Most articles about AI in companies read like press releases. "AI is transforming everything." "The future is here." Zero specifics, zero accountability.
This is the opposite of that.
Companies implementing AI across their operations tend to learn the same thing: some uses move numbers, and others generate impressive-looking demos that nobody uses six months later. This piece breaks it down by department, with adoption patterns, failure modes, and what to measure.
No hype. No hedging.
The Actual State of AI Adoption in 2026
Let's start with numbers that matter. McKinsey's State of AI in 2025 report found that 78% of organizations now use AI in at least one business function, up from 55% just two years prior. But usage and value are different things.
The same report found that only 22% of companies say AI has meaningfully reduced costs or grown revenue. The gap between "we use AI" and "AI works for us" is enormous, and it's almost entirely explained by implementation quality, not technology limitations.
Gartner estimates that through 2025, 85% of generative AI projects will fail to deliver on their intended business outcomes. That's not a technology problem. That's a deployment problem.
Here's what's happening by function.
Finance: A Clear Operating Case
Finance was the first department to get real value from AI, and it's still the most defensible case for implementation.
What works:
Accounts payable and receivable automation. AI can process invoices, match purchase orders, flag discrepancies, and route exceptions while sending uncertain cases to a human reviewer.
Financial close acceleration. AI-assisted reconciliation, variance analysis, and report generation can shorten close work when the underlying ledger and controls are sound.
Anomaly detection in transactions. AI flags unusual spending patterns, potential fraud, and policy violations in real time. This is not a nice-to-have anymore, it's becoming table stakes for any company that processes more than a few hundred transactions per month.
Forecasting. AI models can incorporate more historical and operating signals than a manual spreadsheet, particularly for businesses with seasonal patterns or multiple revenue streams.
What doesn't work:
Replacing CFO-level judgment. AI cannot assess the strategic risk of a major acquisition, navigate a banking relationship, or make a call on whether to extend credit to a specific customer. Companies that try to automate strategic finance decisions discover that quickly.
Implementation on dirty data. Finance AI requires clean, consistent data. If your chart of accounts is a mess, your vendor master is duplicated, or your historical data has gaps, data cleanup must come first.
Operations: High Ceiling, High Effort
Operations is where AI has the highest theoretical upside and the most implementation complexity. The variance between companies that get it right and those that don't is wide.
What works:
Demand forecasting and inventory optimization. Harvard Business Review argues machine learning can improve demand forecasting because it uses data and signals that conventional models ignore.
Predictive maintenance. For companies with physical equipment, AI monitoring can surface failure signals before equipment stops. Track downtime, maintenance lead time, and false alerts.
Process automation (the real kind). Not just form routing, but end-to-end process handling. Customer orders processed, confirmed, routed to fulfillment, and tracked without a human touching any step.
What doesn't work:
Boiling the ocean. Companies that try to automate everything at once end up automating nothing well. Operations AI implementations that work start with one process, validate performance, then expand.
Ignoring change management. Operations staff who feel threatened by AI actively undermine implementations. This isn't a technology problem, it's a people problem that has to be solved before you deploy anything.
HR: Faster Hiring, Smarter Retention
HR was slow to adopt AI but is catching up quickly, particularly in recruiting and retention modeling.
What works:
Recruiting pipeline acceleration. AI screening of resumes and initial candidate qualification can shorten time-to-interview for high-volume roles when humans retain final hiring decisions.
Retention risk modeling. AI can identify employees likely to leave based on engagement signals, tenure, compensation relative to market, and manager feedback patterns. Predictive retention models can give HR teams months of advance warning on flight risk, enough time to intervene before someone walks.
Onboarding automation. Document collection, system provisioning, policy acknowledgments, training scheduling. All of this can run without HR coordinators manually touching every new hire.
Benefits administration. Answering employee questions about benefits, eligibility, enrollment windows, and claims is a huge time sink for HR teams. AI handles this at scale with better consistency than humans.
What doesn't work:
Fully automated hiring decisions. AI screening is a filter, not a decision-maker. Companies that let AI make final hiring calls face legal exposure and miss candidates who don't pattern-match to historical hires. Use AI to surface candidates, humans to decide.
Generic engagement surveys with AI analysis. If your underlying survey questions are bad, no amount of AI analysis makes the data useful. Fix the input first.
Customer Service: The Biggest Win Most Companies Leave on the Table
Customer service is often a strong AI implementation candidate for mid-market companies because the work is high-volume, measurable, and rich in repeatable requests.
What works:
Tier-1 support automation. Common questions, account lookups, order status, and basic troubleshooting can follow defined policies, with ambiguous or sensitive cases escalated to a human.
Agent assist. For complex issues that need humans, AI surfaces relevant knowledge base articles, previous case history, and suggested responses in real time. Agents handle more tickets faster and with higher consistency.
After-hours coverage. AI doesn't sleep. Companies offering 24/7 AI-first support with human escalation during business hours see meaningful customer satisfaction improvements, particularly for international customers in different time zones.
Sentiment analysis and escalation routing. AI that detects frustrated or high-value customers and routes them to senior agents immediately, before they churn.
