Should you take Acumatica 2026 R2?

By Peter Nicholson

Every Acumatica release arrives with more change notes than anyone will ever read, and 2026 R2 is no exception: hundreds of changes, from the platform itself down to a single new field on an e-commerce store setup. Nirav and I can't cover them all, so we did what we did for 2026 R1 back in May. We picked the ones manufacturers and distributors should actually care about, then argued about whether to take it.

I come at this from the data and systems side, inside a business that runs Acumatica day to day. Nirav implements it for a living at AdCirrus. Between us we cover both ends of an upgrade: the person who has to live with it, and the person who has to put it in.

Nirav opened with a strong claim. Unlike most releases, he thinks this is one every customer will need to be on, or risk drifting out of support. The reason is the first thing on our list.

The classic UI is gone, whatever version you're on

We've known for a couple of years that the classic UI was on its way out. The modern UI appeared alongside it as a preview, became generally available in 2026 R1 with a switch to flip between the two, and in 2026 R2 the switch is gone. Modern UI from here on.

What I hadn't realised is that the push isn't really coming from Acumatica. It's the browsers. They're removing XSLT, an old way of rendering web pages, for security reasons, and the classic UI depends on it entirely. Chrome stops running it on 17 November. So whatever version you're on, anything still rendering through classic simply stops working. Acumatica has already pushed hotfixes to earlier versions, which went live for 2026 R1 customers still on classic a couple of days before we recorded, so there's a little more breathing room. But if you have customisations still built on classic screens, they need converting, and depending on how many you've accumulated, that could be a decent-sized job.

I'll be honest about my one grumble. We embraced 2026 R1. We moved to the modern UI on purpose, and we fed back the bugs and glitches we found. A fair few of those have now been fixed in R2, not R1. That doesn't help a customer who doesn't want to pay for another upgrade just to get fixes for problems they reported in the version they're already on. If you want people to adopt early, look after the ones who did.

That aside, I like the modern UI a lot. It's cleaner, it future-proofs the system, and it's already more flexible. I can customise the header and hide fields nobody uses without writing a line of code.

Bring your own AI, but read the pricing small print

The launch event was AI, AI and more AI, and the piece that matters most is the MCP connector. MCP is the standard way of letting an AI model reach into another system's data and tools, and Acumatica now lets you connect the model you already pay for, ChatGPT for business or Claude Enterprise, straight into the ERP.

Nirav's point is that people are already doing this, badly. They export a report to Excel, upload it to their AI of choice and start asking questions. Connecting the model directly takes that conversation a lot further. The first thing customers ask him is whether it's safe to connect a system full of vendor costs, prices and margins. His answer is that it stays inside your own AI account. The question nobody can answer yet is cost: how many prompts, how many tokens, what will it charge? Acumatica says itself that this is phase one, and that it's still learning how customers will use it before it sets proper guardrails on pricing and functionality.

That's where I get twitchy. Connecting your own model kills shadow AI, the person exporting a GI to Excel and pasting it into their personal ChatGPT, and that's a good thing. But consumer AI plans are heavily subsidised, by investors and by the fight between Anthropic and OpenAI for your money. If you saw what your usage would cost at token prices you'd be frightened. I get through thousands of dollars' worth a month on a $100 plan. And "get customers onboarded now, sort billing out later" is a very good sales tactic. Have a little sniff. Build your business processes around it. Then the price goes up, and by then you depend on it. Add the fact that those calls will be logged as API transactions in a consumption-based ERP, and I'd want to see the pricing before anyone builds a process on it.

Data masking doesn't get you out of it either. Acumatica can strip sensitive values before they reach the model, but ask "which accounts are overdue?" with the customer names and amounts masked and you won't get much of an answer.

Nirav's view, which I share, is that no publisher has hit a home run here yet. Anyone trying to right now is probably wasting effort, because it will look completely different in six months. What will still be standing is MCP, and that's what excites me most, because no business lives entirely in its ERP. The contract is in SharePoint. The conversation is in Outlook or Teams. The project timeline is in monday.com. An agent that can reach all of them answers a better question than one that can only see Acumatica.

The catch is that it's only as good as your GIs. If you want to be early, start now. Audit your Generic Inquiries, get rid of the ones nobody uses, make sure the rest are complete, and clean up the master data underneath them. There's also a new tab on GIs where you can tell the AI what an inquiry is for and what each column means in business terms. Acumatica can build the ecosystem. The foundations are ours to lay.

The AI assistant is for people who won't read the wiki

The other AI feature is the assistant in the top-right banner, and it's the one I think most end users will actually touch. Picture a customer service operator with fifteen GIs and no idea which one has the answer, no intention of reading the wiki, and one ERP expert in the building who's already bottlenecked answering everyone else's questions.

Some of Nirav's customers have been running it under managed availability in R1, after signing a waiver, and the feedback has been good. It's quickest for the people who query the system constantly: a sales manager asking for last month's biggest sale or their top five customers, a production manager checking something mid-shift. He's also heard Acumatica talk about launching agents from the assistant to walk through processes such as the month-end close.

Two things from the launch demo won me over. You can ask for a chart, say open AR for your top ten customers, and drag it from the chat straight onto your dashboard. That fixes my long-standing complaint that people would get bored typing the same question every morning. And when the answer is in your instance's own wiki, it doesn't consume AI units at all, which I suspect covers most of the "how do I set up a prepayment?" questions. I'm also hoping it nudges people towards features they've forgotten they own, instead of another spreadsheet on the desktop. It's going to suggest something in Acumatica, not an Excel file.

