Enterprise AI adoption is measured by the number of people who use a tool daily, not by the number of licenses purchased. Tom Gersic switched Salesforce’s Lightning target from customers to monthly active users. A customer with 1,000 people and one login had been counted as adopted. For B2B founders, that swap turns a board number describing a purchase into one describing a behavior.
Quick Answer:
Count monthly active users, segmented by team and role. Gersic’s example is a customer with 1,000 people and one daily login. That account produces the same headline figure as full engagement and means the opposite. WalkMe found 54% of workers bypassed a company-provided AI tool within 30 days.
Last updated 24 August 2026.
About Tom Gersic
Tom Gersic is founder and CEO of YouEx.ai. He spent roughly 12 years at Salesforce, rising to VP of Product Adoption. There, he owned the Lightning Experience adoption program and reported its metrics to the board. He then spent two years at Altimetrik, one of OpenAI’s first implementation partners. That work deployed ChatGPT Enterprise into banks and pharmaceutical companies. Most founders who buy AI this year will hit the same gap between signing a contract and using the product. Gersic has crossed that gap twice, once with software people disliked and once with software they loved. The shape of the problem barely changed.
Watch the Episode
Table of Contents
What Does Enterprise AI Adoption Actually Mean?
Enterprise AI adoption is the share of people who use a deployed AI tool in their normal work. Licenses bought are not the same thing, and Gersic saw the gap directly at Salesforce. Half an organization could sit on Lightning while the other half stayed on Classic.
Adoption was never binary at the account level.
“You could have half of your organization using Classic and half of your organization using Lightning,” he says. Then he asks the question that follows. Are they using it or not?
That question is what makes seat counts useless. A signed contract for 20,000 licenses describes an intention. It says nothing about whether anyone opened the software.
The same confusion between activity and outcome is why so many AI projects fail before production.
The problem is rarely the product. Lightning had real performance issues early on, and Gersic is candid about them. Users also disliked being moved.
“Where did this thing go? Where, you moved my cheese,” is how he describes the reaction.
Frustrated people revert to the old version. Nothing about the license count records that.
Gersic’s framing also exposes what the published guidance leaves out. Across the nine pages ranking for this term, “monthly active users” and “active user” appear zero times. That includes Stanford’s 83-page Enterprise AI Playbook.
Stanford uses the word “seat” four times. Every instance concerns SaaS vendors losing revenue. None concerns a buyer mismeasuring its own rollout.
EC-Council found that 45.6% of organizations do not know their own workforce AI adoption rate. Most companies are managing this number without having it.
The Metric That Changed Everything: Monthly Active Users, Not Logos
Monthly active users replaced customer counts as the Lightning adoption target after Gersic’s team missed year one. Counting customers hid the truth because a single login on a 1,000-person account counted the same as full engagement. The measure decided which activities the team could justify.
Gersic inherited that program around 2017. He is unsentimental about where the original targets came from.
“Cross-checked as to feasibility is not necessarily the role of an executive,” he says. The check happens further down the chain. That is the job he walked into.
His argument for changing the measure is arithmetic. Imagine a million customers, each with one user. Now imagine 10% of customers are engaged with every user.
Gersic’s point is that both produce a similar headline figure and mean opposite things. “It might be the same number of people,” Gersic says, “but it’s either one person per environment, or it’s every environment is heavily using it.”
One of those is experimentation. The other group is customers who would notice if you removed the product. Only the second one is adoption.
The switch worked because it dictated which activities the team could justify. “If you know what your target is, you know how to measure it,” he says. “If you know how to measure it, you know what activities you need to take.”
Without that chain, he compares the work to flying a plane with no instruments. The team missed the first year’s target, narrowly met the second, and beat the third.
Three metrics are proposed for this, each measuring different things.
| Metric | What it counts | What it hides | Use it for |
|---|---|---|---|
| Seats purchased | Licenses on the contract | Whether anyone logged in | Procurement, never adoption |
| Token spend | Volume of model calls | Whether the output had value | API workloads and cost control |
| Monthly active users | People using it in a real month | Depth of use inside a session | The headline adoption number |
| Workflow embedding | Named workflows that now run through AI | Nothing, if you name the workflows | Proving the habit stuck |
Gersic uses monthly active users as the North Star and treats the rest as supporting instruments.
