AI Should Automate the Work Around Relationships. Not the Relationships.

Every founder I talk to is being sold the same story right now: put AI in front of your clients and watch growth take care of itself. Chatbots that sound like you. AI that "personalizes" outreach at scale. Automated everything.

I don't buy it, and increasingly, neither do the people funding these experiments.

In July 2025, MIT's NANDA initiative published The GenAI Divide: State of AI in Business 2025, a study built on 52 executive interviews, a survey of 153 leaders, and an audit of 300 public AI deployments. The finding that traveled: 95% of generative AI pilots showed zero return. Not underwhelming return. Zero. Only 5% of the initiatives studied created measurable enterprise value (MIT NANDA, 2025).

That number hasn't aged out, either. McKinsey's most recent State of AI research, covered by The Register in August 2026, found that only 37% of organizations attribute any earnings impact to AI use — essentially flat year over year — and just 6% qualify as "high performers" capturing real EBIT impact (McKinsey via The Register, 2026). McKinsey's own language for the gap: "organizations' conviction in AI is growing faster than the immediate financial returns" they can measure.

I've spent my career in field marketing, sales ops, and now building AI systems for Main Street and lower middle market firms, and I don't think this is an AI problem. It's a targeting problem. Companies are pointing automation at the part of the business where the ROI was never going to show up — the relationship itself — instead of the mountain of unglamorous work sitting around it.

Where the failed pilots go wrong

Dig into the MIT findings and the pattern is specific: the AI initiatives that failed were almost all customer-facing, front-office bets — sales and marketing tools meant to interact directly with the people companies were trying to build trust with. Meanwhile, the initiatives that worked were something else entirely. Back-office automation — document processing, procurement, risk review — delivered $2 million to $10 million in annual savings for the companies that built it well (MIT NANDA, 2025).

Read that again. The ROI wasn't in the relationship-facing layer. It was in everything underneath it.

That tracks with what I see inside B2B client businesses, and it tracks with what clients themselves are telling researchers directly. Northwestern Mutual's 2025 Planning & Progress Study — a Harris Poll survey of 4,626 U.S. adults — found that when it comes to actually making a financial decision, Americans trust human advisors over AI by a wide margin: 56% trust a human to build a retirement plan versus 13% for AI; 53% trust a human to manage an investment portfolio versus 15% for AI (Northwestern Mutual, 2025). A separate 2025 study from Bread Financial found 65% of people want human involvement when managing their money — and most wouldn't switch to an AI-only model even for a guaranteed higher return.

Clients of trust-based industries — healthcare, wealth management, senior care, financial services — aren't rejecting AI. They're rejecting the idea that AI is the relationship. What they want, per that same Northwestern Mutual research, is an advisor who uses AI well: 47% of Americans said they'd prefer working with a financial advisor who understands and uses AI, a number that climbs to 54% among Gen Z and millennials.

That's the whole thesis in one data point. Not "replace the human with AI." An advisor, backed by AI.

What "automate the work, not the relationship" actually looks like

This isn't an abstraction. Wealth management is already proving the model out. KPMG's research on agentic AI in the sector found that automating meeting prep and admin work is freeing up enough capacity for each advisor to manage 50 to 60 more meaningful client relationships — not fewer, more, because the AI took the paperwork instead of the phone call (summarized via X1 Wealth, 2026). Advisor360's January 2026 survey of 300 firms found advisors are already pointing their AI at exactly this layer: meeting summaries and notes (31%), CRM updates (28%), meeting prep (26%), routine communications (25%). The tasks advisors are not handing to AI: generating actual recommendations for clients — just 3%.

That's the divide MIT is describing, playing out inside a single industry. Automate the scheduling, the notes, the follow-up, the data entry, the "did we log this in the CRM" work — the stuff that eats a founder's or an advisor's week without ever being the reason a client trusts them. Leave the judgment, the conversation, the relationship exactly where it is.

The infrastructure question

Here's what I'd ask any founder in a relationship-driven business before they spend another dollar on an AI tool: is this automating something my clients never see, or is it automating the moment they decide to trust me?

If it's the first, there's real ROI sitting there — the same $2–10 million a year MIT found in back-office automation, just sized to your business. If it's the second, you're building the 95% of pilots that go nowhere, because you're trying to automate the one thing your entire business model depends on a human doing.

This is the argument for building infrastructure — CRM systems your team actually uses, lead routing that works while you sleep, referral tracking that doesn't live in someone's head, follow-up sequences that fire on time every time — instead of chasing the idea of an AI relationship manager. The infrastructure isn't the sexy pitch. It's the one with the return attached to it.

Sources

• MIT NANDA, "The GenAI Divide: State of AI in Business 2025", July 2025
• McKinsey & Company, State of AI research, as reported by The Register, August 2026
• Northwestern Mutual, 2025 Planning & Progress Study, conducted by The Harris Poll, n=4,626, January 2025
• Bread Financial, 2025 consumer research on AI and money management (cited via X1 Wealth, 2026)
• Advisor360, January 2026 GenAI adoption survey, n=300 advisory firms (cited via X1 Wealth, 2026)
• KPMG, agentic AI in wealth management research (cited via X1 Wealth, 2026)