The future arrives in Mike Neal’s world disguised as a purchase order. It might be a grocery buyer deciding when to lock in beef, a processor choosing how far forward to sell, or a mining company staring at a price curve whose clean line conceals a great deal of doubt. The question is never merely what happens next. It is what somebody should do before it happens.
That distinction has occupied Neal for most of his career. He is the CEO and co-founder of DecisionNext, a San Francisco enterprise-software company built for businesses exposed to commodity prices. Its models produce forecasts for price and supply, test scenarios, and help translate those scenarios into choices about buying, selling, inventory, and timing. There is plenty of mathematics beneath the hood. Neal prefers to talk about the moment a person turns the wheel.
His record makes the fixation look less like a pitch than a recurring character trait. DecisionNext is his third analytics company. Before it came SignalDemand, which applied pricing and margin optimization to food manufacturers and was acquired by PROS. Before that came DemandTec, the retail price-and-promotion software company that went public and later became part of IBM. Different markets, similar provocation: businesses routinely make expensive decisions with less rigor than the stakes deserve.
A skeptic learns his numbers
Neal graduated from the University of Florida in 1983 with a degree in economics and a concentration in statistics. The detail he remembers is not a triumphant equation. It is skepticism. Professor Jim McClave taught him to begin quantitative questions from first principles and to be wary of conclusions that look more settled than the analysis permits. Neal spent summers working at McClave’s company, InfoTech, where statistics left the classroom and entered consequential disputes.
The lesson followed him. After an MBA from Duke’s Fuqua School of Business in 1985, his career moved through Deloitte, Accenture, and logistics before arriving at the first of his software ventures. What he carried was a productive distrust: not of numbers, but of numbers granted more authority than they had earned. It is an unusual foundation for an AI founder and a useful one.
The patent trail is a compact history of the problems he chose: econometric engines, merchandise-price optimization, promotion optimization, subset optimization, and ways to capture heuristic knowledge. The vocabulary is dry enough to wick moisture from the air. The commercial point is lively. A retailer has thousands of prices, a limited number of moves, and customers who do not behave like obedient variables. Which lever should it pull?
DemandTec brought that question to retail at the turn of the millennium. SignalDemand, founded in 2004, carried related mathematics into food manufacturing, where volatile input costs and perishable products make pricing especially unforgiving. In 2013, around the period of SignalDemand’s sale, Neal was also running ForwardTrade, a platform forecasting prices for hundreds of boxed-beef cuts. He described its goal with the fine phrase “bid and quibble”: help buyers and sellers reach a forward price without the old ritual of circling one another.
“We use analytics to prescribe answers based on what the future holds.”Mike Neal, 2015
The forecast has to leave the dashboard
DecisionNext began in 2014 with co-founders Robert Pierce and Graeme Stanway. The company would work in commodities-driven industries, initially spanning food, agriculture, mining, chemicals, and natural resources. In 2019 it announced a $7 million Series A. Its product arrived with a fashionable label, prescriptive analytics, but the underlying argument was admirably unfashionable: a forecast that never changes an action is mostly office decoration.
Neal has observed accurate forecasts being ignored. People receive a number, do not know what to do with it, and retreat to the familiar process that produced yesterday’s result. His proposed remedy is almost comically concrete. When a forecast is ready, sit down immediately and write the actions it suggests. Change a price. Alter a promotion. Buy earlier. Sell less forward. The model must cross the small bureaucratic canyon between “interesting” and “we will.”
The forecast-to-action loop
This is also where Neal parts company with the fantasy of the autonomous oracle. A model can produce an unbiased estimate from the information it knows. It cannot instantly understand a breaking story that alters sentiment before the new price data arrives. An industry expert can. DecisionNext therefore lets experts adjust inputs and compare scenarios. The machine is disciplined but late to breakfast; the person has read the news but may bring bias. Put them together and each can inspect the other’s homework.
The partnership is older than today’s generative-AI excitement. Neal has been talking publicly about “human + machine learning” since at least 2019. He does not present human judgment as ceremonial approval at the end of automation. Domain knowledge belongs inside the forecasting process, where an expert can challenge assumptions, weigh an abrupt event, and decide whether a model’s historical relationship still makes sense.
Simple first, complicated later
Neal’s 2023 guidance on market forecasting begins with a rule that could be taped above many product desks: start with a simple, useful model. Forecasters are tempted to add variables because complexity looks industrious. Each addition should instead earn its place by improving results. Accuracy must be tested against history. Different modeling approaches should compete. And a forecast should show a range of outcomes, not a single number dressed as fate.
A forecast is a range, not a prophecy
That modesty matters in commodities, where volatility is the atmosphere rather than an interruption. Supply, demand, weather, policy, transportation, and market sentiment can all move a price. DecisionNext does not remove uncertainty. It tries to quantify the plausible futures, make assumptions visible, and show the financial consequences of the available moves. Certainty would be a suspicious product. A well-tested range is more honest and often more useful.
It also explains why Neal has kept returning to seemingly unglamorous sectors. Grocery procurement and meat processing may not produce many keynote pyrotechnics, but a small improvement can travel across enormous volumes. A fractional pricing error repeated through a supply chain becomes real money. The invisible decision is where the software earns its keep.
“Having a rigorous view matters as long as it’s monetized through optimizing the forward position.”Mike Neal
Endurance, in software and on foot
Neal’s public advice to first-time entrepreneurs is blunter than his model diagrams. Start for the right reasons. Be driven by the idea. Building a company is “hard - really hard,” and the bad days can become bad months. He says obsessive dedication is what gets a founder through them. This comes from someone who has watched one company become public, another sell for a sum below the capital it had raised, and then started again.
There is a biographical footnote that feels almost too neat: Neal is an amateur ultramarathon runner. It would be easy to turn this into a grand metaphor and make the poor man jog through every paragraph. Still, endurance is present in the work. The same thesis has survived different names, sectors, funding climates, and waves of technology vocabulary. Analytics became big data, machine learning, and AI. The stubborn question remained: does it improve the decision?
Recent glimpses of Neal are less solitary. He has written about gathering DecisionNext’s distributed team every quarter, lately in cities near customers and partners. One meeting took the group to North Platte, Nebraska, and a visit to Sustainable Beef. The scene fits his method. Look at the operation. Talk to the people who know it. Bring the software close enough to reality that reality can object.
DecisionNext has also continued adding food-industry partnerships and publishing a Finished Goods Index that follows the shifting cost of familiar products. Neal’s ambition is larger than another dashboard. He expects rigorous forecasting to become ordinary in commodity businesses, much as revenue management became unavoidable in airlines and analytics reshaped retail. Ordinary is the victory condition. A useful technology eventually stops looking clever and starts looking like how the work gets done.
For all the talk of tomorrow, Neal’s real subject is the present tense. The model offers possible futures. The expert interrogates them. Then someone must choose a price, a quantity, a contract, a date. The work ends not with a prettier chart, but with a buyer or seller accepting responsibility for a move. Tomorrow remains gloriously unwilling to sign an affidavit. Today still needs an answer.