← All posts

Perspective

Past the threshold

For most companies, the question of whether AI can create value has been answered. The return now depends on what leaders do with it, and most of the opportunity sits in work that moves between systems and people.

Nate Nelson

Nate Nelson Direction, offerings, operating model · · 14 min read

AT A GLANCE

  • The evidence has shifted. Nearly half of companies now report meaningful value from AI, and spending has roughly doubled in a year.
  • The companies pulling ahead redesign whole workflows and measure the result in the P&L. Most others are still improving individual tasks.
  • The largest untapped return sits in work that moves between systems and is carried by people, where the rules have never been written down.
  • A leadership team can start with one workflow, a baseline and an owner, and have a measured result in weeks.

A little over a year ago, the dominant story about AI in business was usage and adoption, with little value proved. In August 2025, MIT’s NANDA initiative reported that 95 percent of enterprise generative AI pilots had produced no measurable effect on the P&L. In January 2026, PwC’s annual survey found that 56 percent of chief executives had seen neither higher revenue nor lower cost from their AI investments.

The picture is changing quickly. BCG’s 2026 research, published in September, found that nearly half of companies are now creating meaningful value from AI, although only 7.5 percent have reached its top tier. Over the same year, AI spending roughly doubled, from about 1.7 percent of revenue to 3.3 percent.

For most companies, we are past the threshold of whether AI can be valuable. The capability is here and it is not going away. If no new model shipped for three years, what is possible today would still be possible. The question for a leadership team is what to do with it, and how the company changes the way it works.

Our view is that the answer depends more on the operating model than on the technology. The companies capturing value are redesigning how work moves.

EXHIBIT 1

In a year, the evidence moved from disappointment to results

Four statistics. 95% of generative AI pilots showed no measurable P&L impact (MIT NANDA, August 2025). 56% of CEOs reported neither higher revenue nor lower cost from AI (PwC, January 2026). About 50% of companies now create meaningful value from AI (BCG, September 2026). AI spending doubled in a year, from 1.7% to 3.3% of revenue (BCG, September 2026).

Sources: MIT NANDA, The GenAI Divide (August 2025). PwC, 29th Global CEO Survey (January 2026). BCG, The Formula for Agentic AI Value (September 2026).

The return sits in the work between systems

Most early AI investment went to two places. The first was individual productivity, with tools that help people write, summarize, search and draft. The second was point solutions inside a single system. Both create real value and both are worth continuing.

The cost in most operations sits in the handoffs. A customer request becomes a quote, the quote becomes an order, the order becomes a schedule, and the schedule becomes a shipment and an invoice. In many companies that path crosses five systems and four handoffs, and at each one a person re-keys information, chases an answer or waits for one. The software holds the records. The people hold the rules.

This explains why so many pilots disappointed. AI was pointed at the systems, while the knowledge that governs the work stayed in people’s heads. BCG’s findings point the same way. The companies it identifies as leaders apply AI in high value areas, reshape workflows end to end and measure the effect directly in the P&L.

Most companies don’t have an AI problem. They have work that will not move.

CASE IN POINT

Seventy-two rules that had never been written down

A manufacturer selling through retail channels ran its order desk with three people. They keyed hundreds of purchase orders a day into an ERP the company had no plans to replace. Working alongside the team, we documented 72 rules that governed how an order was entered. They covered which customer received which pricing, how a partial shipment was handled, and what to do with an unrecognized SKU. None had been written down.

The rules were codified and built into the ERP the company already owned. The system now handles routine orders, and the team handles exceptions and customers. The desk recovered ten to twenty hours a week, fulfillment moved from two or three days to same day, and the result was measured within six to eight weeks.


AI can now reach workflows that earlier systems could not

There is a reason this knowledge stayed in people’s heads. Every earlier generation of enterprise technology could act only on work that had already been reduced to a form, a field or a rule. Work that required someone to read a document, weigh the context and make a call was out of reach at any budget. Companies compensated with experienced people, and over time those people became the operating system.

AI changes what can be reached. It can read the purchase order, apply the rules, flag the exception and route it to the right person. That makes it possible to move the routine into the system and move people up to exceptions, judgment and relationships.

