CCH Axcess Expert AI [What It Changes]

TL;DR

• CCH Axcess Expert AI is Wolters Kluwer's intelligent layer on CCH Axcess; the June 29, 2026 Scan release is the document-ingestion piece that reads W-2s, 1099s, and variable Schedule K-1s, then sends structured fields into CCH Axcess Tax.

• The constraint that broke is template-only OCR: K-1s and supporting statements do not share one layout, so a scanner that only matches a form image still leaves a person typing boxes.

• Wolters Kluwer, as carried independently by CPA Practice Advisor, says early users see 30 to 70 percent efficiency gains on document-heavy work, and that a 10 to 15 minute K-1 can drop 60 to 80 percent. Those cuts are the vendor's.

• Scan is generally available to firms already on, or moving to, CCH Axcess Tax. It does not file the return, replace reviewer sign-off, or prove your firm's minutes.

Why a two-truck shop should still care

CCH Axcess Expert AI sounds like an enterprise tax-platform announcement. It is. The operational story is smaller than that. A two-truck HVAC company does not buy Scan. It still produces W-2s, 1099s, and, if it has partners or a pass-through, Schedule K-1s. A 10-person marketing agency that pays contractors already lives on Form 1099-NEC. A solo clinic that is an S corporation or partnership already hands its CPA a stack of source documents every spring. The person who used to retype those pages is the same person who now has fewer staff hours and more K-1s.

If your firm is on CCH Axcess Tax, the change is a new intake step: drop the PDF, review the extracted fields, accept or edit, then let the tax engine compute. If you are not on Axcess, the change is still a buying question for your CPA. Ask whether source documents enter the return as structured data or as a person with a keyboard. That is the whole product.

As of June 29, 2026, Wolters Kluwer made AI-powered document intelligence generally available in CCH Axcess Scan as part of CCH Axcess Expert AI. CPA Practice Advisor independently carried the announcement the same day.

What actually shipped

Cathy Rowe, Executive Vice President and Segment Leader of the U.S. Professional Market at Wolters Kluwer Tax & Accounting, framed Scan as the latest proof of a low-touch tax-prep path: Expert AI sits inside daily work so staff spend more time on judgment and client service. Her expert page is the named speaker on the release. The product claim is not a new tax engine. It is ingestion.

The mechanism is plain. A source document arrives. Scan classifies it. Extraction pulls fields. Structured output lands in CCH Axcess Tax. A person reviews before the return is a return. That is the same sequence firms already run with organizers and OCR, with one change: Wolters Kluwer is using Microsoft Azure Content Understanding rather than template-only OCR, so a K-1 that does not match last year's partnership PDF can still be read as a K-1.

The vendor news page is Wolters Kluwer's June 29 release. Treat it as first-party copy. Treat CPA Practice Advisor's write-up as the independent carrier of the same figures.

Sources: CPA Practice Advisor; Wolters Kluwer news.

According to CPA Practice Advisor, early users reported 30-70% efficiency gains on document-intensive workflows. That sentence is Wolters Kluwer's early-user line, republished. It is not a time-and-motion study of your staff.

How a K-1 moves, without the marketing nouns

A Schedule K-1 is not a W-2. Form W-2 is a wage statement with a relatively stable box layout. Form 1099-MISC and 1099-NEC are information returns with named boxes. A partnership K-1 is a partner's share of Form 1065 items, plus footnotes and statements that do not share one template across issuers. The 2025 Partner's Instructions for Schedule K-1 (Form 1065) are the IRS page a reviewer still has to respect after extraction.

That is why template OCR stalls. The form family is known. The instance is not. Azure Content Understanding's job, in Microsoft's own words, is to turn unstructured documents into a user-defined structured output with confidence scores and grounding, so a reviewer can see where a value came from. Wolters Kluwer's job is to map that output into CCH Axcess Tax fields a preparer already knows.

