Corrected 2026-05-31: cluster #2 below collapsed two distinct PwC announcements (May 5 PwC-OpenAI, May 14 PwC-Anthropic) into one and missed the named operational metrics published on May 14. The corrected, asymmetric verdict is in the follow-up piece: PwC’s Second Announcement Had the Metrics, My First Verdict Missed Them. The rest of the baseline below stands as the May 30 snapshot.

Vendor messaging was loud in W21-W22. The buyer record was not.

In the last two weeks, four announcement clusters dominated AI feeds: KPMG-Anthropic on May 19, PwC-Advocate Health on May 14, Opus 4.8 with Databricks framing on May 28, and the continued DeepSeek release wave. If you only tracked announcement volume, this looked like a broad value-capture moment.

If you track named, attributable buyer evidence, the picture is narrower.

This baseline uses one discipline and does not move the goalposts: for each announcement, did a named buyer go on the record about value captured, what exactly did they say, and where are they silent? Silence is data.

1) KPMG-Anthropic (May 19)

Vendor narrative: workforce-scale rollout and responsible AI operating posture.

Named on-record voices: Bill Thomas (KPMG), Rema Serafi (KPMG), Ethan Burris (UT Austin McCombs, included in announcement context).

What was said:

• Rema Serafi made the most concrete workflow claim in the published material, describing a tax-regulation agent build that “used to take weeks now takes minutes” inside Digital Gateway.
• Bill Thomas and the broader KPMG framing emphasized scale, governance, and responsible deployment intent.
• Ethan Burris appeared as a named external voice in the launch context.

Buyer evidence depth:

• Value-adjacent signal exists: weeks-to-minutes is operationally meaningful.
• Named buyer outcome metrics are still thin: no disclosed baseline-to-post customer metric, no published cost-to-serve delta, no customer-facing quality metric.

Verdict: lukewarm.

Why: this is stronger than generic AI partnership language, but it remains mostly operator-side narrative plus one strong internal workflow claim, not broad buyer-side value capture evidence.

2) PwC-Advocate Health (May 14)

Vendor narrative: scaled deployment, training expansion, and cross-function transformation claims.

Named on-record voices: Paul Griggs (PwC), Andy Crowder (Advocate Health).

What was said:

• Andy Crowder was on record in the announcement as the named client voice.
• The public framing emphasized strategic intent and expected impact from expanded AI deployment.

Buyer evidence depth:

• Positive signal: a named client appears in the public record, which is already better than many enterprise releases.
• Core gap: outcome disclosure is limited. The record in this window does not provide a named buyer baseline-to-post metric set for cost-to-serve, throughput, or customer-felt functionality.

Verdict: lukewarm.

Why: named buyer participation is present, but this is still intent-forward rather than value-capture-forward in public evidence.

3) Opus 4.8 with Databricks framing (May 28)

Vendor narrative: better model performance and stronger economics.

Named on-record voice: Hanlin Tang (Databricks CTO).

What was said:

• Hanlin Tang stated Opus 4.8 in Genie ran at “61% cheaper token cost than Opus 4.7.”

Buyer evidence depth:

• This is a clear economics signal.
• It is not yet a named buyer value-capture disclosure. Token efficiency is an input to value, not proof of customer outcome movement.
• No named end-buyer in this window publicly disclosed post-deployment customer metrics tied to that cost improvement.

Verdict: lukewarm.

Why: credible technical-economics evidence is present, but named buyer outcome evidence is still limited.

4) DeepSeek release wave (W21-W22 context)

Vendor narrative: capability momentum and high visibility in model-release discussion.

Named on-record buyer voice in this window: silent.

What was said by named buyers:

• Silent.

Buyer evidence depth:

• Significant announcement and discourse volume.
• Near-zero named buyer evidence in the last 60 days in this vetted set showing value captured in production customer operations.

Verdict: skip.

Why: this is the strongest contrast case. The market signal is loud, but named buyer evidence is sparse. Treating this as proven enterprise value today would be vendor-amplification behavior.

What The Contrast Actually Shows

Across all four clusters, the same pattern repeats:

• Announcement strength is high.
• Named practitioner attribution varies.
• Named buyer value-capture evidence is mostly partial, with DeepSeek close to absent in this source set.

This does not mean the underlying work is weak. It means the public buyer record has not caught up to the scale of supplier claims.

That distinction matters because decision quality fails at exactly this boundary. Teams treat deployment intent as delivered value, then discover six months later that no customer-facing lever moved in a measurable way.

Monday morning rule for CX executives: no expansion decision without one named baseline-to-post metric mapped to one lever - cost-to-serve, capacity reallocation, or product functionality.

The practical rule for operators and buyers remains straightforward:

• If the announcement gives named buyer baseline-to-post metrics, treat as compelling.
• If the announcement gives named participants but mostly intent and setup language, treat as lukewarm.
• If named buyers are silent, treat as skip until public evidence appears.

W21-W22 lands as one lukewarm cluster repeated three times, plus one skip case that clarifies the standard. DeepSeek is not the only example of announcement-evidence divergence, but it is the cleanest one in this two-week window.

If this baseline is wrong, the correction path is simple and public: named buyers publish attributable metrics, and the verdict changes.

If no named buyer metric is published, treat the claim as a communication signal only, not an operating signal for budget or rollout scope.

Until then, silence is not neutral. Silence is a negative evidence signal.