Peak finds the money hiding on the shelf

A warehouse can be full and still disappoint a customer. The wrong products sit patiently on shelves; the right ones are somewhere else. At Eurocell, a British manufacturer and distributor of building products, Peak tackled that expensive contradiction across more than 220 branches. Its January 2025 account reported £1.86 million in inventory released and a 6.7% increase in product availability.

That is a useful introduction to Peak, a Manchester AI company whose business is helping other businesses decide what to buy, where to put it and how much to charge. Its material includes window components, rental tools and construction quotations. The drama happens in working capital and gross margin. You could walk past it every day.

Now part of UiPath, Peak occupies a revealing corner of enterprise AI: the place where a prediction has to become an operational decision. A convincing answer on a screen is only the beginning. Someone, or some software, still has to change the replenishment order.

The shelf and the price tag are connected

Peak sells a cloud software platform, applications and implementation expertise. Its core products, Inventory AI and Pricing AI, address two decisions that are often managed by separate teams. Buying too much creates pressure to discount. Discounting changes demand, which changes what should be bought next. Peak’s proposition is to connect those decisions using a business’s own data.

Inventory AI covers demand forecasting, safety stock, replenishment, reordering and production planning. Pricing AI covers markdowns, promotions, list prices and quotations. These are configurable products for companies with their own products, locations and commercial constraints. The intended users include planners, merchandisers, sales managers and data teams working in retail, consumer goods, manufacturing and distribution.

The platform sits alongside existing systems. It gathers and organizes data, supports models and delivers outputs through integrations and applications. Its Data Bridge connects to established data infrastructure. Peak’s AWS and Snowflake relationships support that approach: customers can add decision software to investments they have already made.

The commercial arrangement follows enterprise software economics. Customers license the technology, with services supporting adoption and delivery; Peak’s license agreement ties charges to the relevant order form. The practical buying decision is therefore a project with a scope, an integration workload and a business outcome attached. A demonstration is the start of that conversation.

Concrete, quotations and a test of trust

At Heidelberg Materials’ UK division, the first weak point was the pricing decision inside an established SAP process. Spreadsheets and sales managers’ experience produced inconsistent quotations. The business struggled to see whether discounts were excessive or prices were costing it sales.

UiPath gathered information from different systems and passed it to Peak. Peak evaluated possible prices and recommended a pricing band. Sales managers could inspect comparable wins and losses, giving them a way to judge the recommendation.

The initial rollout split inquiries evenly between Peak and the existing approach. Positive results encouraged Heidelberg to work toward wider adoption. Peak’s case study reports a 2% improvement in conversion rates, quotes returned in under an hour instead of up to eight, and projected annual savings exceeding 10,000 staff hours.

The instructive detail is the sequence: comparison, evidence, expansion. Trust grew around a decision salespeople already understood. The system also allowed commercial judgment: where materials were time-sensitive, managers could prioritize volume over margin.

The tools come back. That complicates things.

Speedy Hire illustrates why the details of an industry matter. A retailer sells a product and waves goodbye. A rental business expects it back, perhaps needing repairs, before sending it out again. Demand planning must account for rental duration and the asset’s condition on return.

In Peak’s published account, Speedy managed 3,500 products across 200 UK depots and offered four-hour delivery and collection. Its asset and logistics teams used allocation and replenishment applications to decide which equipment belonged where. Peak reported average monthly inventory savings of 4% while satisfying 8% more demand, and allocation decisions taking an hour rather than over a week.

Eurocell’s implementation had a different operating rhythm. Peak produced dynamic restocking targets and fed them into its ERP system. That connection matters: branch staff could work with updated inventory targets inside the company’s operating process. The reported cash release reflects less money held in stock. It should not be read as an identical increase in profit.

These customer accounts describe particular deployments, with particular starting points. They are useful evidence of what teams changed. They are not a universal savings schedule.

A pub conversation meets venture capital

Richard Potter, David Leitch and Atul Sharma founded Peak in 2015. Potter became CEO, Leitch CIO and Sharma CTO. Peak’s own origin story begins with a conversation in a pub, a pleasingly ordinary setting for an enterprise software company.

