Mid-Market AI Returns: What 28 Public Disclosures Show

Executive Summary

AI is paying off for some mid-market public companies and replacing others. Both show up in earnings before they reach revenue.

Across 28 mid-market public companies in eight sectors, three patterns emerge:

Winners do more with the same team. They serve more customers without hiring more people. Upstart: revenue +64% with headcount +18%. BILL: AI-processed transactions +533%. Hinge Health: 47% more members at flat care-team cost.
Losers get replaced. Their customers can now do the work themselves with AI tools the vendor never sold. Chegg: -39% revenue, -67% workforce. LivePerson: guiding to -15-20% in 2026.
Beneficiaries get paid by the AI buildout. They sell into the cycle without deploying AI themselves. Comfort Systems: $11.94B backlog. Ambarella: 80% of revenue from edge-AI chips.

One insight across all three. The income statement is the lagging indicator. What a company discloses now is the leading one. This brief maps the patterns and identifies five archetypes across the cohort.

Disclosure Categories

What companies disclose about AI

Three categories show up across earnings calls and 10-K filings. None of them is "AI revenue contribution."

Partnership announcement. The company names a vendor (Microsoft, Google Cloud, Anthropic), a launch date, and an investment commitment. No operational metric attached. 32% of cohort.
Operational metric. The company reports something specific: transactions processed, members served, automation rate, time-to-completion. 32% of cohort.
Workforce metric. The company reports headcount, hiring restrictions, or SG&A as a share of revenue. 14% of cohort.

Only 11% (three companies) tied AI directly to a revenue or EBIT number. The rest disclosed strategy, vendors, or workforce. Which category a company uses tracks how far along its AI work is: companies with results report operational metrics; companies still in pilot report partnerships; companies losing to AI are forced to quantify the damage.

Exhibit 1

Disclosure splits into three categories. Most companies announce, fewer measure, the disrupted are forced to quantify.

Disclosure Category	Cohort Companies	Share	Representative Metric
Operational (volume, throughput, automation rate)	BILL, Upstart, Hinge Health, GitLab, Kinaxis, Evolent, Health In Tech, SoundHound, Innodata	32%	BILL: AI-processed transactions +533%
Workforce (headcount, SG&A, time-to-fill)	Upstart, Levi Strauss, Hinge Health, Chegg	14%	Upstart: revenue +64%, headcount +18%
Partnership (named vendor, no metric)	Levi (Microsoft), Papa John's (Google), Five Below (Invent.ai), NCR Voyix, UiPath, C3.ai, Sprouts	32%	Papa John's: rollout by end of 2026
Revenue decline attributed to AI	Chegg, LivePerson, Kforce, Upwork (partial)	14%	Chegg: -39% revenue, -67% workforce
Infrastructure backlog (paid by the buildout)	Comfort Systems, Ambarella, indie Semiconductor	11%	Comfort Systems: $11.94B backlog
No AI disclosure	Cracker Barrel, Sprouts (in name)	7%	No keyword present

Source: RBD. analysis of 28 mid-market public companies, primary disclosures (10-K, earnings calls, IR pages) FY2025–FY2026. Some companies appear in multiple categories.

Key Finding

Partnership announcements and operational metrics tie for first place in mid-market AI disclosure. Capital markets price both as comparable AI traction. They describe different things.

Output Gains

Volume up, labor flat

The companies that quantify AI impact disclose more output without more people. The metric is operational.

"In 2025, we grew originations 86% and revenues 64% while growing headcount just 18%. A ratio any business would die for." Dave Girouard, CEO · Upstart FY2025 earnings call

The same pattern across four sectors:

BILL Holdings (SMB fintech). AI-processed transactions +533% at 92% accuracy. The W-9 agent saved roughly 1,000 days of manual work since October launch. Self-service usage moved from 13% to over 40%.
Hinge Health (digital health). Served 47% more members at flat care-team cost. Average async session time dropped 28% sequentially. Engineering output per employee doubled.
GitLab Duo (DevSecOps). First-time Duo customers +35% QoQ. The Ultimate (AI-heavy) tier now makes up 52% of total ARR.
Evolent Health (healthcare AI). AI auto-approves 80% of prior authorization volume.

Exhibit 2

Across five sectors, the disclosed metric is the gap between volume and labor

Source: Upstart Q4 2025 earnings, Feb 2026. BILL Holdings Q1 FY2026, Nov 2025. Hinge Health Q4 2025, Feb 2026. Evolent Health Q3 2025, Nov 2025. GitLab Q1 FY2026, Jun 2025.

