In February 2024, Klarna said its AI assistant was doing the work of 700 customer service agents and handling about 75% of customer chats. By May 2025, CEO Sebastian Siemiatkowski was telling Bloomberg it was critical to be clear to customers that “there will always be a human if you want,” and Klarna was recruiting human agents again, according to Entrepreneur’s report.
AI and humans working together is the strongest operating model for 2026: people alongside AI tools, chatbots and CRM automation, with each one given a clear job. Klarna’s assistant did handle a huge volume of conversations. What changed was the company’s view of quality, which its CEO said had fallen in the AI-only approach.
For US business owners and operations leaders deciding where to automate and where to keep people, the evidence from the past three years points in one direction. The sections below cover what a hybrid model looks like, what the research says, where each task belongs, and how to build one.
“Human Plus AI” in Practice
Human plus AI is an operating model where software handles repetitive, rules-based work and people handle judgment, relationships and exceptions. In a small or mid-sized business, it usually has four parts:
- AI tools: writing, research and summarization assistants that speed up a person’s work.
- Chatbots: the first point of contact on a website or support channel. They answer routine questions and collect details.
- CRM automation: workflows inside a customer relationship management system that log activity, route leads, trigger reminders and update records.
- Human assistants: executive, administrative and operations staff who review AI output, handle exceptions and own the relationships.
The first three are tools. The fourth decides whether those tools produce good work or expensive mistakes.
Research About Humans and AI Working Together
Four sources, all published in the past three years, line up on the same conclusion.
AI makes people faster and helps newer people the most
Researchers from Stanford and MIT studied 5,179 customer support agents using a generative AI assistant. Issues resolved per hour rose 14% on average. For novice and low-skilled workers the gain was 34%, while experienced workers saw minimal impact, according to the National Bureau of Economic Research paper by Erik Brynjolfsson, Danielle Li and Lindsey Raymond.
A reasonable reading for outsourcing: a new team member working with an AI tool should reach full speed sooner. That is GO’s interpretation of the study, not a finding the authors tested for outsourced teams.
Customers want AI to work and a person to be reachable
| 87% of customers say it is essential for companies using generative AI to give them access to a human agent. Source: Gartner survey of 3,566 customers, February to March 2026. |
The same Gartner survey found that 50% of customers say interactions are easier when companies use generative AI. Customers like AI when it works. The risk is what happens when it does not. In Gartner’s words, “when customers are forced through multiple unsuccessful AI interactions before they can reach a person, they are less likely to use that tool again.”
A separate Gartner survey found customers were about three times more likely to use third-party generative AI, such as ChatGPT, Gemini or Copilot, than a company’s own chatbot for service questions. A weak chatbot is competing with free tools your customers already have. It needs to be useful, or it needs to get out of the way and hand over to a person.
Full replacement rarely holds up
Gartner predicts that by 2028 none of the Fortune 500 will have fully eliminated human customer service, and that by 2027 half of the organizations that expected large AI-driven workforce cuts will abandon those plans. Senior Director Analyst Kathy Ross put the reason plainly: AI “excels at handling routine and well-defined problems” but “often struggles with exceptions and high-risk scenarios.”
The payoff comes from redesigning the work, not buying the tool
McKinsey’s State of AI survey (1,719 participants across 97 countries, fielded May to June 2026) found that 73% of high-performing organizations had fundamentally redesigned their workflows around AI. Among all other respondents, the figure was 25%. High performers were a small group, just 6% of respondents, defined by AI contributing at least 5% of earnings before interest and taxes.
The same survey found that 14% of respondents reported workforce reductions from AI, well under the 32% who had predicted them a year earlier. Buying software is the easy part. Deciding who does what once the software is running is where the results come from.
Where AI Should Work and Where People Should Stay in Charge
No rule fits every business, but two questions sort most tasks: how repetitive is it, and what happens if it goes wrong? The table below is GO’s recommended starting point, based on the sources in this article. It is a judgment call, not a research result.
| Task | Best handled by | Why |
| Answering routine questions (hours, service scope, order status) | Chatbot, with a path to a person | The answers are documented and repeat daily. |
| Drafting emails, summaries and first-pass research | AI tool, reviewed by a person | Saves time, but someone must check facts and tone before anything is sent. |
| Logging calls, updating records, routing leads | CRM automation, audited by a person | Rules-based work. Errors multiply when nobody checks the data. |
| Refund exceptions, complaints, contract questions | Person | These carry judgment and legal exposure. |
| Client relationships, renewals, upsell conversations | Person | Trust is built in conversation. |
| Quality checks and process improvement | Person | Someone has to notice what the software is getting wrong. |
How AI Tools, Chatbots and CRM Automation Work With People
AI tools: draft faster, decide slower
Salesforce’s State of Sales report, which surveyed 4,050 sales professionals in 22 countries, found that 87% of sales organizations use some form of AI. Sellers spend about 40% of their week actually selling. They expect AI agents to cut prospect research time by 34% and email drafting time by 36%.
Two cautions. Salesforce sells these tools, and the time savings are expectations from respondents, not measured results. Even so, the direction is useful. An assistant who uses AI to prepare research and first drafts can hand a salesperson material that is ready to review, and the salesperson spends the saved time on calls.
Chatbots: a front door with a person behind it
In February 2024, British Columbia’s Civil Resolution Tribunal held Air Canada liable after its chatbot gave a customer wrong information about bereavement fares. The tribunal said the airline was responsible for all the information on its website, whether it came from a static page or a chatbot. The award was about $650 CAD, according to the American Bar Association’s summary. The amount was small. The principle applies to every company with a chatbot.
Three habits reduce that risk. Limit the bot to answers that are documented and current. Always show visitors a way to reach a person. Have an assistant review chatbot transcripts on a regular schedule and correct anything that no longer matches policy.
CRM automation: only as good as the data
Salesforce’s State of Service report, based on research with 6,500 service professionals, found that companies with unified customer service channel data were 1.4 times more likely to describe their AI implementation as very successful. The same report projects that AI will resolve 50% of service cases by 2027, up from 30% in 2025. Treat that as a vendor forecast, not a fact.
The data finding is the practical one. Automation built on duplicate contacts, stale fields and inconsistent tags produces messy results faster. A data entry analyst or operations assistant who keeps records clean is what makes the automation worth running.
Human assistants: owners of exceptions and relationships
The assistant’s role shifts toward reviewing, following up and handling the cases tools cannot. Salesforce’s service research suggests staff are open to it: 83% of service professionals reported better career prospects with AI, and 82% said they were developing new skills. These are self-reported figures, but retention matters in any outsourced team, and people who see AI as a path to better work tend to stay longer. That last point is an opinion, not a finding.
A Hypothetical Example: AI and Humans in Lead Handling
The scenario below is illustrative. It is not a client case.
Picture a 30-person engineering firm that receives inquiries through its website, email and referrals. A hybrid setup could work like this:
- A website chatbot answers basic questions about services and collects project type, location and timeline.
- CRM automation creates the contact, tags the lead source and assigns a sales owner.
- An AI tool drafts a first reply and a short summary of the prospect’s company.
- An administrative assistant reviews both, fixes anything wrong, adds a personal note and sends the reply the same business day.
- The salesperson calls with the summary in hand. The assistant logs the outcome and schedules the follow-up.
Software handles the mechanics. People handle accuracy and the conversation. If the chatbot cannot answer, the visitor sees a clear way to reach someone. Remove the assistant and step four disappears, so the prospect gets an unchecked draft. Remove the software and the assistant spends the morning retyping. Neither version is as strong as both together.
How to Build a Team Where AI and Humans Work Together
1. Start with one workflow
Pick a single process, such as lead follow-up, invoice processing or support triage. McKinsey’s finding on workflow redesign applies at small scale too. It is easier to redesign one process properly than to add AI to ten processes loosely.
2. Sort tasks by repetition and risk
Repetitive, low-risk tasks go to software. Anything that involves money, contracts, complaints or a client’s trust stays with a person. The table earlier in this article is a starting template.
3. Write the handoff rules
Decide when the chatbot escalates, who receives the escalation and how fast they respond. Gartner’s finding about customers being forced through multiple failed AI attempts suggests setting a low limit. A cap of two failed attempts before routing to a person is a reasonable starting point, and it is GO’s suggestion rather than a tested benchmark.
4. Give someone ownership of the data
Name one person responsible for CRM accuracy: duplicates, required fields, tags and stage definitions. Without an owner, automation drifts out of date.
5. Train people on the tools
The Stanford and MIT study showed the largest gains for newer workers, which means onboarding on the AI tool should be part of the first week, not an optional extra.
6. Measure more than speed
Track resolution rate, customer satisfaction, rework and escalation volume alongside time saved. Klarna’s CEO said the AI-only approach was cheaper but produced lower quality, per Entrepreneur’s reporting. A dashboard that only shows cost would have missed that.
Why the Philippines Fits a Hybrid AI and Human Model
The Philippine IT-BPM industry is reshaping around this model. The industry association IBPAP now projects $42.3 billion in revenue and 1.96 million employees for 2026. It also lowered its 2028 outlook from the 2022 roadmap’s $59 billion and 2.5 million workers to a range of $43.3 to $50.5 billion and 1.85 to 2.14 million workers, citing rapid AI adoption, shifts in buyer behavior and global competition.
IBPAP CEO Jack Madrid has described the new goal as 2 million AI-enabled workers instead of a raw headcount target. The industry is saying openly that the work is changing and that its people need to be trained for it.
For a US buyer, that is useful information. It suggests Philippine teams are being prepared to work with AI tools, and it gives you a fair question to put to any provider: how are your people trained to use them?
Questions to Ask an Outsourcing Partner About AI
- Which AI tools do your team members already use, and how are they trained on them?
- Who reviews AI-generated work before it reaches my customers?
- How do you handle data security when staff use AI tools with client information?
- Can team members work inside the CRM and chatbot platforms I already use?
- How do you measure quality, and does that measurement go beyond speed?
Guided Outsourcing helps US businesses build dedicated teams in the Philippines. Pricing is a monthly flat rate that includes payroll and benefits administration, with no hidden charges. On the admin side, roles include Executive Assistants, Administrative Assistants, Operations Analysts, Research Analysts and Data Entry Analysts, the people who sit next to the tools and keep them accurate. GO also builds IT teams and accounting teams for the technical and financial work that a hybrid model depends on.
Frequently Asked Questions
Will AI replace virtual assistants and administrative staff?
Specific tasks, yes. Whole roles, not on the evidence available. Gartner expects no Fortune 500 company to have fully eliminated human customer service by 2028, and in McKinsey’s survey far fewer companies reported AI-driven workforce cuts than had predicted them. What changes is the mix of work: less retyping and data entry, more review, follow-up and exception handling.
Which tasks should stay with people?
Anything involving judgment, money, legal exposure or a client relationship. Refund exceptions, complaints, contract questions and renewals belong with a person, with AI supporting in the background.
Does a small business need CRM automation?
It helps once follow-ups are being missed or lead information is scattered across inboxes and spreadsheets. Start with the simplest automations: lead capture, reminders and routing. Keep a person responsible for checking that the data is right.
How long before a hybrid model pays off?
No reliable universal timeline exists. McKinsey’s survey found only 37% of respondents reported AI contributing to earnings, so results are not automatic. Run one workflow for a full business cycle, measure quality as well as speed, and expand only once it works.
The Practical Next Step
Choose one workflow this month. List every step, mark which ones are repetitive and which need judgment, and decide who owns the judgment steps. If that person does not exist on your team yet, that is the hire to make first.
If building that team in the Philippines is on your list, talk to the GO team about the roles that fit your workflow.