The Scan step does not decide tax treatment. It proposes values. The tax engine still computes. The reviewer still accepts, edits, or rejects. If your current process is "open PDF, type box 1, type box 2, hunt the footnote," Scan is trying to collapse the typing. If your current process is "the K-1 footnote changes basis and we argue about it," Scan does not remove that argument.

US Tech Automations fits after extraction, not instead of it: a workflow can receive the Scan output, hold unmatched or low-confidence fields in a review queue, and only then write accepted values toward the return. The tax rail stays Wolters Kluwer.

Vendor minutes, shown as a range

According to CPA Practice Advisor, one K-1 can take 10-15 minutes to key by hand, and Expert AI-powered extraction through Scan can reduce that effort by up to 60 to 80 percent. Expert AI can cut that K-1 work 60-80%, in Wolters Kluwer's telling, not in an independent lab.

Those two ranges are the only time figures in the announcement. They are enough to do honest arithmetic. They are not enough to promise a season.

USTA analysis: remaining minutes on the vendor's own range

Inputs, all from the CPA Practice Advisor republish of Wolters Kluwer:

• Manual keying baseline: 10 to 15 minutes per K-1.

• Claimed reduction: 60 to 80 percent.

Remaining time = baseline × (1 − cut).

USTA analysis. Inputs: CPA Practice Advisor. Remaining = baseline × (1 − vendor cut). This is not a customer result.

On the vendor's own band, remaining keying after Scan is 2 to 6 minutes per K-1, and saved time is 6 to 12 minutes per K-1. For 100 K-1s, saved time is 600 to 1,200 minutes, which is 10 to 20 hours, if and only if the vendor percentages hold on your documents. That "if" is the whole evaluation. A footnote-heavy oil-and-gas K-1 is not a W-2.

Do not annualize that 10 to 20 hours into a hiring plan. Do time 20 real K-1s: 10 with Scan plus review, 10 with the old keying path, same reviewer, same complexity mix. Keep the original PDF next to the extracted fields.

What Azure is actually contributing

Wolters Kluwer named Microsoft Azure Content Understanding, not "OCR plus a chat box." According to Microsoft Learn, Content Understanding processes 4 content types (documents, images, videos, and audio) into a schema you define, with confidence scores and source grounding so a human can check the span that produced a field. The same overview lists tax automation as an industry example: generate a unified view from various documents and create comprehensive tax returns. That is Microsoft's example, not Wolters Kluwer's customer proof.

The prebuilt analyzer catalog is the other useful Microsoft page. It documents domain-specific tax analyzers, including US W-2 and a set of 1099 variants, and, in the 2026-06-01-preview wave, Schedule K-1 prebuilts for Forms 1041, 1120-S, 1065, and 8865. According to Microsoft Learn, that preview wave lists 4 Schedule K-1 prebuilt analyzers. Wolters Kluwer did not say Scan is a raw prebuilt-tax.us.1065ScheduleK1 call. It said it is combining tax expertise with Content Understanding to move Scan past template-based processing. Keep those layers separate.

Sources: Content Understanding overview; prebuilt analyzers; service limits; pricing explainer.

Microsoft's GA versus preview table recommends API version 2025-11-01 for production workloads and 2026-06-01-preview for new capabilities. A firm should ask Wolters Kluwer which Content Understanding version Scan calls, whether confidence scores surface in the Axcess reviewer screen, and whether grounding (the highlight on the PDF) is visible to the preparer. If those answers are "not exposed," you are buying a black box with a vendor time claim.

Microsoft also publishes a Transparency note, a Code of Conduct, and a privacy statement. Those pages do not make Scan compliant for your client files. They are the vendor's own responsible-AI and privacy pointers. Your BAA, retention, and permitted-use questions still go to Wolters Kluwer and to your own policy.