Takeaway: The operational metric appears in disclosure one to two quarters before revenue moves. Capital markets pricing on revenue catch the signal late.

Spending and Output

Spending is current. Output is deferred.

Most mid-market companies are spending heavily on AI partnerships. Few have shown operational results yet. Their disclosures describe intent, not result.

Levi Strauss → Microsoft "Super-Agent" Measurable FY2027 Enterprise-wide AI on Azure, embedded in Teams. Target: 40–60 bps SG&A improvement. Rollout begins early 2026.
Papa John's → Google Cloud Gemini Ordering Agent Measurable FY2027 First restaurant deployment of the agent. Nationwide rollout by end of 2026. Cost savings target: $60M supply chain + $25M corporate through 2027.
UiPath → Agentic Automation Measurable FY2027+ Category leader. Disclosure: "agentic automation is not expected to materially contribute to FY2026 revenue." Positioned as "FY2027+ growth engine."

Exhibit 3

Capital markets currently read partnership announcements and operational output as the same kind of evidence. They are not.

Source: Levi Strauss Q4 2025 earnings, Jan 2026. Papa John's Q4/FY2025 results, Feb 2026. UiPath Q3 FY2026 earnings, Dec 2025. Upstart Q4/FY2025 results, Feb 2026.

The substitution pattern

Some companies built their service around human labor for customers. Now AI gives those customers the ability to do the work themselves. Revenue erodes.

Chegg. Homework-help subscription collapsed. Students ask ChatGPT instead. -39% revenue, -67% workforce.
LivePerson. Legacy conversational AI platform replaced by direct LLM deployments inside customer stacks. Guiding to -15-20% revenue in 2026.
Kforce IT staffing. Clients moved IT spend to AI tooling that did not require additional hires. -5.4% revenue.

In each case, customer behavior changed before vendor revenue did. The income statement was, again, the lagging indicator.

Three Disclosure Streams

Three streams. One insight.

No single company describes the pattern across all three categories. Each discloses what its own business model produces. Side by side, they converge on one insight.

Stream 1: More Output Per Employee

9 of 28 companies · output grows faster than headcount

Upstart. Revenue +64%, headcount +18%.
BILL Holdings. AI-processed transactions +533% at 92% accuracy.
Hinge Health. Members +47% at flat care-team cost.
Evolent. 80% auto-approval of prior authorizations.
Kinaxis. 10x planner productivity claim.

Across nine companies in seven sub-sectors, the disclosed unit is output per employee. Not revenue.

Stream 2: The Substituted

4 of 28 companies · customers doing the work themselves

Chegg. -39% revenue. ChatGPT replaced the homework-help subscription.
LivePerson. Guiding to -15-20% revenue in 2026. LLM chatbots replacing the legacy platform.
Kforce IT staffing. -5.4%. Clients spending on AI tools, not staff.
Upwork low-value categories. Translation and basic writing declining. Tasks moved in-house to LLMs.

The mechanism: customers can now do what the vendor used to do for them.

Stream 3: Selling to the AI Buildout

3 of 28 companies · paid by the cycle, not deploying

Comfort Systems USA. $9.1B revenue (+30%), $11.94B backlog. HVAC and electrical for AI data centers.
Ambarella. 80% of revenue from edge-AI chips.
indie Semiconductor. $7.4B strategic backlog driven by automotive AI.
Innodata. $252M revenue (+48%), $68M in pretraining-data contracts. Federal AI unit launched at $25M.

Leading indicator is upstream: hyperscaler capex commitments ($660-690B for 2026, per Comfort Systems' CEO).

The convergence insight. Mid-market AI returns surface in disclosures one to two quarters before they reach the income statement. The three streams describe one underlying pattern: winners do more with the same team; losers get replaced by customers using AI themselves; beneficiaries get paid by the AI buildout cycle. The income statement lags. The leading indicators are output per employee (winners), customer behavior change (losers), and backlog tied to AI capex (beneficiaries). The discipline: knowing which disclosure to read.

Exhibit 4

Three streams describe the same pattern: AI returns appear before revenue or EBIT can recognize them

Stream	Disclosure Unit	Leading Indicator	Income Statement Lag
More Output Per Employee	Volume per FTE, automation rate, members per care worker	Output-per-employee gap widening	1–2 quarters before revenue acceleration
Customer Substitution	Customer behavior change, category-level volume decline	Subscriber count, GSV by category, time-to-cancel	1–2 quarters before revenue decline
AI Buildout Backlog	Backlog tied to AI buildout, tech-sector share of revenue	Hyperscaler capex commitments, data center starts	2–4 quarters before revenue (construction cycle)

Source: RBD. analysis of 28 mid-market public-company disclosures, FY2025–FY2026. Leading indicators inferred from cohort-level disclosure patterns, not from individual company guidance.

