How CFOs Should Evaluate Sales AI: The Questions Nobody Asks

The sales team runs the demo. The vendor delivers the pricing deck. A per-seat number appears in the proposal, and by the time the request for approval reaches finance, the complexity has been compressed. The sales leader is excited. The ROI deck shows a payback period. The per-seat cost sounds reasonable next to what you already pay for Salesforce.

What finance typically does not see: the consumption-based AI pricing model buried in the addendum. The integration development cost that will appear as a professional services engagement in Q2. The data synchronization maintenance that will require a dedicated RevOps hire. The shelfware risk — the statistical probability that 40% of licenses will be underutilized within 18 months while the contract runs for three years.

CFOs are the last line of defense against bad software purchases. The problem is that most CFO sales AI evaluation ROI frameworks are built on the visible cost structure, which vendors have optimized for concealment. The real cost lives in the footnotes, the supplemental terms, and the decisions that don't get made because data lives in 26 different systems.

How the Cost Gets Hidden

Sales technology vendors have converged on a consistent pricing strategy: lead with a per-seat number that passes the smell test, bury consumption and add-on costs in supplemental terms, and let the buyer's optimism do the rest. A $150/user/month Salesforce Enterprise price looks reasonable until you build the full stack: AI features, Marketing Cloud, Service Cloud, Revenue Cloud, integration costs, Salesforce admin labor, and the inevitable professional services engagement for customization.

The same dynamic plays out across the vendor ecosystem. Gong is $350/user/month — but it only sees calls. You still need a CRM, a sequencing tool, a forecasting tool, and enrichment. Clari is $200/user/month — but it only does forecasting, and it needs clean CRM data to produce clean forecasts. ZoomInfo is $300/user/month for enrichment that doesn't automatically trigger your sequences. Each vendor solves one problem, prices for one problem, and integrates with everything else at your expense.

The total when you add them up — CRM, sequences, call intelligence, forecasting, enrichment, e-signatures, proposals, scheduling, support ticketing — approaches $3,350 per user per month. On a 50-person sales team, that is $167,500 per month before professional services, before training, before the integration maintenance that compounds each time a vendor releases a breaking API change.

The Total Cost of Ownership Calculation

A proper sales technology total cost of ownership calculation for a fragmented 50-person sales stack has five components that rarely appear in the same budget conversation:

License cost: The per-seat fees across all tools. For 50 reps across a representative stack — Salesforce Enterprise at $150/user, Gong at $350/user, Outreach at $150/user, Clari at $200/user, ZoomInfo at $300/user, plus scheduling, e-signature, proposals, and support ticketing — you are at $1,200 to $1,500/user/month in core licenses before AI add-ons. Monthly: $60,000 to $75,000. Annually: $720,000 to $900,000.

Integration development cost: Most sales stacks require custom integration work to pass data between systems. CRM to sequencing tool. Call intelligence to CRM. Enrichment to CRM. Each integration is a project — scoped, built, tested, and deployed. A modest stack with 8 integrations at 40 hours each at $150/hour engineering time is $48,000 upfront. That is before ongoing maintenance when vendors change their APIs, which happens multiple times per year.

Integration maintenance: Each integration is a liability. It breaks when vendors release new versions. It drifts when data schemas change. It produces silent failures — data that looks synced but isn't — that corrupt your pipeline data and your forecasting models. Budget 20% of initial integration cost annually for maintenance. On a $48,000 initial build, that is $9,600/year, compounding as the stack grows.

Training and onboarding: A 50-person sales team cycling at typical B2B attrition (20–30% annually) means 10–15 new reps per year who need to be onboarded onto 8–10 different tools. Each tool has its own certification, its own workflow, its own quirks. Budget conservatively at 4 hours per tool, 8 tools, 12 new reps: 384 hours annually of onboarding time, at fully-loaded rep cost of $75/hour: $28,800/year, not counting the productivity drag while new reps are learning the stack.

Opportunity cost of fragmented data: The hardest cost to quantify but often the largest. When your call intelligence, your CRM, your sequencing tool, and your forecasting tool don't share a data model, you make decisions with incomplete information. A deal risk that is visible in call sentiment data does not propagate to the pipeline review unless someone manually bridges it. Forecast accuracy suffers. Win rate analysis is limited to whatever data ended up in the CRM, which is a fraction of what actually happened. This is not a soft cost — it is a measurable gap in decision quality.

