Five studies of five different buyers, none designed to agree, land on the same rule: buyers select on proof, pay a premium only after it, and want it from companies like their own.
Every study agrees
Proof beats claims at selection
Buyers choose vendors on demonstrated results, not on how advanced the AI sounds.
5 of 5 studies
Chief financial officers
Accuracy is the top selection criterion (8.88 of 10); AI maturity is last of nine (6.67).
Accounts payable automation
AI and machine learning ranks second to last of nine selection criteria (6.8 of 10).
AI services buyers
95% want to see production deployments; only 8% are very confident in vendor claims.
Compliance software
AI that 55% of customers never see earns the vendor no credit.
AI security buyers
Buyers want measured risk reduction on their own traffic, not benchmark scores.
The premium is paid after proof
Buyers will pay more for AI, but only once it has proven itself in their own business.
5 of 5 studies, 4 measured directly
Chief financial officers
95% would pay a premium; 28% would accept 15 to 30% more.
Accounts payable automation
Comments on willingness to pay are almost all about AI that is already deployed.
AI services buyers
97% would pay 10% more for genuine AI specialization, 65% would pay 20% more and 38% would pay 30% more.
Compliance software
73% accept a price increase, most often 6 to 10%.
AI security buyersIndirect
Would budget more for coverage of AI agents (directional).
References at own size and domain
Buyers trust proof from companies like their own: the same size and the same industry.
5 of 5 studies, 3 measured directly
Chief financial officers
In their own words, they ask for references at their size and in their industry.
Accounts payable automationIndirect
Praise for industry-specific knowledge is what drives the highest scores.
AI services buyers
Domain knowledge is the most repeated gap; 35% rank it in their top five.
Compliance softwareIndirect
Regulated buyers want vendors their regulators already trust.
AI security buyers
They want controls that are audit-ready for regulated workflows.
Four of five studies agree
Layer on the system of record
Buyers want AI added to the systems they already run, not a replacement for them.
4 of 5 studies, 3 measured directly
Chief financial officers
77% prefer AI layered onto existing systems; 15% want a standalone tool.
Accounts payable automation
88% have it integrated with procurement; 38% say switching would be difficult.
Compliance software
Buyers stay on the specialist system their regulators already trust.
AI security buyersIndirect
Two thirds run on cloud defaults picked for convenience, the case that tests the pattern.
Not measured
AI services buyers
Human stays on the exception path
Buyers want routine work automated and a person deciding the exceptions.
4 of 5 studies, 3 measured directly
Chief financial officers
80% want a human in the loop; 96% say language models alone are not enough.
Accounts payable automation
89% process invoices touchless, with approval rules routing exceptions to people.
AI services buyersIndirect
Governance and model operations are named as worth paying more for.
AI security buyers
Policy-driven controls running in production are the bar for evaluation.
Not measured
Compliance software
Per-seat pricing loses to fixed or outcome
Buyers prefer a fixed fee or paying for results over paying per user.
4 of 5 studies, 3 measured directly
Chief financial officersIndirect
93% are shifting labor budget to AI, which points to pricing against the labor it replaces.
AI services buyers
76% prefer fixed or outcome-based fees; 6% want time and materials.
Compliance software
47% reject per-user pricing, the most refused model.
AI security buyers
They want a custom enterprise agreement or a flat annual fee, not usage-based billing.
Not measured
Accounts payable automation
Three of five studies agree
Spend runs ahead of proof
AI budgets are rising faster than the evidence that AI pays back.
3 of 5 studies
Chief financial officers
57% face board pressure to adopt AI; 59% expect headcount to fall.
AI services buyers
84% expect their AI spend to rise.
AI security buyers
Budgets are moving into the $5M to $15M range.
Not measured
Accounts payable automation, Compliance software
Skills gap is why they buy
Companies buy AI because they cannot hire the people to build it themselves.
3 of 5 studies, 2 measured directly
Chief financial officers
16 to 18% name the skills gap as their main barrier; 95% would buy rather than build.
Accounts payable automationIndirect
50% of the leading vendor's customers cite its implementation method.
AI services buyers
68% say hiring senior AI talent is very difficult or their top challenge.
Not measured
Compliance software, AI security buyers
Two of five studies agree
Pilots do not reach production
Most AI pilots stall before they are used in day-to-day operations.
2 of 5 studies
Chief financial officers
Only 4% say more than half their pilots succeed; 81% of piloting enterprises convert a quarter or fewer.
AI services buyers
61% move less than a quarter of their pilots into production.
Not measured
Accounts payable automation, Compliance software, AI security buyers
Shipped AI is invisible to the base
Vendors have shipped AI that most of their own customers never notice.
2 of 5 studies
Compliance software
55% see no AI in the product; only 2 of 20 had noticed an AI feature.
AI security buyers
47% have not heard of the specialist vendor; the gap is awareness, not capability.
Not measured
Chief financial officers, Accounts payable automation, AI services buyers
02
Accuracy ranks first, AI maturity last
Finance leaders are not buying AI, they are buying outputs they can trust: accuracy is their top criterion at 8.88 of 10, and AI maturity comes last of nine at 6.67.
