Innovators: Kimaru AI and the Case for Decision Intelligence
<p>We recorded late in Tokyo, and <a target="_blank" href="https://evanburkosky.com">Evan Burkosky</a>, CEO of <a target="_blank" href="https://kimaru.ai">Kimaru AI</a>, laid out a claim that is both obvious and ignored. Most supply chains still run on spreadsheets. People glue together ERP exports, POS reports, CRM notes, and a flotilla of pivot tables, then hope the next week behaves like the last one. It rarely does.</p><p>Kimaru calls its approach decision intelligence. Strip away the hype, and you get a layer that sits above the systems of record, learns from the metrics the business already tracks, and proposes concrete actions that a human reviews before anything happens. Instead of looking backward at what sold last quarter, planners see forward, with specific guidance on replenishment, pricing, safety stock, and routes, all framed by the constraints that actually govern their work.</p><p>The company builds what they call a decision digital twin for each user and stakeholder. That twin encodes the choices a role can make, the outcomes that matter, and the limits that cannot be crossed. A set of software agents handles the tedious jobs that eat time, from connecting data and cleaning it, to reconciling mismatched fields across vendors and partners. Once that groundwork is in place, the system runs structured simulations and produces a single best recommendation. The user adjusts or approves, and the order flows back into the existing tools to open a purchase order or move a shipment. Nothing woolly, no free-running bot that buys five million widgets on a whim, and a clear circuit breaker that the user controls.</p><p>The need is plain in any complex chain. An electronics maker in Taiwan hunts for copper, chips, and specialty parts, sells into high-end audio, and now faces tariffs, shifting routes, and new suppliers in places like Vietnam. A missed signal upstream turns into idle inventory and missed revenue downstream. Many teams still try to manage this with manual reports that take days to compile. Kimaru’s pitch is that the same work can take half a minute once the model understands the business and the user’s risk tolerance.</p><p>Large firms often have parts of this effort underway. Data lakes in Snowflake or Databricks. Early agents that score demand or smooth seasonality. Kimaru’s value is the connective tissue. The architecture keeps raw data on site through federated learning, shares patterns without moving sensitive records, and records actions for compliance. It plugs into the stack that already exists and tries to make it useful, rather than selling an expensive rip and replace.</p><p>Under the hood, the tools are not science fair novelties. They are the engines that have powered recommendations and prescriptive planning for a decade, from Monte Carlo to random forests to modern neural nets. The twist is in how those parts are arranged, how cross-company collaboration is modeled, and how the system learns from every correction a planner makes. Evan talked about chaos engines and fractal simulation from his CTO’s doctoral work, and even exploratory talks with quantum groups. The point is not to impress with jargon. The point is to give a planner a credible option on Tuesday morning that shortens a meeting and prevents a stockout.</p><p>This is not a product for crane operators. It serves the inventory lead at a regional grocer, the VP who sets policy for a chain of factories, the manager who runs a distribution center and needs to pick a lane now. Kimaru has spoken with hundreds of people in those seats. All of them admit they live in spreadsheets because the official systems cannot keep up with the chaos outside the building.</p><p>There is also the mood to consider. A wave of flashy pilots has soured many buyers on artificial intelligence. Reports claim that most generative pilots fail to produce value. Evan’s answer is blunt. Language toys are probabilistic by design, which makes them risky as a control surface. Operations need structure and memory. The decision layer gives the model something firm to run on, it narrows the error bars, and it keeps people in charge at the points that matter.</p><p>Kimaru just finished Alchemist, closed out a pre-seed, and is opening a seed round. Interest is strong because the problem is large, boring, and very expensive. The global supply chain ties up vast sums in safety stock to hedge against shocks, and wastes more when plans lag reality. Every hour pulled out of manual reconciliation is an hour that can move product or cut costs.</p><p></p><p></p><p>That is the theme worth noting. The most useful advances rarely sparkle. They remove friction that everyone has learned to tolerate. They give professionals a way to make the same decisions they make today, only faster, with clearer guardrails, and with less risk of groupthink. If Kimaru can turn the spreadsheet habit into a system that thinks ahead, it will not be glamorous. It will just be how the work gets done.</p> <br/><br/>This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit <a href="https://www.keepgoingpod.com/subscribe?utm_medium=podcast&utm_campaign=CTA_2">www.keepgoingpod.com/subscribe</a>