We describe this as humans and AI working as one. The aim is to augment and elevate the people who know the business, and to turn what they know into an asset the company owns.

EXHIBIT 2

The same company can run a different way

Diagram comparing a company today and with Vai. Today, work moves through people across five systems, quote, order, schedule, make and invoice, with four handoffs, and the week is about 70% moving data and 30% deciding. With Vai, the system runs the routine and people handle exceptions, judgment and relationships, and the week is about 20% data and 80% deciding and selling. Illustrative.

Source: Vai

Four kinds of return

When work moves this way, the return shows up in four forms. Leadership teams make faster progress when they decide which one they are pursuing before they start, because that choice determines the workflow, the baseline and the measure.

EXHIBIT 3

Each kind of return has a different starting point

Table of four kinds of return. Efficiency: routine work moves into the system. An order desk recovered ten to twenty hours a week and moved to same day fulfillment. Productivity: people spend their time on the cases that need a person. A receivables team works a ranked queue, and the operation produced a multiple return in its first year. Growth: capacity scales without matching headcount. A distributor built a new operating backbone in a fraction of the usual time and cost. Innovation: work that was never practical becomes possible. One view of the business, assembled from systems that never connected, showing what is happening, what is at risk and what to do next.

Source: Vai founder and client experience. Client details withheld.

What it takes to capture the return

Four practices separate the programs that produce a number from the ones that produce a pilot.

Start with a workflow and a number. The unit of change is a workflow that matters to the P&L, with a baseline agreed before work begins. A list of use cases or a tool selection is a weaker place to start, because neither comes with an owner or a measure.

Build beside the business you have. Redesigning a workflow inside the organization that runs it today is slow, because the existing process has to keep working. A small team with permission to work differently can build the new version alongside the old one, and the old one keeps running until the new one has proved itself. What the first workflow proves sets the standard for the next.

Fund in waves against measured results. Each wave carries a value case, a measured number and one named owner for each automation. The next dollar follows the result. This is how a CFO funds any other investment, and AI should meet the same standard.

Design security and governance in from the first day. Data stays in the company’s own environment. A person remains in the loop on the decisions that matter, with an audit trail on all of it. In regulated supply chains, controls are informed by NIST 800-171, CMMC and AS9100 where customers require them. BCG found that only 5 percent of companies have the full set of controls needed to give AI agents real decision authority, which makes governance a practical limit on how fast a company can move.

Five questions for the leadership team

These are leadership questions. They belong to the CEO, the CFO and the COO, and they cannot be handed to a vendor or left to the IT function alone.

  1. Which workflow costs us the most in handoffs, rework and waiting?
  2. Who holds the rules for that workflow today, and what happens when they leave?
  3. What number would prove the change worked, and what is that number today?
  4. Who will own the new workflow once it is running?
  5. If we were building this part of the company today, with everything now possible, what would we build?
Three colleagues working through a process sketch at a whiteboard

How to start

The first step is small. One workflow, one number and one owner are enough to show a leadership team what is possible in its own operation.

Three steps. 01, Get aligned: agree on the work that matters and the number that will prove it. This takes days. 02, Pick one workflow: build it in weeks, beside the business you have, with a small team that has permission to work differently. 03, Measure it. Run it: finish with a working system, a measured number and your own team running it. Then choose the next workflow.

HOW VAI WORKS

Vai modernizes the way businesses work, with humans and AI working as one. We are experienced operators and AI builders, and we do the work in two parts. Direction aligns a leadership team on the work that matters and the number that will prove it. Build delivers one workflow at a time, in weeks, measured against the agreed baseline and handed to the company’s own team to run.

We work full stack, across leadership, operating model, workflows, and data and AI architecture, and we hold no allegiance to a single AI model. Every engagement leaves a working system, a measured number, and a team that can run it.

If you run a company with work that will not move, bring us one costly workflow and we will tell you what we would do with it.


About the authors

André Wright, Mark Roboff and Nate Nelson are the founders of Vai. Vai is incubated by Verify, which has served the aerospace and defense supply chain for fifty years. Vai works with companies that make, move, fix and serve things, from the middle market to the enterprise.