Segment the number before you report it. Company size, region, industry, and role each hide different failures. A blended figure tells you nothing about which segment stalled.
Monthly active users became the standard adoption measure for subscription software because it counts humans rather than contracts. Gersic’s team at Salesforce adopted it after account-level counting produced targets nobody could act on. Gersic segments by size, region, industry, and role, which turns the number into a decision rather than a report.
Why Does Enthusiasm Fade Before Adoption Starts?
Enthusiasm fades because it is a reaction rather than a habit. Gersic puts the decay at two or three days, well before any workflow has changed. He saw the same long tail at Salesforce and on ChatGPT Enterprise rollouts into banks.
Gersic expected the banks to resist. They were the opposite. One insurer’s CEO mandated 20,000 ChatGPT Enterprise seats outright.
Financial services had already used AI for years. What changed was the meaning of the word. An earlier AI program meant hiring data scientists and data engineers.
Rolling out ChatGPT Enterprise turned out to be an engineering and user-experience problem instead. “It became a human adoption user experience task,” Gersic says, “more about how are people going to use it.”
The rollout still took the same AI adoption work that Lightning did. That symmetry is the argument. Dislike did not stop Lightning, and excitement did not carry ChatGPT.
The constant is that nobody plans for how people will use the thing. “Enthusiasm’s great for a day or two, and then the third day comes in and people are onto something else,” Gersic says.
He calls adoption “the long tail of what happens after people are excited.” Excitement is the first step of adoption.
Gersic’s warning is that treating it as a result produces activation figures which decay quietly.
Founders hit the same trap when their AI thought leadership strategy measures output rather than trust. The number looks fine for a quarter. Nobody checks whether the same people came back.
OpenAI’s own data makes the case against the seat count. OpenAI’s state of enterprise AI report found that 95th-percentile workers send six times the median employee’s message volume.
OpenAI presents this as evidence that leading users are pulling ahead. Read from the buyer’s side, it says something else. A healthy-looking average can rest on a small, unrepresentative core.
WalkMe surveyed 3,750 workers across 14 countries. WalkMe found 54% had bypassed a company-provided AI tool and done the task manually within 30 days.
Enthusiasm and adoption are different measurements, and Gersic found that excitement fades within two or three days. Adoption is the sustained usage that follows, measured in people rather than licenses. OpenAI reports that 95th-percentile workers send six times the median employee’s message volume.
Real Sponsorship Is Presence, Not Proclamation
Executive sponsorship of an AI rollout is presence, not an announcement at an all-hands. Gersic’s test is whether the sponsor keeps turning up after launch week. Stanford found sponsorship was the most frequent acceleration factor across 51 enterprise deployments, at 43%.
Most founders believe they are already doing this. Gersic’s definition is narrower and harder to fake.
“Real sponsorship looks like presence,” he says. It looks like a founder who stayed for the after-party. It looks like one who leaned on the executive team when targets slipped.
On one customer training program, the CEO sat through the entire session. Gersic says nobody questions the priority when that happens. The room stops debating and starts learning.
When the CEO cannot attend, he recommends a designated sponsor on a fixed schedule. A daily 15-minute meeting works. Attendance itself carries the signal.
The corollary matters as much as the presence. “When that person leaves, that’s a signal that maybe this isn’t as important as it once was.”
Sponsorship also decides who carries the number. Gersic is direct about what a founder owes the board when it is bad. “You can’t blame your team, ever,” he says.
Inheriting a target does not transfer the accountability. He is blunt about the position you are put in. “We didn’t hit the target, I failed, this is how we’re gonna change it.”
Stanford’s Enterprise AI Playbook reached the same conclusion from 51 enterprise deployments. Its fourth finding states that sponsorship is about actions rather than approval. Effective sponsors clear blockers weekly and tie adoption to corporate OKRs.