What the IRS pages still require

Scan does not replace the form. The IRS Form 1065 page is still the partnership return. The K-1 partner instructions still tell the partner what each box means. Form W-2 is still the wage statement. Form 1099-NEC is still nonemployee compensation. Form 1099-MISC is still miscellaneous information. According to the IRS, Form 1099-NEC is 1 information return for nonemployee compensation, not a K-1 substitute.

If you e-file, the e-file for tax professionals page is still the IRS door, not Scan. The AICPA Tax Section is still the professional-standards neighborhood for reviewers who have to sign. None of those pages endorse CCH Axcess Expert AI. They are the documents the extracted fields have to land on.

A reviewer who cannot open the original PDF next to the extracted box is not reviewing. A firm that lets Scan populate, then files without a named person accepting the K-1, has changed the control, not just the keystrokes.

Honest limits

Three limits are in the source material, not in a forecast.

First, the time figures are Wolters Kluwer's, including the 30 to 70 percent early-user band and the 60 to 80 percent K-1 cut. CPA Practice Advisor republished them. No independent sample size, document mix, or review-time offset is in the public copy.

Second, availability is scoped. Scan is generally available to firms on, or moving to, CCH Axcess Tax. It is not described as a standalone OCR product for Drake, UltraTax, or a spreadsheet. The CCH Axcess platform page is the suite context; Scan is a module inside it.

Third, Expert AI is a platform layer. Scan is one capability next to workflow, tax preparation, and advisory. Buying "AI" as a slogan does not tell you which module is on, which documents are in scope, or who owns rejects.

Price is unpublished in the announcement. Ask Wolters Kluwer. Do not invent a seat cost.

Where practice management still sits

Scan solves ingestion into Axcess Tax. It does not collect the document from the client, rename the file, chase the missing K-1, or deliver the finished return. Those jobs still live in practice management. If you are comparing that layer, start with the live pages on Canopy alternatives for accounting firms and accounting practice management software. Reporting tools are a different decision; see Fathom vs Jirav.

US Tech Automations can sit between Scan and the reviewer: trigger on a new source-document batch, route low-confidence K-1 fields to a named person, and keep an accept/edit/reject record. It should not be configured to file.

A useful first week is narrower than a platform tour. Pick one document family (K-1s from one partnership software vendor, or W-2s only). Turn Scan on for that family. Require the original PDF in the same review screen. Count false accepts, false rejects, and minutes through review. Only then add 1099s.

Write the sample protocol on one page so a partner can sign it. Ten documents from issuer A, ten from issuer B, one planted error. For each document record: time to first extraction, fields accepted without edit, fields edited, fields rejected, whether the original PDF was visible, and whether the reviewer would have filed. Do not average those rows into a single "percent accurate" until you have all 21. A W-2 batch will look better than a K-1 batch. That is expected. Publish the split internally so a later sales slide cannot hide it.

If Scan cannot show the source span, treat every accepted field as an ungrounded guess. Microsoft documents grounding and confidence on Content Understanding. If Wolters Kluwer does not surface those to the preparer, your control is the human who opens the PDF, not the model. Budget the human. The what's new and document overview pages are the Microsoft side of that question. They will not answer Wolters Kluwer's SKU or price.

Staffing math comes last. The USTA analysis band is 10 to 20 hours per 100 K-1s if the vendor cut holds. A 400-K-1 season is 40 to 80 hours of keying, not a headcount. Review time can eat the savings. A partner who files 400 K-1s still needs a named accepter on each one. That is not optional because the IRS partnership and e-file pages did not change on June 29.

The Journal of Accountancy and Thomson Reuters Tax are not sources for this Scan release. They are the neighborhood a reviewer already reads. Do not paste their unrelated benchmarks next to Wolters Kluwer's 60-80%. If a later independent study of Scan appears, add it. Until then, your 21-document log is the only number that belongs in a partner meeting.