The Per-Conversation Pricing Trap

The most acute version of the hidden cost problem in current AI budget evaluation is per-conversation AI pricing — specifically, Salesforce Einstein at $2 per AI conversation.

The number sounds reasonable. Two dollars. Less than a cup of coffee. It becomes less reasonable when you run the adoption math. A 50-rep team. Each rep does 20 AI interactions per day — querying deal status, generating emails, asking for coaching suggestions, summarizing meeting notes. That is not an aggressive usage assumption; it is what happens when AI is genuinely integrated into daily workflow.

50 reps × 20 interactions × $2 = $2,000 per day. $40,000 per month. $480,000 per year. On top of existing Salesforce seat licenses.

Scale to 100 reps and $960,000 per year in AI fees alone becomes the realistic outcome. This is not a theoretical maximum — it is what happens when AI adoption succeeds. The better the AI, the more reps use it, and the higher the bill climbs. The vendor's revenue is directly correlated with usage, not with outcomes. The full breakdown of what Salesforce Einstein actually costs at different team sizes makes this relationship explicit.

Flat-Rate vs. Consumption Pricing: What Each Signals

The choice between flat-rate and consumption-based pricing is not just a financial preference — it reveals the vendor's incentive structure, which reveals how the product will be positioned over time.

With consumption pricing, the vendor profits when AI interactions are frequent. The vendor's incentive is to make each interaction feel valuable, to encourage usage, to surface AI suggestions in every workflow. The customer's incentive, after the first invoice shock, is to ration AI interactions — to think carefully before invoking the AI, to discourage reps from using it casually. This is the adoption-limiting consequence of consumption pricing: it trains cost-conscious organizations to use the AI less, which means the AI improves less, which means it delivers less value, which compounds into a bad investment narrative.

With flat-rate pricing, the vendor profits when customers renew. Customers renew when the product is valuable. The product is valuable when it is used habitually. The vendor's incentive is therefore to maximize AI adoption — to remove friction from every AI touchpoint, to encourage reps to invoke the AI for everything, to make AI the default rather than a considered choice. Flat-rate pricing aligns vendor and customer incentives in a way consumption pricing structurally cannot.

When a vendor pitches per-conversation pricing as "pay for what you use, not for what you don't," the correct response is: "At what usage level does our bill exceed a flat-rate alternative, and what happens to our costs when adoption succeeds?" If the vendor can't answer that question precisely, the pricing model is not designed for transparency.

Seven Questions CFOs Should Ask Before Approving Any Sales AI Purchase

These questions are designed to surface the cost structure and risk profile that vendors are not volunteering:

1. What is the true all-in cost at our usage level, including AI interaction fees? Ask for a simulation: at X reps, Y average AI interactions per day, what is the monthly invoice in month 12? Require a written answer. If the vendor cannot produce one, the pricing model is not designed to be understood.

2. What is the integration maintenance cost annually? Not the initial build cost — the ongoing maintenance. How often does the vendor change their API? How much notice do they give? Who is responsible for updating integrations when they break? For a 50-person sales team with 8 integrations, this number often exceeds $50,000 annually in engineering time.

3. What is the data portability policy if we terminate? You will accumulate years of deal history, contact records, call recordings, sequence data, and commission calculations in this platform. What format is it exported in? How long does the export take? What happens to data after 30 days of termination? A vendor with unfavorable portability terms is calculating switching cost into the contract price.

4. What is the audit trail capability? Can you produce an immutable record of any AI action for a compliance audit? If an AI agent updated a deal stage, enrolled a contact in a sequence, or sent an email — can you prove exactly what happened, who authorized it, and when? For regulated industries and for commission disputes, this is not optional. An append-only audit log written by a service role — not modifiable by users — is the only acceptable implementation.

5. How does commission accuracy connect to your data model? CRM data quality directly affects commission calculations. If deal data is stale, duplicated, or inconsistently staged, commission calculations are wrong. Wrong commission calculations cost 2–4 hours per rep per month in dispute resolution — for a 50-person team, that is 100–200 hours monthly of non-productive time. A platform with an append-only commission audit trail, where every calculation is reproducible from the underlying events, eliminates these disputes.