Sponsorship was the most frequent acceleration factor in that sample, at 43%. McKinsey found 44% of leading companies have CEO or board-level AI ownership. Among bottom performers, the figure is 17%.
Brainstorm Inc measured the activation gap directly. Organizations with executive sponsorship reached 50% Copilot activation within 90 days. Those without reached 28%, on identical license counts.
Is Your Revenue Team Already Using AI You Cannot See?
Go-to-market teams do most of their AI work on free accounts their employer cannot see. Harmonic Security measured free-account AI hours at 28.6% and enterprise-plan usage at 10.1%. That gap names exactly which teams your sanctioned rollout has failed to serve.
Harmonic Security analyzed 1,935,247 classified AI-session minutes. The split is one most executives would not predict.
Go-to-market teams account for 28.6% of free-account AI hours. They account for only 10.1% of enterprise-plan usage. Legal, by contrast, accounts for 32.3% of enterprise-plan activity.
Read together, those numbers say your real AI adoption is happening where you cannot see it. WalkMe’s 54% bypass figure says the same thing from the other direction. The work happens either way.
Gersic thinks the personal accounts came first. He compares it to the iPad arriving in the enterprise. Executives bought one at home, then asked how to run business apps on it.
That blend of home and work technology is what makes the pattern so hard for any employer to police. “If you don’t roll out technology for your employees, they will find ways to do it,” he says, “and it’s hard to catch.”
His recommended response is supply rather than restriction. Governance, in his framing, is what makes speed possible.
He expects most people want to do the right thing. Give them a sanctioned tool that does the job and the shadow usage stops being worth the effort.
Regulation is moving the same way. Gersic points to the EU AI Act changing how companies think about watermarking. Images are already detectable as AI-generated, and he treats that as a good thing.
He is direct about where accountability lands. If an AI drafted the message and you sent it, you own what it said.
Reviewing it first is your decision and your risk. A policy can require the review. Only a person carries the consequence of skipping it.
Do Not Fund AI by Cutting the People Who Create Demand
Cutting sales development headcount to fund AI removes the people who generate the pipeline. An observability company tried to cut 20% of its SDRs but couldn’t. Gersic reads that failure as a lucky escape rather than a setback.
He does not read it as a failure at all.
“Trying to replace an SDR is insane,” he says. “Make them, find their superpower and make it even better.” He treats the failed cut as a lucky escape.
The goal in a competitive market is growth rather than lower cost. His argument about entry-level work is specific. The tasks change and the roles persist.
What a great enterprise seller does is research obsessively and hold relationships. They can name the buying committee before they walk in. They know who knows whom.
The useful application is handing a junior rep that same map before the call. That is why every lead in YouEx.ai arrives with a research report attached. Other agents then score it and route it.
Gersic’s aim is to raise the floor. “How do you make your newest, weakest perform as good as your best SDR?”
Evidence on AI adoption cuts both ways, and founders should know it. ICONIQ Growth found that 71% of SDR teams now have most of their staff using AI regularly. That is the highest of any go-to-market function.
Stanford found that headcount reduction was the most common outcome in 45% of deployments. The other 55% chose hiring avoidance, redeployment, or no reduction at all. Cutting is not the majority behavior.
There is also a trend that predates AI entirely. Bridge Group data shows SDR-to-AE promotion rates fell from 34% in 2020 to 16% in 2024.
That path was already narrowing before any AI SDR shipped, on Bridge Group’s numbers. Gersic’s point is that the codified half of the role is what AI removes cleanly.
The tacit half is different. Gersic argues that live objections, multi-threading, and discovery are where a rep learns the job. Automate those and Stanford’s finding on codified work stops protecting you, because the training ground goes with them.
Surviving the Trough When the Budget Is Annual
AI programs stall around month six, after the integration work and before the value lands. Gersic calls this the trough of disillusionment and says the way to lose money is to quit while in it. Annual budget cycles turn that dip into a permanent cancellation.
