The redesign is the value creation

In 1899, electric motors supplied less than 5 percent of the mechanical drive in American factories. It took roughly 20 more years for them to supply half. Electrification did not lift productivity growth in manufacturing until the early 1920s, 40 years after the first central power station opened for business.

Much of the wait was for the factories, which took decades to change around the motors.

Electricity paid off once the work was redesigned around it, and so did the network. Companies that bolted the new technology onto the old way of working ended up paying for both.

AI is the next power source of this kind, with one difference that matters to a sponsor. A company in the lower middle market can redesign around it inside a single hold period.

For most of those years, owners electrified the plants they already had, using group drive. Each electric motor turned a section of shafting that drove a group of machines, and the original belts, shafts, and engines stayed in place.

The building stayed the one steam had required. Long shafts lost power to friction, so plants rose several stories to keep the shafts short, and every machine sat where a belt could reach it.

The retrofits paid for themselves in lower power use and better control of machine speed. The old equipment stayed on as capacity, though, so each plant carried 2 power systems, and measured productivity gained little.

The plant ran on a new power source and kept its old layout.

The delay was rational. Replacing a plant that still worked rarely paid, so the redesign waited for new construction.

The fastest-growing industries built new electric plants first. Electrification gathered real momentum only after utility rates fell sharply in real terms, from 1914 to 1917. The wider rebuilding came with the investment boom of the 1920s.

Much of the gain came with unit drive, a motor on every machine. With the overhead shafting gone, a building no longer needed heavy bracing, and plants could go to a single story. Machines could be placed in the order of the work and moved when the product changed.

A change in one department no longer meant shutting down power to the whole mill. The work itself was redesigned, from the layout of the floor to the way materials moved through it.

The economic historian Paul David found that growth in factory motor capacity during the 1920s accounts, statistically, for about half of that decade's acceleration in manufacturing productivity growth. Electric power had been on sale for 40 years. The acceleration came in the decade industry rebuilt its plants around it.

The pattern repeated 70 years later.

Dell had been building computers to order and selling them direct since 1984, taking orders by phone. In 1996 it moved that model onto the web and automated build-to-order. Each order told the factory what to build, and components were bought as orders called for them.

The economics followed the design. Dell avoided the unsold machines that lose value on a shelf, skipped the cost of retail stores, and passed the savings to customers as lower prices.

Blockbuster pressed Netflix harder than many people remember. Its online service started as a business completely separate from the stores. The stores were not even connected to the internet, and store managers and franchisees resented an online service that took their sales.

In 2006 Blockbuster let subscribers return an online rental in a store and leave with another, and nearly 1 million signed up in 2 months. The stores were still not connected, so each visit gave away a free rental, paid for by a store business in decline and heavy with debt.

Netflix had predicted that falling store revenue would leave Blockbuster struggling to fund its online business. A new chief executive moved the money back to the stores, and Blockbuster went bankrupt 2 years later.

Both companies had the same network. Dell connected its orders, its factory, and its buying to it, on a model already built for direct sales. Blockbuster kept it outside the stores and made the stores pay for it.

Many companies are adding AI the way factories added motors in 1910. A chat assistant on every laptop, a feature switched on inside the CRM, a pilot in customer service. The work stays the same, the license fees land in SG&A, and the gain never reaches EBITDA.

Writing in 1990, David saw the same overlay with computers, as companies kept paper procedures running beside the new systems, sometimes to the detriment of both. He also cautioned that computers are not dynamos, and AI is not either. The overlay is the part of the story that repeats.

Today's version of the serviceable plant is the process that still works: the quoting routine, the pricing spreadsheet, the monthly close. Nobody tears it down, so the new technology gets added on top.

The math that kept the shafts turning still applies. The old process works, so replacing it looks like cost, and adding a tool looks like progress.

Redesign starts from the work. Take 3 jobs found in almost every company a sponsor buys.

Quoting. A seller reads a request, checks the contract price in one system, stock in another, and lead times in a spreadsheet, then types the quote. Redesigned, price, stock, and lead time sit in one system of record, and an agent drafts each quote at the contract price. A seller approves it, every quote is logged, and the measures are hours from request to quote and quotes per seller.

Cost-to-serve. The typical company knows gross margin by product and estimates the cost of serving each customer: freight, warehousing, returns, rush orders, the account that calls every day. Redesigned, those costs are allocated from the ledger and the warehouse system to every account and SKU each month. Prices, order minimums, and freight terms are then set against contribution margin the company can see.

Inventory. Reorder points live in a planner's head, set once and adjusted after a stockout. Redesigned, demand and lead time are forecast by SKU, and safety stock is set from the forecast's own error. The planner works the exceptions, and the aim is less working capital at the same fill rate.

In each case the model is the smallest part of the work. The data has to come from one place, the process has to be redrawn, and one person has to own the number.

The old process works, so replacing it looks like cost, and adding a tool looks like progress.

That sets the order of the work. First, learn how the work actually happens, from the people who do it. Then build one system of record, and the unit economics on top of it.

Agents come after that, once they have clean data to work from and a number to move. A cadence comes last, and it keeps the gains in place after the build is done.

Until recently, rebuilding a company's systems meant an ERP program: a multimillion-dollar budget, years of implementation, and software the company bent itself to fit. Many companies in the lower middle market could not afford it, and few sponsors could wait for it.

That has changed. Custom systems that once took years can now be built in a fraction of that time, around the way the company works. A sponsor can redesign the work in the first year of a hold and still show a buyer years of results.

A custom system is an asset only if the company can keep it running. The company owns the code and the data, and the system is documented well enough for its own team, or anyone it hires, to support.

That is the opportunity in companies that never re-engineered, bought at yesterday's multiple. The redesign is the value creation.

Redesign is also easy to imitate. A new tool, a dashboard, and a slide about AI can pass for progress for a year. So we hold every initiative to 3 tests, before it starts and again after launch.

1. No extra tools. It replaces work or software the company already has, and the old way goes away.
2. Real adoption. The people it was built for use it without being chased.
3. Attributable return. A baseline is set before launch, so the gain can be traced to this work.

The tests sound obvious. They are easy to skip, because each one requires deciding in advance what failure looks like. After launch, an initiative that fails a test by the date set before it started is stopped.

Electricity and the network sorted companies by what they rebuilt. We expect AI to sort them the same way. The difference this time is speed: the redesign that took factories decades can now happen inside one hold.