The ambition was broad: make AI available to businesses that could not assemble a large specialist team themselves. Peak called its field “decision intelligence,” bringing data science and commercial decision-making together. Its current product focus gives that phrase a more concrete meaning: inventory and prices.

Investment followed. An extended Series A brought $12 million in April 2020, led by MMC Ventures and Praetura Ventures. Oxx led a $21 million Series B in February 2021, with MMC, Praetura and Arete participating. That August, SoftBank Vision Fund 2 led a $75 million Series C. Peak said the money would support product development, research and international hiring.

Its public culture is unusually explicit. The company lists five values: smart, curious, open, driven and responsible. It calls employees “Peakers,” publishes a culture manifesto and runs AltitudeX, a gathering for commercial and technical audiences. Those are statements of intent and visible company practices. They explain the organization Peak wants to build without pretending to describe every employee’s experience.

“Every company needs its own AI.”

Peak’s stated proposition

The acquisition price needs its small print

UiPath acquired all of Peak’s equity on March 7, 2025, and announced the deal five days later. For the buyer, the attraction was specific: pricing and inventory intelligence, established retail and manufacturing relationships, and technology that could support specialized AI agents.

The cost becomes clearer in subsequent accounting. UiPath’s quarter-ended-April-2026 filing records total purchase consideration of $38.1 million: $30.3 million in initial cash and $7.8 million in deferred and contingent consideration, measured at acquisition-date fair value. An earlier provisional figure was $40.1 million. These are acquisition-accounting amounts, distinct from the capital Peak previously raised.

UiPath’s strategic explanation also points toward SAP customers, where pricing and inventory are familiar problems. There is a practical fit: automation moves information between systems; Peak supplies decisions about products and prices. The Heidelberg example had already put those capabilities together.

By its December 2025 review, Peak said it had launched Agentic Commercial Pricing, Agentic Inventory Management and Agentic Merchandising. The change in vocabulary carries a change in scope: the offering now extends from recommending decisions to executing parts of the workflow.

Who gets to press the button?

Commercial Pricing combines forecasting and price recommendations with actions such as quote generation and system updates. Peak describes approved operating parameters and human oversight for complex situations. Its product page still labels the Negotiation Agent “coming soon,” a reminder to separate the available product from the roadmap.

Agentic Inventory Management extends across purchasing, production and fulfillment. Agentic Merchandising combines inventory and pricing across buying, allocation, promotions and markdowns. Their shared design reflects a commercial reality: changing one variable can upset another. A promotion that sells well can still be awkward if replenishment cannot keep up.

Debenhams Group provides a retail example. Its announcement describes rolling out Peak across brands including PrettyLittleThing, BoohooMAN, Karen Millen and Debenhams, replacing manual pricing work with automated decisions for thousands of SKUs. The group reported improvements in sales, gross margins and cash flow. Peak’s late-2025 updates continued to feature the relationship.

This is a competitive market. PROS supplies AI-based price optimization; RELEX combines forecasting, inventory, pricing and retail planning. Peak’s case rests on configurable applications, a shared platform and its connection to UiPath’s automation. Buyers should compare those capabilities against the actual decision they need to improve, including the option of developing it internally.

Copy the experiment before buying the ambition

Potter’s published advice is refreshingly manageable: start with one complex, data-rich business area and define the desired outcome. Waiting for an organization-wide data transformation can postpone useful work for years. Starting narrowly lets a company develop evidence and experience before expanding.

A practical lesson from these deployments is to choose a decision with a visible owner and measurable consequences. Record the existing process. Compare recommendations with current practice. Track availability alongside inventory, or conversion alongside margin. Heidelberg’s split rollout offers a concrete example of testing a change before extending it.

The conditions matter. As an inference from these implementations, a business with little usable trading history, unreliable inventory records or no route back into operational systems will have more groundwork to do. If nobody owns exceptions, greater automation can simply make confusion travel faster. Small, straightforward operations may also have less to gain from a specialist enterprise implementation.

Peak’s appeal is clearest when complexity has already become expensive: too many product-location combinations, too many quotations, too many discounts to assess individually. The useful question for a prospective customer is wonderfully unglamorous: which decision is costing us money every week, and how will we know when it improves?