Crossover CFO study, finance leaders, January 2026
6789 of 10Mean of nine 7.61
Accuracy and reliability of outputs
8.88
Total cost of ownership
8.34
Integration with existing stack
7.95
Vendor support quality
7.74
Implementation speed and complexity
7.72
Security certifications and compliance
7.34
Reference customers in similar industries
7.03
Product roadmap and innovation pace
6.81
AI maturity and capability
6.67
03
The proof gate
Size decides which side of the gate a company is on. Enterprises are three times as likely to be in production (37% against 12%), while half of companies under 500 employees run no pilots at all.
Crossover CFO study, finance leaders, by employee count
Under 500 employees
500 to 2,499 employees
2,500 or more employees
All sizes
Already in production
12%
10%
37%
17%
Do no pilots
50%
21%
30%
37%
Buy only, no internal AI
50%
28%
17%
36%
Board pressure, moderate or strong
53%
49%
77%
57%
04
The pilot graveyard
Running pilots is not the same as shipping AI: 32 of the 82 finance leaders who pilot have not moved a single one into production, and only 5 see half or more get there.
Crossover CFO study, finance leaders who run pilots
Under 500 employees
None reached production10
Under 25%11
25 to 50%8
50% or more1
500 to 2,499 employees
None reached production16
Under 25%9
25 to 50%2
50% or more4
2,500 plus employees
None reached production6
Under 25%11
25 to 50%4
50% or more0
05
The premium is real, and it is paid after proof
The premium is real but bounded: 97% of Fortune 1000 technology buyers would pay 10% more for genuine AI specialization, 38% would pay 30% more, and none would pay 50% more.
Crossover customer study, Fortune 1000 technology buyers, answering on price
06
Pressure precedes deployment, doubt stays with the undeployed
Board pressure is highest where AI is already live (82% against 52%), and doubt about the return concentrates where it is not: 58% of enterprises without AI in production cannot justify it.
Crossover CFO study, finance leaders, by deployment status
0%25%50%75%100%
Enterprise, in production
36%91%
+55 points
Enterprise, not in production
58%68%
+10 points
Midmarket, in production
14%71%
+57 points
Midmarket, not in production
34%51%
+17 points
Cannot justify the return on investmentall four 38%Under moderate or strong board pressureall four 61%
07
The gate by cohort
The proof bar is the same in every cohort. Industry, role, team size and data readiness change how far a buyer stands from it, not what it takes to clear it.
Finance leaders, cut by industry, role, team size, pressure, posture and data readiness
Cohort
What differs
Numbers
Industry
B2B software and technology
Furthest through the gate and the most return on investment-skeptical, because they have seen the receipt
33% in production; 50% cannot justify the return on investment; 83% layer on existing systems
financial services and fintech
Regulated and stalled: lowest production, highest doubt, lowest premium
10% in production; 70% cannot justify the return on investment; 20% would pay 15% or more
healthcare and life sciences
Most likely to skip pilots entirely; cost-driven by reimbursement pressure
45% do no pilots; 10% in production; 62% name inaccuracy
manufacturing
Nobody will replace core systems; most want a human on every decision
0% would replace core systems with AI-native; 65% prefer assistive only
professional services
Highest board pressure and highest premium acceptance, with modest production
89% under board pressure; 50% would pay 15% or more; 17% in production
Role
CFO vs vice president of finance
The operator is twice as skeptical as the principal
return on investment doubt 25% vs 48%; cost concern 38% vs 63%
Finance team size
3 to 4 vs 40 or more
The small team cannot execute under pressure; the large team can
In production 0% vs 37%; do no pilots 55% vs 22%; under board pressure 73% vs 33%
Board pressure
none vs strong
Pressure is the engine of pilots and deployment, not of conviction
Do no pilots 77% vs 39%; in production 8% vs 30%; return on investment doubt roughly flat, 31% vs 35%
Posture
buy only vs primarily buy with some internal work
The firm that cannot generate proof pays less for it
Do no pilots 61% vs 19%; would pay 15% or more 22% vs 52%; want AI inside existing systems 85% vs 76%
Data readiness
excellent
Clean data is the precondition for production, and the cohort that then stops piloting
50% in production; 75% do no pilots; 12% return on investment doubt
08
By employee count
Size changes the starting point, not the requirement: enterprises are further into production and under more board pressure, while smaller companies are further from piloting at all.
