legal drafting
- pharmaceutical and compliance workflows
In these cases, the value comes from doing a business process better, faster, or more reliably—not just from generating content.
Human-in-the-loop creates a tradeoff
- AI often still needs human review to avoid errors and hallucinations.
- But if a human must check everything, the productivity gains shrink.
- Davenport says this is one reason some AI deployments are less transformative than advertised.
Risks and Problems
“Work slop” and “process slop”
Davenport distinguishes between casual low-quality AI output and something larger:
- Work slop: bad or low-value AI-generated content passed between colleagues.
- Process slop: when AI-generated low-quality output spreads across an entire business process.
Examples he gives include:
- recruiting pipelines
- college admissions
- scientific peer review
- legal and consulting workflows
His concern is that AI may lower trust in organizational processes if everyone is generating, reviewing, and re-generating mediocre content.
Backlash is growing
- He notes growing skepticism, especially among younger users who use AI regularly but are also becoming wary of its effects.
- Many people like AI for personal use but distrust it in schools, hiring, writing, and creative work.
- Davenport suggests the optimism around AI’s productivity impact is still much stronger than the evidence.
How Davenport Uses AI
Practical, not ideological
He says he uses AI for tasks he dislikes or wants help with, such as:
- analyzing survey data
- working with genealogy/family tree files
- drafting and restructuring articles
He finds it useful for:
- finding patterns in data
- producing rough drafts
- doing tedious background work
But he still edits heavily and does not trust it to write polished work independently.
AI in teaching
As a professor, Davenport encourages students to use AI—but only transparently and critically:
- try multiple prompts
- verify factual claims
- check citations
- add personal insight
- improve the quality beyond what AI gives them
He says students often resist because proper use takes away the “easy productivity” gains.
Looking Ahead
The future is disciplined experimentation
Davenport argues companies need:
- clearer process ownership
- better workflow measurement
- a culture of experimentation
- a smaller number of high-value AI use cases
He is especially interested in:
- knowledge management, which AI may revitalize
- agentic workflows, though he thinks most companies are not yet ready
- European firms, which he sees as more process-oriented than U.S. firms
Bottom line
Davenport’s overall view is cautious but not dismissive:
- AI will matter a lot
- many companies are using it too loosely
- the biggest benefits require structure, measurement, and thoughtful deployment
- widespread job destruction is not imminent
Notable Quotes and Ideas
- “It’s all just predictions.”
- “No deployment, no employment.”
- AI is useful, but most current use cases are still more about convenience than provable business value.
- The real challenge is not whether AI can generate output, but whether organizations can turn that output into reliable, measurable outcomes.