A data warehouse? Sort of

Acumatica is calling this a native data warehouse. For now it's an enhancement to GIs that stores their results, a bit like a materialised view, instead of recalculating ten thousand rows on every refresh when 9,900 of them never change. I can see it helping with snapshots too: the order book on the first of the month against the month-end position. It won't replace a proper data warehouse or data engineering, and it doesn't need to.

Nirav was less sure who it's for. For a manufacturer turning over $5–10 million, he couldn't yet see what it changes. He thinks it fits the upper end of Acumatica's customer base better, the $100–200 million, multi-company businesses. We'll both be watching what customers actually do with it.

Available to promise, or: dates, dates, dates

This is the one I was most pleased to see, and the one I missed by a single release. Available to promise calculates the earliest date the full quantity on an order line can ship. It looks at stock on hand plus incoming purchase orders, transfers, production orders and kit assemblies, and sets all of that against demand that's already committed. Setup is three steps: switch it on, choose which order types count, and set a look-ahead horizon.

I've seen the alternative first-hand: a worksheet on a purchaser's desktop with item lead times, updated after conversations nobody else was part of, and a constant back-and-forth of "when are we getting more of this?" Go back to a customer three or four times with a new date and it isn't a promise any more. It's a guess.

The demo also showed availability and an ATP date for every warehouse in one pop-up. That matters when the customer wants it next week. The quickest location isn't always the right one if it means expedited freight from the far end of the country, and now you can see the trade-off and choose.

It all depends on dates. Someone I know who listens to the show told me that for six months of their project, all they heard was "dates, dates, dates". They were right. Purchase due dates and promise dates, sales ship dates and commitments, production ready dates: keep them honest and Acumatica will do an awful lot of the working out for you.

The shop floor kiosk, and a workshop I sat in

This one is personal. At Summit 2025 in Las Vegas, Acumatica ran a two-hour round-table on a shop floor kiosk it was toying with. The screens were primitive, and the room was VARs, customers and developers in groups, talking through what they actually needed. Every publisher says it listens to its users. This is the first time I've watched one take an idea from that kind of room all the way to a release, and some of what's on the screen now are things I remember arguing for.

R1 gave us labour reporting, order completion, scrap, and a view of the materials a job needs. R2 lets operators issue materials to a specific operation, including substitutes and items that weren't on the bill of materials, capture lot and serial attributes, and correct quantities on the spot. Traceability belongs with the people who did the work, not passed through three pairs of hands until the day supervisor arrives to clean up after the night shift.

Nirav has demoed it and calls it slick: one screen with only what an operator needs. His caution is a useful one. This isn't an MES, and it won't integrate with your CNC machines. It's a sensible entry point for manufacturers who've never tracked shop floor labour, and he thinks anyone touching inventory and machines should have something like it. He just hopes Acumatica resists the urge to turn it into a screen that does fifty jobs.

I couldn't resist a daydream. A few releases from now, with the kiosk, AI and advanced planning and scheduling working together, a machine goes down halfway through a job. The operator taps "machine down". An agent blocks out the work centre, suggests work centre seven for the job in progress, reschedules the rest and tells the supervisors. The operator confirms, walks over, logs in and carries on. We'll be there by 2028 R2. Probably.

The smaller wins

Nirav's favourite is wonderfully specific. If you run Shopify through Acumatica's connector and charge a restocking fee on returns, the return has always synced but the fee hasn't, so someone has been keying it in by hand on every return order for years. A new fee setting on the e-commerce store fixes it. For US customers, 1099s can now be generated and filed electronically with the IRS from inside Acumatica, instead of printed onto the paper form, scanned and uploaded.

Then the rapid-fire round. Bank matching that learns from the matches you've confirmed. A vendor approval workflow that keeps new or changed vendors out of purchasing and AP until they're approved, which any controller will appreciate. Refunds posted to the current open period without reopening closed ones. Warehouse picking strategies by pallet, case and each, from the right zone, plus handing over a half-finished pick list without losing what's been picked. Rate shopping that sorts by cost and shows in green or red whether each service meets the customer's requested date, with address checks before the label prints. Shopify exchanges that arrive already reconciled, and BigCommerce B2B accounts that sync both ways. CRM AI that flags which opportunities and cases need attention, and parent-child projects. On the platform, mobile clock-in with offline support, an API monitoring tab showing which integrations are hammering the system and how close you are to your licence limits, and resource units renamed into plain English.

So, should you take it?

Nirav's advice: take the version, but not the first build. Wait for the next one, which puts most customers in a November-to-December window, a quieter time for many businesses anyway, and check the known-issues list in the community before you go.

I agree, and I'd go further: never take the first build. Two releases a year is a bit like the iPhone. Ask "should I buy now or wait?" and there's always another one coming, so you end up waiting forever. Look at what's in it and make a value judgement. Don't upgrade for one shiny thing; upgrade when there's enough in it to help the business use the system better. If you're on an older version, there's enough here. And thanks to the classic UI, the decision is partly being made for you.

We'll both be at Acumatica Summit 2027 in Seattle, 24 to 27 January, and the plan is to broadcast live from there. If you'd rather someone else read the release notes and plan the upgrade for you, that's Nirav's day job at AdCirrus. He still hasn't paid me to say that.

Listen to the full conversation: Acumatica 2026 R2 on The ABCs of ERP & Beyond, with Peter Nicholson and Nirav Shah.