Ben Marans

The through-line

Ben Marans keeps returning to a single conviction: the metal service center sector is the most technologically neglected corner of the American economy, and that neglect is a business opportunity rather than an accident. He describes arriving at it by elimination, having analysed 33 industries in depth while doing vertical SaaS investing, and concluding that this one is "by every metric the most Legacy most old school" [1]. The reasoning is that the epicentre of the industrial supply chain is the service center layer, and that legacy software there "cause tremendous inefficiencies hindered growth loss profits" and keeps an industry that "has basically provided America with its industrial success in the Dark Ages" [1]. He pairs that commercial thesis with an explicitly patriotic one: steel and aluminium are everywhere, the country was built on them, and the sector has been "completely ignore[d] and underserved" by the venture-backed technology world he came from [1][2].

What has shifted over roughly three years of building is where he attacks first. Early on the ambition was framed around replacing the core system of record, achieving ERP feature parity and then layering AI on top [1]. Later the emphasis moves to the commercial side of the house, with CloudForge described as a sales CRM and prospecting tool for service centers and distributors, sold on the argument that ripping out an ERP is frightening and slow while a growth tool "has very limited downside" [2]. The underlying philosophy is constant: build for the workflows that already exist in this specific industry, and make adoption cost the customer almost nothing in effort, because "it's our job to deliver value rather than just deliver a tool" [2].

On why this industry has the oldest software in America

Marans traces the problem to early success. Manufacturing was among the first categories to adopt business software in the 1980s because it is process-driven, which is why the sector is still full of "green screens MS DOS as400 Matrix based" systems [1]. Service centers, which he treats as a variant of manufacturers, adopted early and got stuck. The verticalised incumbents he names, Nmar, Invera and PS data, were founded during or shortly after the Cold War, and in his account they have not meaningfully changed their platforms since, making them "the definition of a sleepy incumbent" [1]. The functional gaps he lists are specific: no CRM, no quote management, no customer engagement tools, no e-commerce portals, no dashboarding inside the tool, inventory views so clunky you have to export a PDF to read them, and no use of AI whatsoever despite the volume of data sitting in the system [1]. He adds a workforce consequence: as service centers hire younger people, those people "have no idea how to use" the old interfaces [1].

He is careful to distinguish service centers from manufacturers, because the distinction explains why generic tools fail. With a manufacturer, "you take 10 products and make one" against a bill of materials; in a service center, "you take one product and you make 10", cutting a 65-foot tube into pieces, laser cutting, galvanising, sending it out for further processing, then tracking and inventorying every piece along the way [1]. That inversion is why he sees plenty of venture-backed activity modernising manufacturing ERPs and quoting, and almost none on the service center core system of record [1]. Where the sector has seen technology arrive is on the system-of-engagement side, and he points to Reibus as the example, a marketplace that has raised over a hundred million dollars, while noting he does not think "the model with that has been like nailed perfectly" [1].

On the AI applications he thinks actually matter

Price forecasting sits at the top of his list, because rapid fluctuation in steel and aluminium prices is what makes quoting slow and margin capture unreliable [1]. He wants to analyse buying patterns, historical sales and purchasing data, and supply and demand dynamics to forecast where hot rolled coil goes next month or in 2026, and to push that granularity down to "the Alloys the forms the grades the dimensions the chemistries the region the End Market the buying Persona down to the psychographic data" [1]. His critique of the status quo is that the industry's index is at best a month lagged and reflects mill pricing, meaning "it's not even representative of how the service center needs to Price", only what they pay the mill [1].

From there he builds outward. Demand planning off the same data; automated quoting calculations driven by granular ERP data points so quotes come back fast at the optimal price; an AI agent using natural language processing to take text-based RFQs, check inventory and return a quote; and pricing suggestions derived from the customer's own transaction data to "meet the price takers at the highest possible price point" [1]. The most ambitious step is centralised sourcing: a fabricator specifies a product and lead time, and the system circulates the RFQ to the service centers most likely to hold it, which requires classifying each product category into a single standardised data model [1]. The payoff he describes is a self-reinforcing one, a trove of real-time transaction data under a standardised nomenclature of metal products, granular enough to support not just an index but a prediction of it [1].