Signal vs Speculation

Sourced signal, as of June 29, 2026: Wolters Kluwer made AI document intelligence generally available in CCH Axcess Scan under CCH Axcess Expert AI. Scan is described as reading W-2s, 1099s, and variable K-1s plus supporting statements, sending structured output into CCH Axcess Tax, and using Microsoft Azure Content Understanding instead of template-only OCR. Independent coverage is CPA Practice Advisor. Vendor figures are 30-70% early-user efficiency on document-heavy work and a 60-80% cut on a 10-15 minute K-1. Microsoft documents Content Understanding, confidence scores, grounding, and tax prebuilts, including K-1 analyzers in a preview catalog, on Learn.

Our read: if Scan exposes confidence and the source highlight, and if a firm times its own K-1 mix, this can become the default first pass on document-heavy Axcess work over the next 12 to 36 months. That is a conditional forecast. It does not predict that every K-1 footnote extracts cleanly, that review time falls by the same percentage as keying time, or that firms off Axcess Tax get an equivalent product.

Our read: teams already routing source documents through US Tech Automations workflows will treat Scan as a model swap on the extraction step, not a reason to rebuild the review queue. The hard part remains: unmatched fields, missing statements, and a person who will not file a guessed box 20.

Key Takeaways

• CCH Axcess Expert AI is the platform layer; Scan is the June 29, 2026 GA ingestion module into CCH Axcess Tax.

• The new mechanism is Azure Content Understanding plus Wolters Kluwer tax mapping, not a new filing engine.

• Vendor time claims are 30-70% on document-heavy work and 60-80% on a 10-15 minute K-1. Label them as Wolters Kluwer's.

• USTA analysis of those inputs: remaining keying 2-6 minutes per K-1, saved 6-12 minutes, if the percentages hold on your documents.

• Review, original PDF, and a named accepter are still the control. Scan does not e-file.

Frequently Asked Questions

What is CCH Axcess Expert AI?

CCH Axcess Expert AI is Wolters Kluwer's intelligent layer across CCH Axcess. The June 29, 2026 Scan release is the document-intelligence piece that extracts source-document fields and sends them into CCH Axcess Tax.

Does Scan replace a tax reviewer?

No. Wolters Kluwer describes structured output into CCH Axcess Tax and a low-touch prep path. A reviewer still has to accept, edit, or reject extracted fields before a return is a return.

Are the 60-80% K-1 savings independent?

No. CPA Practice Advisor republished Wolters Kluwer's figures. Time 20 of your own K-1s before you staff to those percentages.

Which documents are in scope?

The announcement names W-2s, 1099s, and highly variable Schedule K-1s and supporting statements. Confirm your document mix, including state K-1s and footnotes, with Wolters Kluwer.

Do you have to be on CCH Axcess Tax?

Yes, as described: Scan is generally available to firms on, or moving to, CCH Axcess Tax. It is not documented as a standalone extractor for other tax engines.

Next step on the Scan queue

Do not start with "turn on AI." Start with one K-1 issuer, a reviewer, the original PDF, and a reject bucket. When that path needs a visible trigger, exception queue, and approver, map an agentic workflow around the Scan review gate before you widen the document types.

The useful test is boring on purpose. Ten K-1s from the same partnership software. Ten from a second issuer with extra statements. One known error planted in a PDF. If Scan misses the planted error, or if the reviewer cannot see the source span, you do not have a production control yet. You have a demo.

Keep a claim ledger with three columns: what Wolters Kluwer published, what your 20-document sample showed, and what the partner is willing to sign. When a new percentage appears in a sales deck, attach the document family, the sample size, and whether review time was included. That habit is worth more than the 60-80% headline.

If Scan is not on your stack, the still-valid question for the HVAC shop, the agency, and the clinic is whether their CPA's intake is structured or typed. Ask to see one redacted K-1 next to the return fields. If the answer is a person rekeying box by box, you now know what CCH Axcess Expert AI is trying to retire, and what it is not allowed to retire: the signature.