His pattern for failed programs is consistent. Learning is what happens down in the trough.
The trap is structural rather than emotional. Most companies measure year over year. In Gersic’s experience, a program without momentum inside one fiscal year rarely gets a second one.
Gersic warns that budgets get reallocated to other things. Three clocks collide to produce that outcome.
Novelty fades within weeks. One analysis of adoption curves puts peak usage at week four.
By month six, it settles at about a fifth of target users, regardless of the tool. Reporting then demands a result at the quarter.
The budget does not reopen for twelve months. Nothing on the current first page for enterprise AI adoption addresses that collision.
Funding discipline is drifting in the wrong direction too. Open Future Forum found 34% of companies have no clear AI budget at all. A month earlier, the figure was 23%.
That drift matters more than it sounds. A program with no named budget line has no renewal conversation. It simply stops when attention moves.
Gersic’s counter is to treat the dip as expected rather than as evidence of failure. Programs that survive it keep collecting feedback and shipping improvements while the enthusiasm is gone.
The practical response is to make the funding decision before enthusiasm dies. Gartner’s 2026 guidance is that funding expands fastest where outcomes are measurable inside the spend’s own cycle.
Agree the kill-or-continue metric at kickoff, while the program still has goodwill. Do not improvise it in month seven when the number already looks bad.
The Four Conditions Every Rollout Needs, Sized for a Small Company
Four conditions carry every rollout Gersic has run: executive support, communications, enablement, and measurement. Executive support comes first because it removes the debate about whether the work matters. Measurement comes fourth and governs the other three, because untracked adoption cannot be managed.
Gersic’s most successful customer programs had the CEO in the room.
Communications comes second, and one launch email does not qualify. He compares it to selling, where the first email is only the first email.
Enablement is separate again. Telling people a tool exists is not teaching them to use it. The two get collapsed constantly.
Measurement is fourth, and it governs the other three. That makes this as much a decision-making framework as a rollout plan. “You can’t focus on adoption without really being able to know what the measures are,” he says.
Then there is the rollout shape, which most teams get backward. Gersic recommends starting with five people who are genuinely excited. Salesforce called those people trailblazers.
You want the ones who give hard feedback because they care. Let them see that feedback take hold. Demand then pulls the rest of the organization along.
“Turning it on for 1,000 people and good luck. That’s a recipe for disaster. But turning it on for five people and making it a scarce commodity that only the handful of really exciting people have is, okay, now everybody else wants it.”
Size changes the sequence. At Salesforce, the smallest customers could simply have Lightning switched on. The largest needed the rollout planned with their own IT first.
Under 250 people, this gets simpler rather than harder. You do not need a center of excellence or a governance board. You need one named owner per function and a standing 15-minute meeting.
He also warns against generic empathy for the resistant. Most people are not refusing the tool. They are protecting a job that already works.
“It’s rarely anybody’s job to learn a new technology,” he says. “They’re trying to get their job done, and you’re telling them that they need to change how they get their job done.”
Frequently Asked Questions
How do I measure enterprise AI adoption if seats are meaningless?
Count monthly active users, then segment them by team, role, and company size before reporting anything. Tom Gersic switched Salesforce’s Lightning target from customers to monthly active users. Account-level counting had produced targets nobody could act on. A blended figure hides the difference between one user per account and full usage by a tenth of them. Gersic advises checking whether sales usage clusters at month end, which indicates quota compliance rather than value.
What is a healthy monthly active rate for an AI tool?
AI adoption rates depend entirely on the role, and expecting a single number across the company is a mistake. Gersic points out that a customer service representative sits in their tool all day. Gersic expects adoption there to approach 100% unless they are on holiday. A traveling enterprise seller is a different case because no one pushes a top performer to log activity. Set the expected rate per role before the rollout, then measure each role against its own number.
My team loved the demo, and now nobody uses it. What happened?
Enthusiasm was measured and mistaken for adoption. Gersic found excitement typically lasts two or three days before attention moves elsewhere. He calls adoption “the long tail of what happens after people are excited.” The usual missing pieces are a communications plan beyond the launch email, enablement kept separate from communication, and a sponsor who kept turning up. WalkMe found that 54% of workers bypassed a company AI tool and completed the task manually within 30 days.