On winning in the near term with unglamorous features

Alongside the AI roadmap, Marans is blunt that the immediate wins are basic. First is feature parity across procurement, sales, production, shipping, accounting, billing, AR and AP, which he concedes is "a fundamentally like massively robust software platform", plus better UX that simply shaves off time [1]. Then two focused bets. The first is CRM and quoting: today a quote becomes a PDF and disappears, while a salesperson should be able to log in, see a filtered table of open quotes and follow up in a few clicks, because "you're going to make more money that way" [1]. The second is dashboarding, which he calls the low-hanging fruit, replacing complex report and CSV exports with views of quote win ratio by product and customer, inventory by pounds by location, open and delayed purchase orders, and invoice aging [1]. His argument for why nobody has solved this with existing BI tools is a sizing one: "when you're a one to two location service center doing 8 to 20 million in Revenue you've got 10 people like you don't have an analyst just managing a powerbi" [1].

On the two halves of industrial sales, and the half nobody does

Marans splits any industrial distribution sales job into account management and business development, and observes that unlike software sales the bulk of distribution revenue comes from existing clients through repeat orders, RFQs, quotes and purchase orders, which makes personal account management critical and consuming [2]. Business development, he argues, is effectively "non-existent" at the capacity AI now allows, and prospecting has become an afterthought for a rational reason: "why would you sit there prospecting when one hour of prospecting yielded very little ROI?" [2]. His answer is a database of industrial businesses built by a crawler that identifies and enriches companies, an AI-based semantic search that takes your company profile and target criteria and returns matches, and an automated outreach engine sitting on top of CRM workflows [2]. He claims the discovery layer is the harder half, asserting that no service center he has found has "their entire total addressable market discovered and accounted for and documented cleanly" [2]. In practice a user asks for something like data center cooling system manufacturers in the southwest with over 50 employees and gets results with contacts in fourteen seconds, then generates configured outbound with a human in the loop to review, followed by call tasks and notes [2].

He positions the product as slice-and-dice rather than rip-and-replace: it can substitute for a Salesforce or HubSpot, or be used purely as an outbound business development tool, though he thinks the best case is the circular one where prospecting and existing-customer management run in the same place [2]. Existing deployments range from single users to scores of salespeople, and one integrates with the ERP to track sales usage and automatically follow up to reactivate lapsed customers [2]. All of the outbound tooling built before, he argues, was designed for one-time software-style sales, which is "a whole different model" from distribution [2].

On adoption, culture and the fear of change

Marans segments the market by willingness rather than size: roughly 25 percent of service centers are eagerly looking for innovation, about 50 percent think AI sounds interesting but will figure it out later, and 25 percent want to shake your hand and do business the old way [1]. Finding the right ones is the first challenge; the second is that replacing a core system of record can put a business at risk, so "the sales friction is going to be in the fear of changing rather than a value realization", since buyers see the value quickly even at MVP stage [1]. He blames that fear on prior experience with horizontal enterprise software, where SAP arrived, cost ten million dollars and took forever "because they're basically building you software from scratch driven by product managers who don't care" [1]. His counter is verticalisation: because the software is built for workflows consistent across the industry, there is little configuration, and implementation becomes fast, supported by data transformation and migration tooling such as one schema [1]. He treats implementation as a hiring priority in its own right [1].

Later he frames the same resistance as habit rather than principle. Traditional service centers often have no buying workflow for software at all, so they see value and still stall, and sentiment has been soured by large companies that entered the market over the past several years with good intent and did not deliver what was promised [2]. He puts his fit rate at 80 to 90 percent of the companies he speaks with, and defines the exclusion sharply: the only bad fit is "those that don't have the capacity or the desire to grow" [2]. The ICP question reduces to whether you want more business and more clarity in handling current business [2]. His response to the industry's tech-forward deficit is to remove effort rather than argue: the system is agentic and wrapped in a managed service, effectively "selling you an outbound sales rep" [2]. Against all of it he sets a flat statement of stakes: in steel, aluminium or non-ferrous, if you want to grow "you have to do things differently" [2].