Five Archetypes

Five mid-market AI archetypes

The 28 companies sort into five archetypes based on how AI shows up in their disclosures.

Archetype 01

The Output Winners

MORE OUTPUT PER EMPLOYEE · QUANTIFIED

Companies: BILL, Upstart, Hinge Health, GitLab Duo, Kinaxis, Evolent, Health In Tech, SoundHound, Innodata.

AI lets them do more with the same team. The disclosed metric is output per employee. They had the absorption capacity to convert AI capability into capacity before deploying it.

Archetype 02

The Switzerland Layer

CROSS-PLATFORM ORCHESTRATION · ARCHITECTURAL BET

Companies: UiPath, Manhattan Associates, Domo, BILL, Kinaxis.

Positions the company as an orchestration layer across competitor AI platforms. The bet: AI models commoditize, the orchestration layer endures. Operational results show up later, in customer expansion data.

Archetype 03

The Infrastructure Beneficiaries

PAID BY THE CYCLE · NOT DEPLOYING

Companies: Comfort Systems, Ambarella, indie Semi, Innodata Federal.

Revenue tracks hyperscaler capex commitments. Leading indicator is upstream: Microsoft, Google, Amazon, Meta. Income statement responds with a 2–4 quarter construction lag.

Archetype 04

The Substituted

CUSTOMERS DO THE WORK · REVENUE FALLS

Companies: Chegg, LivePerson, Kforce (IT staffing), Upwork (low-value categories).

The company built its service around human labor customers can now do themselves with AI. Hardest hit: subscription services where the value was access to a person doing a task.

Archetype 05

The Announcers

PARTNERSHIP NAMED · METRICS DEFERRED

Companies: Levi Strauss, Papa John's, Sprouts, NCR Voyix, Five Below, C3.ai, Veritone.

Discloses partnerships, product roadmaps, and forward-looking targets. No operational results yet. The open question: do FY2027 earnings show output gains, or do capital markets reread the partnerships as the absence of traction?

Horizon

When the patterns sort

Phase 1 · Now – Q4 2026

The Quantification Demand

Analyst questions shift from "what is your AI strategy" to "what is your output per employee this quarter."

Output Winners: multiple expansion
Announcers: pressure to convert intent to result
The Substituted: accelerated subscriber decline

Phase 2 · 2027

The Compression

The Announcers either show output gains or drop AI claims.

Switzerland Layer: platform consolidation pressure
Infrastructure Beneficiaries: backlog growth moderates
Substitution wave 2: insurance brokers, mid-tier accounting, parts of consulting

Phase 3 · 2028+

The Resolution

Mid-market settles into three durable patterns:

AI-Native. Output gains built into the original model.
Operationally Integrated. Output gains achieved through retrofit.
Substituted. Business model replaced when customers do the work themselves with AI.

"Announcing" disappears as a credible disclosure category.

External Factors

Catalysts and barriers

Catalysts

Capital market scrutiny of AI disclosures

Analyst questions are migrating from strategy to operational results. Each cycle that AI-Native companies disclose output-per-employee gains, pressure on the Announcers grows.

Operational metrics becoming standard

Upstart, BILL, Hinge Health, and Evolent have normalized a format that ties AI directly to output gains. Companies that do not adopt it stand out by absence.

Hyperscaler capex visibility

2026 capex commitments are publicly tracked at $660–690B. Infrastructure Beneficiaries can disclose backlog with traceable upstream drivers.

Barriers

SEC disclosure conventions

10-K and 10-Q structures are organized around revenue and EBIT. Operational metrics live in press releases and transcripts. Filing-based analysts miss them.

Boards reward announcement velocity

Speed of partnership announcement and named-vendor selection get rewarded. The measurement discipline required for operational disclosure gets rewarded less.

The 93/7 spending inversion

Deloitte (2025): organizations allocate 93% of AI budgets to technology and 7% to people. Without investing in the people side, the technology can't produce the output gains capital markets will start asking about.

Implications

Four priorities for leadership teams

Stop measuring AI as a revenue contribution percentage

Track: output per employee, transactions per FTE, automation rate, members served at flat labor cost.
Replace: the "AI revenue contribution" reporting line with an output-per-employee ratio.
Cadence: track monthly, disclose quarterly. The income statement effect arrives one to two quarters later.

Audit the business model for substitution exposure

The four companies losing revenue to AI share one feature: their service was built around human labor a customer can now do with AI.