Should I restrict personal AI accounts or replace them?
Supply the sanctioned alternative first and read the unsanctioned usage as product research. Harmonic Security found that go-to-market teams account for 28.6% of free-account AI hours, compared with 10.1% of enterprise-plan usage. That tells you precisely which teams the official rollout failed. Gersic’s position is that governance should enable speed, and that people find their own tools when the company provides none. Restriction without replacement makes the usage invisible rather than absent.
Is token spend a better AI adoption metric than monthly active users?
Token spend measures volume rather than value, and Gersic compares relying on it to counting a developer’s lines of code. It matters for API workloads, where matching the right model to the task prevents real waste. For seat-based products such as ChatGPT Enterprise or Copilot, tokens are included in the license. They reveal little about whether a person got value. Monthly active users answer the question a founder is actually asking about their own team.
How long should I wait before cutting an AI rollout?
Set the kill-or-continue metric at kickoff rather than deciding after the enthusiasm fades. Gersic’s failure pattern is six months of integration work followed by the evaporation of interest, which is the trough of disillusionment. Annual measurement cycles compound it. Gersic notes that a program without momentum in one fiscal year rarely receives a second year. Gartner’s 2026 guidance is that funding expands fastest where outcomes are measurable inside the same cycle as the spend.
Conclusion
Gersic’s two rollouts sit either side of the same finding. Lightning had performance problems and a user base that loved the old version. ChatGPT Enterprise arrived with a CEO mandate and genuine excitement. Both produced the same long tail. The variable that mattered was never how people felt about the software.
For a B2B software company, the adoption number on your board deck probably describes a purchase rather than a behavior. Switching your enterprise AI adoption measure to monthly active users, segmented by role, changes which problems you can see. It also changes the conversation with your revenue team. You move from defending a rollout to fixing the segment that never embedded it.
Sproutworth’s ghostwriting for B2B SaaS founders is built for founders with a point of view they have not yet published.
The founders who gain most from AI will be the ones who planned for human behavior on day one.
Related Resources
- Why AI Projects Fail: 95% of Pilots Deliver No Return
- 5 Reasons Your AI Thought Leadership Strategy Isn’t Working
- Content That Earns Trust: A B2B CEO’s Pre-Sale Playbook
- Decision-Making Framework: 3 Signs You’ve Outgrown Yours
- LinkedIn Ghostwriting for SaaS Founders: Pricing, Process, and ROI
Related Links
Some topics we explore in this episode include:
- Product Adoption Strategies: Focus on user workflows, not just product features, to drive real adoption.
- Monthly Active Users (MAU) as Key Metric for Enterprise AI adoption: Moving from vanity metrics to MAU for measuring true engagement.
- User Experience and Performance Issues: How usability and perceived speed barriers affect adoption.
- Executive Sponsorship: The critical impact of hands-on leadership on rollout success.
- Hype Cycle and Disillusionment: Persevering through the “trough of disillusionment” when initial excitement fades.
- Enthusiasm vs. Sustained Enterprise AI Adoption: Why excitement isn’t enough for long-term, meaningful use.
- Shadow AI Use and Governance: Risks of unapproved AI tool use and the enabling role of strong governance.
- Embedding AI in Workflows: The imperative to deeply integrate AI into daily business processes.
- Measuring Productivity and Value: Aligning metrics to true business outcomes instead of surface-level activity.
- AI in Regulated Industries: Unique challenges and lessons from deploying AI in sectors like finance and insurance.
Listen to the episode
Subscribe to & Review the Predictable B2B Success Podcast
Thanks for tuning into this week’s Predictable B2B Podcast episode! If the information from our interviews has helped your business journey, please visit Apple Podcasts, subscribe to the show, and leave us an honest review.
Your reviews and feedback will not only help me continue to deliver great, helpful content but also help me reach even more amazing founders and executives like you!