Finance leaders, every measure by size band
Measure
Midmarket, under 500
Upper midmarket, 500 to 2,499
Enterprise, 2,500 or more
Deployment
Already in production
12%
10%
37%
We don't do pilots
50%
21%
30%
No pilot budget at all
40%
28%
23%
No finance function with AI in production
58%
46%
33%
Production is 2 to 3 years out
42%
28%
23%
Buy only, no internal AI
50%
28%
17%
Of those who pilot: zero reached production
33%
52%
29%
Of those who pilot: zero or under 25% reached production
70%
81%
81%
Board pressure, moderate or strong
53%
49%
77%
Concerns
inaccuracy of model or agents
77%
62%
70%
cost
50%
46%
30%
can't justify the return on investment
32%
31%
50%
Barriers
integration complexity
22%
15%
7%
unclear return on investment or vendor immaturity
20%
23%
20%
Differentiators
higher automation accuracy
68%
64%
50%
better integrations
55%
46%
37%
all-in-one platform
30%
31%
17%
Expectations
Will pay 15 to 30% more for AI
23%
31%
33%
Finance work automated today under 5%
72%
67%
57%
Expect 11 to 50% automated in 3 years
73%
70%
73%
AI agent could become the system of record
75%
62%
70%
CFO owns the AI decision fully
63%
46%
50%
09
Doubt and pressure
Doubt follows the pilots that did not ship, not the ones that did: 38% of finance leaders without AI in production cannot justify the return, against 23% of those with it live.
By deployment status: return on investment doubt, board pressure and cost concern
Group
Cannot justify the return
Board pressure
Cost concern
Enterprise, in production
36%
91%
18%
Enterprise, not in production
58%
68%
37%
Midmarket, in production
14%
71%
14%
Midmarket, not in production
34%
51%
55%
All, in production
23%
82%
23%
All, not in production
38%
52%
49%
By pilot outcome: the doubt sits with the pilots that did not convert
Pilot outcome
Cannot justify the return
In production
Board pressure
We don't do pilots
34%
13%
45%
Zero pilots reached production
38%
12%
59%
Under 25% reached production
42%
19%
65%
25 to 50% reached production
29%
36%
79%
10
The other four studies
Four further Crossover studies, each fielded for a different sponsor and a different buyer, check the finance leader findings from the outside in.
Study
Who was surveyed
Numbers used
accounts payable automation customers, 2025, one vendor overweight
Accounts payable automation software customers
AI and machine learning ranked 6.8 of 10, second to last of nine selection factors. 89% touchless on routine invoices. 38% of the overweight vendor's customers call switching difficult vs 0% at six peers; 88% integrate accounts payable with procurement and purchase orders. 50% cite implementation methodology as a strength vs 29% across vendors. 90% interested in new modules after AP proves out.
Fortune 1000 AI services buyers, Jul 2026
US Fortune 1000 class buyers of AI and digital engineering services
84% expect spend to rise. 61% ship under 25% of AI projects to production. 95% cite production deployments as the top proof of vendor depth. 68% say senior AI hiring is very difficult or their top challenge. 76% prefer fixed or outcome pricing. Mean willingness to pay premium +21%, 65% would pay 20% or more. 8% find "traditional IT services plus AI" credible.
Compliance software customers, 2026
Regulated mid-size buyers of compliance and quality software
55% of the vendor's customers report no AI touchpoints; 2 of 20 have seen AI in product; peers' customers 28% none. 73% accept an AI uplift, most often 6 to 10%. 47% reject per-user AI pricing. 2 of 3 lapsed accounts cite a vendor with stronger AI (base of 3). Renewal intent 5.9 vs peers 6.6.
Enterprise AI security buyers, Jun 2026
Enterprise AI safety, security and trust-and-safety buyers
100% run AI in production; 90% expect agentic AI in production within 12 months. 65% report shadow AI; 20% say current controls detect it. Two thirds rely on cloud-provider built-in safety tools. Budgets cluster at $1M to $5M.
11
Objections, taken seriously
The 6 strongest objections to the argument, each answered with the data rather than around it.
“This is the diffusion curve. Large firms always adopt first.”
Partly. But the diffusion curve says late adopters wait for the technology to mature and the price to fall. These buyers are waiting for neither: 95% will pay a premium, and AI maturity ranks last when they choose. What they want is evidence from a peer. The curve treats that as one factor among five, but our data says it is the deciding one.
“Better models will make proof unnecessary.”
Models have improved every quarter for three years, yet accuracy that falls short of production is still the top complaint. What matters is how the model performs on the buyer's own data, inside its workflow, with its auditor's sign-off. A better model makes proof cheaper to produce, but it does not replace it.
“We saw this with process automation.”
It is the closest precedent: boards were enthusiastic, pilots multiplied, and few reached scale. The vendors that survived sold outcomes and published them. The difference now is that the data needed to measure results already sits in the core system, ready to be read.
“AI is already everywhere.”
Casual use is everywhere: 65% of enterprise security buyers report unapproved AI use, and only 20% can detect it. What is held back is AI that posts to the ledger or touches a customer invoice, because someone owns the mistake when it goes wrong. That is also where the software revenue is.
“The sample is small.”
It is. The enterprise group is small, and we tested eight comparisons. The argument rests on the two differences that hold up statistically, buying without building and running in production, plus four separate studies that point the same way. The 2027 repeat survey will test the rest.
“Crossover is talking its book.”
Yes. We sell verified evidence and we are arguing that the market runs on it. So we hold ourselves to the same test: our checker rejects some of our own first drafts, the rejected rows stay on record, and our own production numbers appear on the same page as our clients'.