On commitment, headcount and what separates the best customers

The differentiator among his customers is not size but resolve. The most successful ones have "a cultural DNA, a top-down impact" where adopting the software is treated as a real initiative rather than an experiment, because "Ideas are great. Execution is everything" [2]. He connects that directly to staffing decisions: when someone leaves, a committed customer may not need to backfill, and one recent client, a subsidiary of a publicly traded company, is using the platform instead of hiring more people, with the tech stack ready for the few hires they do plan [2]. Asked how the product de-risks hiring for a short-staffed service center, he gives three things: a steady and impactful top-of-funnel pipeline, better organisation of tasks across current and new customers, and visibility, "clarity" and single-place tracking that give a manager "the ammo to control your day" [2]. He acknowledges that many service centers are simply too chaotic to make the time and space for change, which is precisely why the product is designed to require very little effort [2].

On vertical SaaS, industrials and raising money on hard mode

On investors, Marans finds the vertical SaaS conversation entirely bifurcated: "it's either you get it and you're excited about it or you don't", and after five words he can tell which, at which point the conversation is not worth having [1]. Among those who do get it, attention clusters in healthcare, retail and e-commerce, with some construction, while industrials and manufacturing are the gap, mostly because investors do not know the space [1]. He expects capital to funnel there and expects the resulting revolution to be experienced by frontline workers [1]. On the funding climate, he is unsentimental that 2023 is far harder than 2020 through 2022, when peers with similar metrics raised on "crazy valuations off of a three sentence pitch deck", and he treats the correction as a return to fundamentals: "you basically get to play the video game on hard mode" [1]. His operating conclusion is focus. Companies that raised heavily and allocated in too many directions fell apart and pulled back, whereas CloudForge has run lean, doing "a lot a lot" with essentially two people and some contractors, spending effectively and descoping deliberately [1].

On learning an industry from the inside, and telling its story

Marans came to metals with no understanding of the supply chain until the end of 2022, from a venture capital role in New York researching niche old-school vertical markets, where he was responsible for building 12 companies and underwrote 12 million dollars of investment, a million into each [2]. Metals began as pure curiosity, on the logic that the material is ubiquitous and he knew nothing about it, and within weeks he concluded distribution was the segment where technology was adequate for existing workflows but nowhere near harnessing what modern software and AI could do [2]. What turned analysis into commitment was an emotional attachment to the industry's importance to American manufacturing, sharpened by reshoring and what he calls the renaissance of American manufacturing [1][2]. His method of learning it was physical: three-plus years embedded in service centers, travelling the country, sleeping near facilities and arriving with the first shift [2]. He also runs a media effort around the same conviction, publishing podcasts on the CloudForge blog, YouTube and LinkedIn to bring service center executives and operators' accounts of the work to a wider audience, and hosting Forgecast [2][3][6]. That instinct for recording what an older era leaves behind shows up outside work too, in the HK Ghost Signs project documenting Hong Kong signage from earlier eras, technologies and cultural movements in a fast-changing city [4].

Takeaways

• Service centers invert manufacturing logic, taking one product and making ten from it, which is why generic manufacturing ERPs and horizontal tools do not fit their tracking, inventory and quoting workflows [1].
• The industry's pricing index is mill-based and up to a month lagged, so it reflects what a service center pays rather than how it should price, which is the gap real-time transaction data and forecasting are meant to close [1].
• Buyers see the value fast; what stalls deals is fear of replacing a core system of record, so implementation speed and verticalised, low-configuration workflows are the real sales lever [1].
• Roughly a quarter of service centers actively want innovation, half are curious but passive, and a quarter refuse outright, so targeting matters more than persuasion [1].
• Prospecting collapsed in industrial distribution because an hour of it produced almost no return; AI-based semantic search over an enriched company database changes that arithmetic by finding the total addressable market first [2].
• The only genuinely bad-fit customer is one without the capacity or desire to grow; everything else is a matter of habit and buying process [2].
• Vertical SaaS investors split cleanly into those who understand the category and those who dismiss it as a small market, and industrials remain the least covered vertical [1].
• Running lean with two people and deliberately descoping focus is what he credits for surviving a funding market where fundamentals, not hype, decide outcomes [1].