6. What is the switching cost in Year 3 if the vendor changes pricing? Sales platforms lock in data, workflow, and rep behavior. The switching cost at Year 3 is higher than at Year 1. Ask explicitly: if this vendor increases per-seat prices by 40% in Year 3, what would it cost us to migrate? If the answer is not calculable, you are accepting an open-ended cost risk.

7. What percentage of features are we paying for but not using? Shelfware is endemic to enterprise software. Research consistently shows that organizations use 40–60% of the features they pay for. Ask for a feature usage report from existing customers at comparable team sizes. If the vendor doesn't track feature adoption, they are not measuring whether their product is actually valuable — and neither can you.

Commission Accuracy and the Finance Connection

Commission disputes sit at the intersection of CRM data quality and finance's reporting obligations. The chain is direct: reps log activities in the CRM. Deal stages change. Revenue is booked. Commission calculations run against that data. If the CRM data is stale, duplicated, or missing, the commission calculations are wrong.

The average commission dispute takes 2–4 hours to resolve between rep and finance — a back-and-forth of screenshots, emails, and manual reconciliation. On a 50-person team with 20% of reps disputing at least one calculation per month, that is 20 disputes × 3 hours each = 60 hours per month. At a blended fully-loaded cost of $75/hour, that is $4,500 per month, $54,000 per year, in dispute overhead — not counting the trust erosion between reps and finance that comes from any commission calculation that can't be explained.

An append-only audit trail for commission calculations — where every event that affects commission is logged with a timestamp, user, and cryptographic integrity check — eliminates these disputes. Every calculation becomes provable. Reps can see exactly what triggered their commission. Finance can reproduce any calculation from the underlying event log. The dispute disappears not because someone won the argument but because the facts are unambiguous.

This is a CRM ROI argument that finance can own independently of the sales team's velocity metrics. Reducing commission dispute overhead by 80% on a 50-person team produces a measurable annual savings that appears in the finance ledger directly.

The Audit Trail Question Separates Enterprise from Amateur

In every enterprise software evaluation, the audit trail question distinguishes vendors who have built for compliance from those who built for demos. The question is simple: "Show me the audit log for an AI action that happened three months ago. What does it contain?"

A mature implementation shows: timestamp, user ID, session ID, action type, resource affected (with ID), input parameters, output result, IP address, and a hash that links this log entry to the previous one, making the chain tamper-evident. This is not just a compliance feature — it is operational insurance. When a rep claims the AI sent an email they didn't approve, the audit log either confirms or refutes the claim in seconds.

When an AI agent enrolls a contact in a sequence based on a trigger rule, and that contact later complains about inappropriate outreach, the audit trail shows exactly what happened — what rule fired, what data triggered it, what actions resulted. Without this, the organization has no defense and no ability to learn from errors. A Revenue Operating System built for enterprise treats audit logging as infrastructure, not as a feature to be added later.

The Consolidation Math for Finance

The argument for platform consolidation is often presented to finance as a cost reduction play. That framing is partially correct and partially misleading. The cost reduction is real — collapsing 26 tools into one platform at $699/user/month versus $3,350/user/month is an 80% reduction in license cost. On a 50-person team, that is $132,550/month saved, or $1.59 million annually.

But the more significant financial argument is risk reduction, not cost reduction. Every additional vendor in the stack is a renewal negotiation, a contract risk, a data portability risk, and an integration maintenance liability. A stack of 26 vendors has 26 renewal dates, 26 pricing change risks, and 26 API deprecation risks. A single platform has one.

The ROI calculator that makes this concrete for finance requires three inputs: current stack cost, current integration and maintenance cost, and current administrative overhead cost (training, onboarding, RevOps time). The output is a fully-loaded cost comparison that makes the consolidation case without requiring faith in vendor-provided ROI projections.

For more on how the evaluation plays out at the executive level: see the CRO's first 90 days with a Revenue OS and what VPs of sales evaluate when choosing AI tools. The evaluation process is designed to produce the data that CFOs need — not just the data that sales leaders want to show them.