B2B's Agentic ROI Isn't in the Storefront. It's Locked Behind Legacy Architecture.

A distributor's system detects a tornado approaching one of its warehouses, checks a live weather feed, and reroutes an order already in progress to ship from a different state. The customer never learns any of this happened. Andy Hoar, who recently published a book on B2B automation called Bot to Bot, uses that example in an interview with Watson Weekly to make a point: the biggest agentic returns in B2B commerce sit somewhere other than where most companies are currently spending.

B2B leaders would be wise to take that point seriously.

Most B2B agentic budget goes toward the storefront and the last mile, the parts a customer or a budget committee can actually see. Hoar's argument is that the real value sits further back, in pricing calculation, inventory allocation, and order routing: operational work that nobody outside the company ever witnesses, and that rarely gets funded as a result.

The argument is valid, but it leaves out why the middle became unglamorous to begin with. Most B2B commerce still runs its pricing rules inside an aging monolith, its allocation logic inside an OMS few people at the company fully understand anymore, and its order routing through a stored procedure that survives on institutional memory rather than documentation. A storefront refresh can happen without going near any of that. Automating pricing, allocation, or routing cannot.

ALDO Group's CIO, Matthieu Houle, described a version of this when he walked through the company's AI investment framework. When he took the role, he found ALDO's real supply chain work happening in a parallel system built out of Excel and email, running alongside an ERP that was too rigid to use well. The official infrastructure existed; nobody could get their actual job done inside it. His reasoning for why he put ALDO's biggest AI bet in supply chain rather than the front end was direct: "With an all-in-one solution, you get the feature set out of the box. You find yourself in a place where you're trying to keep up with the features."

A storefront is easy to modernize precisely because it's replicable. A competitor can match it by switching platforms. A pricing and allocation system built from years of undocumented workarounds isn't something a competitor can copy, which is exactly why it's worth the harder investment.

This is where composable and MACH-aligned architecture already does the necessary work: decoupling pricing, inventory, and order logic from any single application and exposing it as services that other systems, agents included, can call directly. Hoar's weather-triggered reroute depends on that kind of access existing underneath it. Most B2B sellers haven't built it. What they have is a working monolith nobody wants to reopen, plus, as Houle found, a shadow system built to route around it.

The gap is closing, but unevenly. Gartner's research, cited by the MACH Alliance, puts headless or decoupled architecture at 67% of enterprise retailers, up from 38% in 2022. Nearly all of that movement has come from consumer retail, where storefront modernization has had a decade's head start. B2B and distribution have faced no comparable pressure, since no customer ever complained loudly enough about a slow pricing engine to get it rebuilt.

None of this is a quick fix, and it's worth being direct about that. Houle's own framework treats this tier of investment as capital-intensive, built on the same business case rigor as any major capex decision, not something bought off a shelf and switched on. Untangling pricing or allocation logic that's been buried in a stored procedure for a decade, or rebuilding a workflow that's quietly been running in spreadsheets for years, is genuinely hard, multi-year work. Anyone selling it as a fast follow to a storefront project is selling something else.

What a company can do in the meantime is find out where it actually stands.

• Can a pricing change ship without triggering a full regression cycle on the core platform?
• Does allocation logic live behind an API any team can call, or only inside the one application that owns it?
• Is there an actual system of record for these rules, or does the knowledge live in three people's heads, or in someone's spreadsheet?

The answers determine whether the sequencing advice is something a company can act on this year or something it's still years from being ready for.