Ask: what share of what we deliver is human labor a customer could do themselves with ChatGPT?
High exposure: pivot toward platform access, asset access, or judgment work.
Low exposure: the business model has a natural floor.

Treat the Switzerland bet as an architectural commitment

Positioning as a cross-platform orchestration layer (UiPath, Manhattan, Domo, BILL) requires the kind of architectural independence most mid-market firms haven't built.

The bet: AI models commoditize, the orchestration layer endures.
The cost: integration depth, partner-platform discipline.
The middle path fails: partial orchestration on platform dependencies collapses when the platforms consolidate. Choose deliberately or compete on price.

Read AI investment as a workforce-design decision first

The disclosure pattern is the tell.

Tech-stack spenders disclose partnerships and platforms.
Workforce-design spenders disclose output per employee.
The 93/7 spending inversion explains why most mid-market AI investments will not generate the metrics the winners already report. Without investing in the people side, the technology cannot convert into operational output gains.

Decision Support

Mid-market AI disclosure diagnostic

Eight questions across the Four Capability Bands from The Intelligence Organization™: Right-Fit Technology, People & Purpose, Operational Integration, Adaptive Governance.

Exhibit 5

Diagnostic: which mid-market AI archetype describes your current disclosure pattern

Capability Band	Assessment Question	Score 1–5	Archetype Implication
Band 1: Right-Fit Technology	How concentrated is the AI portfolio? Three to four use cases with operational metrics, or six or more partnerships without?	___	Score <3: Announcer archetype. Concentrate to three to four use cases with output metrics before adding capacity.
Band 1: Right-Fit Technology	What share of AI capex is tied to long-lived infrastructure that cannot be redeployed if a use case fails?	___	Score <3: the technology investment is trapped by design. The disclosure cannot pivot to operational metrics.
Band 2: People & Purpose	What is the output-per-employee trajectory across operations AI has touched? Widening, flat, or narrowing?	___	Score <3: absorption capacity is missing. The AI exists but is not being converted into output.
Band 2: People & Purpose	What share of what you deliver is human labor a customer could do themselves with general-purpose AI?	___	Score >3 (high exposure): Substituted archetype likely within 2–4 quarters. Pivot the value proposition.
Band 3: Operational Integration	Have core workflows been redesigned around AI, or has AI been added to existing workflows?	___	Score <3: Pilot Factory pattern. Operational metrics will not scale because the processes were not redesigned.
Band 3: Operational Integration	When one workflow is redesigned, are the upstream and downstream workflows redesigned together?	___	Score <3: output gains in one workflow create handoff friction elsewhere.
Band 4: Adaptive Governance	Does the earnings deck or board reporting disclose operational output metrics tied to AI, or only partnership announcements?	___	Score <3: the disclosure architecture itself signals Announcer. Add the metrics before analyst questions force them.
Band 4: Adaptive Governance	Are AI output metrics defined as governed KPIs with owners, measurement methodology, and quarterly accountability?	___	Score <3: metrics may exist operationally but will not survive board scrutiny. Disclosure without governance produces revisions.

Source: RBD. analysis. Framework aligned with The Intelligence Organization, Four Capability Bands (Starkey, 2026).

Score interpretation

32–40 · Output Winner. Disclosure aligns with cohort winners. Sustain the output metric as a quarterly disclosure.
20–31 · Mixed. Some operational metrics, partnership disclosure dominant. Address the lowest-scoring Band before adding new AI commitments.
8–19 · Announcer, Pilot Factory, or Substituted (depending on Band 2 Q2). More partnership announcements will not move the disclosure. Start with the output metric.

How to apply it

AI-Native, already disclosing output gains. Hygiene check. Focus Band 4 (Governance) to make sure the metrics survive board scrutiny.
Announcer, named partnerships, no operational metric. Score Band 2 Q1 (output per employee) at the most granular level you can support. That score is the starting point for the metric your disclosure pattern needs.
High substitution exposure (Band 2 Q2 > 3). The diagnostic is a deadline. Reallocate revenue concentration toward segments not dependent on bundled labor before the substitution disclosure begins.
Infrastructure Beneficiary. Diagnostic largely irrelevant. Track upstream: hyperscaler capex + customer concentration risk.

Decision support aligned with The Intelligence Organization · Band 2 (People & Purpose) as the absorption-capacity gate · Operational disclosure as the leading indicator of AI returns

Sources

References

Primary Sources (Cohort Company Disclosures): BILL Holdings, "BILL Launches New AI Agents to Power Touchless Transactions for the Fortune 5 Million," BusinessWire, Oct 28 2025. BILL Holdings, Q1 FY2026 Results, BusinessWire, Nov 6 2025. Upstart, "Fourth Quarter and Full Year 2025 Results," BusinessWire, Feb 10 2026. Upstart, "Projects 35% CAGR Through 2028," Seeking Alpha, Feb 2026. Hinge Health, Q4 2025 Earnings Transcript, Motley Fool, Feb 10 2026. Evolent Health, Q3 2025 Results, PR Newswire, Nov 6 2025. GitLab, Q3 FY2026 Earnings Transcript, Seeking Alpha, Dec 2025. GitLab, Q1 FY2026 Results, BusinessWire, Jun 10 2025. Kinaxis, Q4 2025 Results, Globe and Mail, 2026. Manhattan Associates, Q4 2025 Earnings Transcript, Motley Fool, Jan 27 2026. SoundHound AI, "Record Annual Revenue $169M," IR Page, Feb 2026. Innodata, Q4/FY2025 Results, IR Page, 2026. Health In Tech, Q4/FY2025 Results, PR Newswire, 2026.

Aspirational Disclosure Cohort: Levi Strauss & Co., "Partners with Microsoft to Build Next-Gen Super-Agent," IR Page, 2025. Levi Strauss, Q4 2025 Earnings Transcript, Motley Fool, Jan 28 2026. Papa John's and Google Cloud, "Reimagine the Future of Food Ordering," PR Newswire, Jan 11 2026. Papa John's, Q4/FY2025 Results, BusinessWire, Feb 26 2026. UiPath, Q3 FY2026 Earnings Transcript, Seeking Alpha, Dec 2025. UiPath, "Outlines Q4 Revenue Target as Agentic Automation Momentum Accelerates," Seeking Alpha, 2025. C3.ai, Q2 FY2026 Results, BusinessWire, Dec 3 2025. Five Below, "Taps Invent.ai Platform," Retail Tech Innovation Hub, Jun 12 2025. NCR Voyix, Q4 2025 Earnings Transcript, Motley Fool, Feb 26 2026. Veritone, "Reports Q4 2025 Results and Announces Strategic Deal with Oracle," BusinessWire, Mar 2026.

Substituted Cohort: Chegg, "Reports 2025 Fourth Quarter and Full Year Results," IR Page, 2026. Chegg, "To Remain a Standalone Public Company," BusinessWire, Oct 26 2025. Chegg Workforce Reduction Coverage, CNBC, Oct 27 2025. LivePerson, "Announces Fourth Quarter 2025 Financial Results," PR Newswire, 2026. Kforce, Q4 2025 Earnings Transcript, Motley Fool, Feb 2 2026. Upwork, Q4/FY2025 Results, IR Page, 2026.

Infrastructure Beneficiary Cohort: Comfort Systems USA, Q4/FY2025 Results, BusinessWire, Feb 19 2026. Comfort Systems USA, Q4 2025 Earnings Transcript, Seeking Alpha. Ambarella, Q3 FY2026 Earnings Transcript, Seeking Alpha. Ambarella, "Raises FY2026 Revenue Growth Target as Edge AI Accelerates," Seeking Alpha. indie Semiconductor, Q4 2025 Results, BusinessWire, Feb 19 2026. Innodata, "Anticipates 35% Revenue Growth in 2026," Seeking Alpha.

SEC Filings (10-K, 10-Q): C3.ai 10-K, FY ending Oct 2024, SEC EDGAR. Ambarella 10-K, SEC EDGAR. GitLab 10-K, SEC EDGAR. EDGAR full-text search for "artificial intelligence" in 10-K filings, 2025–2026 reporting periods.

Comparative Industry Research: McKinsey, "The State of AI," Mar 2025 (1,933 respondents). BCG, "From Potential to Profit: Closing the AI Impact Gap," AI Radar, Jan 2025. Deloitte, "The State of AI in the Enterprise, 2026" (3,235 leaders). Gartner, "Worldwide AI Spending Will Total $2.5 Trillion in 2026," Jan 2026. Deloitte CTO Bill Briggs, Fortune, Dec 2025 (93/7 spending ratio).

RBD. Research: Starkey, M.C., The Intelligence Organization, 2026. RBD., "Enterprise AI Investment 2026 Outlook: From Technology-First Budgets to Capability-First Returns," RB-Q2 2026. RBD., "3 Foundational Decisions: What Separates the 6% Generating EBIT from AI," SI-Q4 2025. RBD., "Enterprise AI Tool Strategy: The Copilot Adoption Gap," SI-Q1 2026. RBD. cross-industry mid-market AI disclosure synthesis, 2026.