What doesn't work:
Deflection-first AI. Customers are not stupid. If your AI chatbot exists to prevent them from reaching a human rather than to actually help them, they will notice, and your NPS will tell you about it.
Deploying AI on top of bad processes. If your return policy is confusing, your AI will confuse customers faster than your humans did. Fix the process, then automate it.
Sales: Efficiency Gains, Not Replacement
Sales teams have the most mixed relationship with AI of any department. The tools are genuinely useful. The adoption rates among salespeople are genuinely low.
What works:
Lead scoring and prioritization. AI that ranks inbound leads by likelihood to close, company fit, and urgency signals means reps spend time on the right opportunities. Across the industry, top-performing sales teams are consistently more likely than underperformers to lean on AI for scoring and prioritizing leads.
Call recording and analysis. AI that transcribes sales calls, identifies objections, tracks competitor mentions, and flags coaching opportunities gives sales managers visibility they never had before.
Outreach personalization at scale. AI-generated personalized outreach based on prospect research, recent news, and company signals. Not "Hi [First Name]" personalization, but genuinely relevant context.
Pipeline forecasting. AI-driven forecast accuracy is meaningfully better than rep-reported pipeline in most organizations. Fewer surprises at quarter end.
What doesn't work:
AI for complex B2B relationship selling. Enterprise deals, family office relationships, high-stakes negotiations. These require human judgment, rapport, and reading the room. AI can support the process, it cannot drive it.
Forcing adoption without incentive alignment. If AI tools add reporting burden without reducing rep workload elsewhere, adoption rates will be low regardless of mandate. Give reps time back.
Marketing: The Fastest-Changing Function
Marketing has arguably been changed more dramatically by AI than any other department over the past two years. The pace of change is still accelerating.
What works:
Content production at scale. Blog posts, social content, email sequences, and ad copy. AI does not replace marketing strategy, but it can reduce the production bottleneck when humans review the output.
Audience segmentation and personalization. AI-driven segmentation that goes beyond demographics to behavioral patterns and purchase intent. Personalized email sequences that adapt based on engagement. Campaign ROI improves materially.
SEO at scale. AI tools for keyword research, content briefs, internal linking, and technical SEO audits compress what used to take weeks into days.
Ad optimization. Automated bidding, creative testing, and audience targeting in Google and Meta can use campaign data on the schedule set in your SOW, while the marketing team retains budget and brand controls.
What doesn't work:
Brand voice on autopilot. AI content that goes out unreviewed is a brand risk. The quality distribution is wide. The best AI content is excellent. The worst is incoherent or factually wrong. Human review of AI output is not optional.
Replacing brand strategy with AI. Positioning, messaging architecture, campaign concepts. These require human insight into customer psychology and competitive dynamics. AI helps execute strategy, it does not create it.
What the Failures Have in Common
AI implementations that fail tend to share a pattern.
They start with the technology. "We bought Copilot, now what?" Or "We deployed an AI chatbot." The technology precedes the use case. The team never articulates what problem they're solving or what success looks like.
They underinvest in change management. AI changes how people work. People resist change. If you don't actively manage the transition, including what happens to the roles that change, passive resistance can undermine adoption.
They don't measure the right things. Vanity metrics like "AI hours saved" or "tickets deflected" do not show whether cycle time, error rates, throughput, or customer outcomes improved.
They try to do too much at once. The companies that get the most from AI pick one department, one process, deploy carefully, validate the operating metrics, and then expand. The companies that fail try to transform everything in parallel and end up transforming nothing.
Two Related-Company Examples
Here are two examples from companies owned by Leverwork's founder.
JSVHQ (a company owned by Leverwork's founder, Jonas Sørensen; a related company, not an independent client, and its results are not a prediction of yours) uses AI for company research, financial analysis, market sizing, and initial screening. A person handles judgment calls and relationship-sensitive decisions.
SORNA Limited (also owned by Leverwork's founder; a related company, not an independent client) cut its compliance cycle time by over 85% by automating document collection, verification, cross-referencing, and flagging, with one human review gate.
What Good Implementation Costs
Transparency on pricing, because most vendors aren't.
A proper AI workforce implementation is scoped to your actual business processes and integrated with your existing systems. Leverwork uses a fixed-price setup plus a monthly retainer, scoped on a 20-minute call. Talk to us about the workstream.
The scope includes process mapping, system integration, AI configuration, monitoring, and performance tuning.
Compare proposals by scope, exception handling, integration ownership, monitoring, and the operating metrics written into the SOW.
The Department Where You Should Start
If you're a mid-market company considering AI and don't know where to start, the answer is almost always customer service or finance.
Customer service is often a useful starting point because repeatable work is easy to measure and escalation rules can be tested before wider rollout.
Finance because the accuracy requirements force proper implementation discipline, the data is usually cleaner than other departments, and the operating metrics are visible.
Start there. Prove it. Then expand.
Find Out Where AI Will Have the Most Impact in Your Business
The department breakdown above is directional. What actually matters is your specific situation, your processes, your team structure, and your data.
We run a structured assessment that maps your current operations, identifies the best-fit AI opportunities, and produces an implementation sequence. Your 20-minute assessment estimates the figures for your roles.
If you'd rather talk through your situation directly before committing to anything: