Mike HeapMike is an experienced Product Manager who focuses on all the “non-development” areas of My AskAI, from finance and customer success to product design, copywriting, testing and more.
Table of Contents
- Why is the in-house vs outsourcing debate stuck on the wrong question?
- The repeat, judgment and presence split
- Repeat work goes to AI
- Judgment work stays in-house
- Presence work is what's left to outsource
- Why is the outsourcing window closing?
- What does this look like in real rollouts?
- YouGarden: 66% AI resolution, 965 hours a month
- TravelJoy: 80% AI resolution with no night shift
- Edel Optics: 53 countries at 92% AI CSAT
- Kriptomat: 62% AI resolution through the surges
- What to do this week
- How do I get AI to run the three-band split for me?
- When does outsourcing still win?
- The takeaway
- FAQs
Created time
Jul 30, 2026 04:02 PM
Title length (<60)
Author
Last optimised
Ecomm?
Image
Publish date
Jul 29, 2026
Video
Slug
outsourcing-customer-service-pros-and-cons
Featured
Type
Article
Ready to Publish
Ready to Publish
You're running the in-house-versus-outsourcing spreadsheet on the wrong denominator: most of the volume you're pricing doesn't need a person at all.
You cost out both options the same way, seat by seat. Keep support in-house and you pay salaries, recruitment, tooling and management. Send it to a BPO and you pay a lower rate per agent, with a trade in control and brand voice.
The pros and cons on both sides are real, and every ranking page lists roughly the same eight of them. None of them ask how many of the tickets in that quote need a human at all. I'd start there: ask how much of the queue needs staffing before you ask who staffs it.
Answer that first and the outsourcing decision shrinks to a smaller, much more specific contract, because software now absorbs the repeat volume that used to take people.
I'm Mike, co-founder of My AskAI. We run AI support agents for 200+ ecommerce and SaaS businesses inside Zendesk, Intercom, Freshdesk, Gorgias and HubSpot.
Our agents have resolved more than 1,000,000 tickets, and the customer base holds a 72%+ resolution rate on a rolling 30-day basis. Four of those rollouts (YouGarden, TravelJoy, Edel Optics and Kriptomat) supply the numbers in this post.
Why is the in-house vs outsourcing debate stuck on the wrong question?
TL;DR: The standard comparison prices two ways of staffing the whole queue. It never asks what share of that queue is documented repeat work.
The consensus case is well made, so I'd read it before arguing with it. Zendesk's outsourcing guide gives the clean definition: "Customer service outsourcing is when businesses hire third-party companies to handle their customer service."
Business.com's guide to outsourced customer service opens on the discomfort operators feel, that "weighing whether you should reassign your customer service to a third party is complex and causes unease for many business owners", then prices the two sides: $30-$65 per hour for dedicated agents, $1-$2 per minute plus $50-$400+ monthly for answering services, against $6,000+ per in-house rep in equipment and setup.
TDS Global Solutions puts the weight on partner selection, arguing that "outsourcing is not simply about hiring the lowest-cost call center". PartnerHero comes closest to breaking the binary: "Sometimes, you don't have to pick just one. A hybrid model lets you blend in-house and outsourced support for maximum flexibility." That hybrid is still two human models blended; my third option is a tool that makes one of them cheaper.
SupportYourApp's own arithmetic puts five in-house agents at $32,500 a month against five outsourced agents at $12,000, a saving of about 63%. Take the exact figure with a grain of salt, since the outsourcer is the one publishing it.
I'd want one more line in that comparison. Every line in it prices a seat; none of them counts how many of those tickets need one.
A team sizes a BPO contract against total monthly volume, signs seats and a 4-6 week training window to match, and then discovers that the majority of what it just bought bodies for was answerable from the help center plus one order lookup.
You now own a minimum commitment sized against a number nobody examined.
The searchers know something is off. I counted three Reddit threads inside the top ten for this search, sitting alongside five pages published by outsourcers and helpdesk vendors.
When a buying question pulls that much forum traffic into the results, I read it as buyers wanting a source that isn't selling them the answer.
The consensus gets one thing right: the humans you keep are the valuable half of the function. This was never an argument for firing everyone.
One of our customers described AI support as protecting their team, taking the Monday-morning trawl through hundreds of not-always-polite tickets off people who then spend the day on work worth doing. It shields the band you should be least willing to rent out.
The repeat, judgment and presence split
TL;DR: Support volume splits three ways (repeat work, judgment work, presence work) and each band has a different right answer. Picking one model for the whole queue is what makes the decision hard.
Support volume is three kinds of work that happen to arrive through the same inbox, each with a different winner. I call it the repeat/judgment/presence split.
Band | What the work is | Who should own it |
Repeat | Documented answer plus a system lookup | An AI agent in your helpdesk |
Judgment | Discretion, exceptions, retention, complaints | Your own team, in-house |
Presence | Time zones, languages, surges you can't staff | A BPO, scoped to this band only |
Do that split first and you walk into the outsourcing conversation knowing what you're buying.
Repeat work goes to AI
High volume, low variance, answerable from documented knowledge plus a system lookup. Order status, password resets, policy questions, where-is-my-refund, plan and billing queries.
It's the biggest band in almost every queue we see, and boring by definition: the same forty questions, arriving thousands of times.
The answer has to exist in writing somewhere, and the agent has to be able to read the customer's live account state. Neither requirement is headcount. Where both are true, this band stops reaching people.
Across the field, 195 rated AI deployments spanning 38 vendors put the median handling rate at 70%, with the middle half of the field between 56% and 80% (I posted the full spread of that dataset on LinkedIn if you want the range). Those are aggregate, self-reported numbers from vendors publishing their wins, and every vendor defines its own numerator, so treat them as where the field sits.
If you want the difference between the metrics being quoted at you, we've written up what deflection rate measures.
The test for putting a ticket in this band is whether the answer depends on a rule you could write down. "Where's my order" depends on a lookup and a shipping policy, both of which exist in writing or in a system. "Can you make an exception for me" depends on a decision, which doesn't.
Plenty of tickets in this band end in an action. Resending an invoice, updating an address, canceling a subscription inside the window your policy allows.
Whether the agent completes those on its own or drafts them for someone to approve is a setting you choose per action. Either way, how documented the work is puts the ticket in this band.
Judgment work stays in-house
Discretion, policy exceptions, commercial calls, retention saves, the angry customer with a legitimate point. It's a small share of volume carrying a large share of the value. Being wrong here is expensive in a way no rate card captures.
This is the work I'd keep in-house whatever the rate card said: it needs someone who knows the business, carries the authority to make an exception, and will be there next quarter to live with the precedent they just set.
It's also the work that makes a support job worth having. Hand it to a third party and you're renting out the part of the function that compounds.
A rate-per-seat comparison treats all three bands as interchangeable labor, so the judgment band gets priced at the same number as the password resets and gets sent to the same place.
Cost is the weakest reason I'd give for taking repeat volume off your team.
That Monday-morning trawl is what grinds a support team down, and the day gets better when what reaches a person is the work that needed one. Judgment work is what your team should be spending that day on. Outsourcing it doesn't give you that day back.
Decide the boundary in advance, because the escalation rules you write at setup are what keep a retention conversation or a fraud flag away from the automated path in the first place.
And I'd treat the band as a choice about who should own the decision: a policy exception goes to a person because you want a person accountable for it.
Presence work is what's left to outsource
Bodies in a time zone, a language or a channel at a volume you cannot staff yourself: overnight phone lines, a seasonal surge that needs trained people inside two weeks, a language your team doesn't cover.
This is exactly what the consensus lists as the outsourcing pros. Zendesk, SupportYourApp, PartnerHero and the small-business guides all name 24/7 coverage, multilingual reach and elastic scaling as the top reasons to outsource.
The split is their own list with the other two bands taken out of it.
The fourth standard pro, a lower cost per seat, is the one I'd separate out, because it's an argument about the price of absorbing repeat volume and has nothing to do with presence. That argument usually carries most of the weight in a quote, and the split takes the ground out from under it.
Coverage, languages and elasticity survive the split intact. The rate is what justifies seats you may not need.
Presence work is specific enough to size. I'd size it before reading another rate card. A time zone is a set of hours, a language is a share of the inbox, a surge has a start and an end date, and each of those can be counted separately from your total volume.
Outsourcing has historically answered repeat and presence in one contract, because absorbing repeat volume took people. The BPO business case now rests on presence alone, a much smaller slice than the contract you're being quoted for.
And the split moves. Day one is the worst it will ever be: the repeat band grows as coverage grows, in the order I'd build it, from connecting your knowledge, to connecting your user data through an API, to building tasks for the workflows behind the questions.
A multi-year seat commitment fixes your denominator for its term. Your actual repeat band keeps growing.
Why is the outsourcing window closing?
TL;DR: A BPO's whole reason to exist is absorbing the repeat band. On that band an offshore agent now has less context and less access than the AI does, and the outsourcers know it, because they're putting AI agents in front of the queue themselves.
A BPO exists to absorb exactly the work I'd now hand to AI: high-volume, documented, repeatable tickets. That was a people problem for thirty years, so you rented people.
AI does that band now.
In most companies tier-1 is the offshore tier, so when AI takes tier-1 the offshore need shrinks and what escalates goes to tier-2 and tier-3 inside the company, which usually means faster answers on the hard tickets. I'd take that into the internal conversation ahead of any rate card: the same people move up the chain to the work that was queued behind the repeat volume.
On the repeat band I'd take the AI over the outsourced agent.
AI reads the whole help center, every past ticket, and the customer's live account state through an API. An outsourced agent works from a training deck, a limited systems login, and whatever the last handover note said.
Edel Optics was sitting at 20-30% AI resolution on our agent, then connected a User Data API so the agent could see order, delivery, return and tracking information, and resolution went to 75-79% overnight.
Nothing about the product range or the customers changed. What changed was what the agent was allowed to see.
The industry's own 4-6 week ramp to tier-1 proficiency is a measure of how long it takes to load a fraction of that same context into a person, once, per person, every time one leaves.
The outsourcers are putting AI agents in front of the queue themselves, some building and some buying, and I'd read their product pages before signing one of their contracts.
SupportYourApp, which ranks tenth on this exact search, sells a proprietary AI product suite bundled into its service tiers: SupportBrain for agent assist, SupportResponse as an autonomous agent, plus SupportVoice and SupportCRM. Its own marketing for SupportResponse claims "81% of chat tickets resolved without a human agent".
Concentrix, one of the largest CX outsourcers in the world, rolled out pre-built agentic AI in December 2025, built on its own iX Hello platform. The four launch agents are Product Support, Order Status, Appointment Scheduling and Collections. That's the repeat band, itemized by the company selling you seats to cover it.
I'd make the same move in their position.
PartnerHero, whose "which support model is right for you?" guide holds a position-one feature block on this search, was acquired by the AI CX platform Crescendo in October 2024, bringing "over 200 customers and 3,000 human-in-the-loop CX professionals" with it. An AI vendor now owns the company publishing the article that tells you to choose between in-house and outsourcing.
TaskUs announced partnerships with Decagon and Regal in May 2025 and stood up an agentic-AI consulting practice. Its CEO put a number on it that surprised me:
"Agentic AI from Decagon and Regal should enable us to reduce the cost of customer support by 25-50% while significantly improving quality"
That's Bryce Maddock, CEO of TaskUs, in his own press release. The same release describes deployment specialists applying agentic AI "upfront to automate many simple, repetitive customer service functions."
TELUS Digital got there earliest, launching its Fuel iX GenAI platform back in October 2023.
That's five outsourcers publicly committed to automating the band their seat contracts are priced on. You can pay a BPO to put an AI agent in front of your queue, or you can put one there yourself. What I wouldn't do is buy seats for that band from either.
Outsourcing survives all of this. If anything it makes the presence band more important, because that's the band that lasts.
I've made the longer version of that case on LinkedIn: anything a support team does happens on a computer, so with enough access there's no structural reason the automatable share stops where it sits today.
What does this look like in real rollouts?
TL;DR: Across four rollouts the repeat band runs 62-80% of volume, and YouGarden's share alone is 965 hours a month, about six full-time agents.
The repeat band looks different once it's measured. These are our own customers, with the numbers as their case studies report them.
Customer | Helpdesk | Tickets per month | AI resolution | Hours saved per month |
YouGarden | Freshdesk | Around 12,000 | 66%, peaking near 82% | 965 |
TravelJoy | Zendesk | 2,500-2,700 | 80% | 193 |
Edel Optics | Zendesk | 4,000+ | 75-79% | 150 |
Kriptomat | Intercom | Around 1,700 | 62% | 172 |
YouGarden: 66% AI resolution, 965 hours a month
YouGarden is a UK online garden center running support on Freshdesk, handling around 12,000 tickets a month. Their AI resolves 66% of them, peaking near 82%, at 78% AI CSAT across 11,785 tickets.
The number I'd take into a BPO negotiation is 965 hours saved a month, which at five minutes a ticket is about six full-time agents. That's the repeat band expressed in the exact unit a BPO contract is priced in.
They got there by connecting the Freshdesk help center plus 3,000+ pages from their own sites, building a custom User Data API for recent orders, tracking and delivery, and running in Freshdesk Notes mode for a month before flipping to direct replies. AI tagging classifies each incoming message so reporting stays consistent and so they can decide which categories the AI answers at all.
They came to us after switching off DigitalGenius, following a tender against Freddy AI and Intercom Fin.
TravelJoy: 80% AI resolution with no night shift
TravelJoy is a platform for travel advisors, on Zendesk, at roughly 2,500-2,700 tickets a month. Their AI resolves 80%, up from 24% on Zendesk's own AI, at 86% AI CSAT and 193 hours saved a month.
Their result is the direct answer I'd give to the 24/7 pro on every outsourcing list. They cover their time zones on availability, without staffing an overnight shift.
The setup mixes two levers: the AI Tagging Zendesk app auto-classifies incoming tickets and replaced manual contact-reason tagging, while Handover guidance rules decide what goes to a person. On email they run the AI as a copilot, replying to only the first message so the team keeps the thread.
Edel Optics: 53 countries at 92% AI CSAT
Edel Optics is a European eyewear retailer selling across 53 countries, on Zendesk, at 4,000+ tickets a month. Our agent resolves 75-79% of them, with a 92% AI CSAT across 4,067 tickets and 150 hours saved a month.
Multilingual coverage is the other big reason to go offshore. This rollout covers 53 countries at 92% CSAT. Language is auto-detected per message, defaulting to German and switching for everyone else.
Two controls keep the judgment band away from the AI: AI Tagging classifies the incoming message and blocks AI replies on categories they don't want it touching (faulty items, for instance), and an "I don't know" handover sends anything unanswerable straight to a person.
Kriptomat: 62% AI resolution through the surges
Kriptomat is an EU-licensed cryptocurrency platform on Intercom, at around 1,700 tickets a month. Their AI resolves 62%, up from about 50% at go-live, saving 172 hours a month at ten minutes a ticket.
They run on Guidance rules, with Handover configured for legal and fraud topics, so the judgment band stays in-house by design and never has to be caught after the fact.
They evaluated Intercom Fin before us and turned it down as uneconomical on the per-resolution pricing it carried at the time.
On the seasonal-surge argument that every BPO page sells, their support team's own experience is the reply:
"Personally, I'm a big fan of the direct integration with our pre-existing help articles, and how easy it is to re-train the agent when it's providing outdated information. It has helped our team out immeasurably especially during heavy inquiry surges over the past few months!!"
That's Hannah DiBella on Kriptomat's support team.
Four rollouts across four industries and three helpdesks: 62%, 66%, 75-79%, 80%. That's the range of what the repeat band turns out to be once it's counted, with the working shown in each of our case studies.
What to do this week
TL;DR: Before you price a single seat, tag one month of tickets into the three bands. The split is usually the whole answer.
None of this commits you to anything. It exists to produce the number the quote in front of you is missing: how much of your queue needs a person.
Do them before you price a seat, because everything you sign afterwards is sized against whatever you believed at this point.
- Tag last month's tickets into repeat, judgment and presence. Export the month, sample a few hundred if the volume is large, and sort them into the three bands by hand. Budget about two hours. You come out of it with the denominator the BPO quote was missing, and with a defensible number for the internal conversation. Three of our four rollouts used tag-based classification for exactly this before deciding what the AI should answer.
- Re-quote the outsourcing contract against the presence band only. Send the same RFP back with the volume figure replaced by your presence number and ask what changes. Thirty minutes and one email. The answers you get back tell you how much of the original quote was priced against work that no longer needs a seat.
- Check whether your repeat band is actually written down. Pull ten of your most common repeat questions and try to find the answer in your help center. If it isn't in writing anywhere, no delivery model fixes that. It's the same blocker for a BPO: the outsourcing industry's own onboarding guidance, like rethinkCX's, says a searchable, current knowledge base covering product info, FAQs, common resolutions and policies has to exist before onboarding begins. Luckily you don't have to write it all by hand if you're starting from nothing: Train on Historic Tickets auto-drafts starter articles from past resolved tickets (5,000 by default, more on request) so a team with no help center can get to a first draft in an afternoon.
- Trial the repeat band for two weeks. Before signing anything longer than a month, run the AI in internal-notes mode inside your helpdesk so it drafts every reply for your team without a customer seeing it, then compare those drafts against what your agents actually sent. Notes mode is a deployment mode you choose, and you switch to direct replies when the drafts stop needing edits. YouGarden ran a month of it before going live and Edel Optics started there too, and our AI Copilot Chrome Extension does the same job across any of our integrated helpdesks. Then compare the two commitments in front of you: mid-market outsourcing specialists start at roughly 10-25 seats and the largest providers at 50+, against a 30-day free trial with all features unlocked, unlimited tickets and no card. If you want the longer version of this sequence, we've mapped it out as a 30-60-90 day plan.
How do I get AI to run the three-band split for me?
Step one is paste-and-classify work, which is what I'd hand to an LLM before spending my own afternoon on it. Copy this into ChatGPT or Claude with your own export; it only sees what you paste, so I'd treat the output as a head start on the two-hour pass and spot-check it against the raw tickets.
I run a customer support team and I'm about to price an outsourcing contract. Classify my tickets against the repeat/judgment/presence split:
1. Repeat — the answer is written down somewhere (help center, policy page, past tickets) or reachable by a system lookup (order status, delivery, account state). No discretion required.
2. Judgment — the reply depends on a decision rather than a rule: policy exceptions, commercial calls, retention saves, a complaint with a legitimate point.
3. Presence — the ticket needs a person in a time zone, channel or language we don't currently cover, however simple the question is.
My ticket export (subject lines or first messages): [paste 100-300 tickets]
What our help center and connected systems already cover: [list your knowledge sources and any customer-data API or integrations]
Our coverage today: [e.g. 5 agents, weekdays 9-6 UK, English only]
For each ticket, assign one band and give a one-line reason. Where you can't tell from what I've pasted whether the answer exists in writing, write "unverified, check knowledge coverage" instead of guessing.
Output a table with ticket, band and reason, then the percentage split across the three bands. Finish with my presence band expressed two ways: tickets per month, and the coverage hours it falls in. That second number is the only one an outsourcing quote should be sized against.When does outsourcing still win?
TL;DR: Four situations where a BPO beats both in-house and AI, and one where nothing works until you fix the knowledge.
All of that is an argument about the repeat band. In a few situations a BPO is still my recommendation.
Phone-heavy support at volume is a presence-band job and BPOs are built for it. The entire industry builds its cost structure, staffing model and training pipeline around voice floors.
It's also the hardest work in the sector to keep staffed: CallForce cites ContactBabel putting offshore voice floors at 45-60% annual turnover against a 30-45% industry average from QATC, with voice training running 3-6 weeks against 2-6 for non-voice. Running that anyway, at scale, with a bench, is what I'd pay an outsourcer for.
Regulated work needs a licensed or certified human. Financial advice, clinical triage, anything where the constraint is a qualification attached to a person.
No delivery model dissolves that, and the structural answer is the same whether the licensed human sits in your office or your partner's. Kriptomat's legal and fraud handover rules are the in-house version of the same decision.
Seasonal spikes where you need trained bodies inside two weeks are where BPOs mobilize faster than you can: SupportYourApp puts outsourced setup at "less than one month" against weeks-to-months to build the equivalent in-house, and some providers advertise a team operational in 7-10 business days.
The caveat that keeps it from being a blank check comes from the outsourcing side too: ramp to tier-1 proficiency runs 4-6 weeks for routine work and 8-12 weeks for complex or regulated work, and rethinkCX notes that "compressed timelines (under 4 weeks) usually produce agents who underperform during the first 90 days." Would you rather have an AI absorbing that spike, or hire, train and then release fifteen people inside a six-week window?
Languages with thin model coverage are the fourth case. For the major languages this has flipped, and Edel Optics running 53 countries at 92% AI CSAT is what that flip looks like.
But thin coverage is a real category, and where quality drops below your bar, a native speaker in the queue is still the answer.
No documented knowledge is the last one. Where a team has nothing written down, AI has nothing to work from, and a BPO's training window is the answer, because a person can be told things a retrieval system cannot read.
"Nothing written down" is a smaller category than it sounds, because most teams we onboard have answered the questions thousands of times already, in tickets. Train on Historic Tickets turns those resolved tickets into starter articles, which is the first draft of a help center you already own.
If that's your situation, run it before you price a training window. An afternoon of editing tells you whether the knowledge was missing or just never written up.
The last piece is what you're buying when you sign. Helpware, itself an outsourcer, puts per-contact pricing at $1-$7 with an industry average near $4, and reports that contract minimums run 50+ seats at the largest providers (Teleperformance, Concentrix, TTEC) and roughly 10-25 seats at mid-market specialists, with minimums that "often require paying for 25+ agents even if only using 15".
Add setup fees of $50,000-$200,000 at the enterprise end and around $5,000-$25,000 at mid-market, a 4-6 week ramp, and 30-45% annual attrition that resets a share of that ramp every year. All of it is priced against the whole queue, repeat band included.
Helpware's own guide argues the cost of a quality failure usually outruns the rate saving inside a year.
The takeaway
TL;DR: Split the queue before you pick a model. Repeat work goes to AI, judgment stays in-house, presence is what's left to outsource.
The in-house versus outsourcing debate is a fight about who staffs the queue, held without anyone asking how much of the queue needs staffing. Answer that first and the argument gets smaller and easier. In our own rollouts the repeat band runs 62-80% of the queue.
Repeat work goes to AI. Judgment work stays in-house, because that's the band where being wrong is expensive and where your team's experience compounds.
Presence work, the time zones, the channels and the surges you can't cover yourself, is what's left to outsource. I'd buy that at a fair price all day, once it's the only thing in the contract.
If you do one thing this week, tag a month of tickets into the three bands. Sorting them takes an afternoon and puts a percentage on your repeat band.
Once you've decided what stays in-house, routing a single ticket across AI, AI-assisted and human handling inside that team is a separate exercise, and we've written it up as a decision framework alongside the longer piece on building an AI-first support team.
If you want to see what the repeat band looks like in a business your size before committing to anything, YouGarden's numbers are a reasonable place to start, and our own trial runs on your own tickets.
FAQs
What are the pros and cons of outsourcing customer service?
The pros are real: a lower cost per seat, specialist CX expertise, fast scalability, 24/7 and multilingual coverage without hiring, and less recruitment and training load on you. The cons are equally well documented: less control over brand voice, a learning curve on your product, security exposure, and limited effectiveness on complex or unusual requests.
The thing missing from both lists is the denominator: most of the volume being priced in that comparison is documented repeat work that no longer needs a person, and the published field median for AI handling sits at 70%.
Only the scope changes. Split the queue into repeat, judgment and presence work first, and I'd only apply the pros and cons to the last of the three.
When should you outsource customer service rather than hire?
Outsource when the need is presence: a time zone you can't staff, a channel you can't cover, a language with thin coverage, or a surge with a hard deadline. That's where a BPO's lead time is the advantage, with SupportYourApp putting outsourced setup at under a month against weeks-to-months to build the same capability in-house, and rethinkCX putting ramp to tier-1 proficiency at 4-6 weeks.
Hire in-house when the work is judgment: discretion, exceptions, retention, anything where being wrong is expensive. I'd push back on treating overnight coverage as automatically a staffing problem, because in our own rollouts it's usually the band that solves itself first: TravelJoy covers its time zones on availability, without staffing an overnight shift.
If the need is neither presence nor judgment, it's repeat work, and staffing is the wrong lever.
What is business process outsourcing in customer service?
Business process outsourcing means handing an entire business function to a third-party provider that runs it for you; in customer service, that's a partner staffing and managing the agents who answer your tickets, calls and chats. Zendesk's definition covers it: "customer service outsourcing is when businesses hire third-party companies to handle their customer service."
Commercially you're buying capacity, and it's sold by the hour, by the seat or FTE, by the contact or ticket, occasionally on an outcome basis, and almost always with a minimum you cannot drop below. Every one of those units is measured against your total volume. I'd test that assumption before you sign.
Is outsourcing customer service to the Philippines still the default?
It's still the largest single destination and it isn't going anywhere, with decades of accumulated CX capability and a workforce built around English-language support.
The industry's own outlook has changed: coverage of the Philippine IT-BPM sector's revised targets reports the association's earlier growth goals being cut, with rapid AI adoption named as one of the drivers. If you're sending the repeat band offshore, you're paying for headcount to do work the destination industry is automating itself.
Should a small business outsource customer service?
The size argument for outsourcing is weaker than it looks in both directions. Across 195 rated AI deployments, the median handling rate moves very little with company size, running 72% at 1-49 employees and 75% at 5,000+, so being small doesn't cost you the repeat band.
Meanwhile the minimums bite hardest at small scale: mid-market outsourcing specialists start at roughly 10-25 seats, or 5-25 agents depending on the provider, and Helpware notes contracts that "often require paying for 25+ agents even if only using 15". The smallest teams we work with hold much the same repeat share as the largest ones, while the contract minimums are written for companies several times their size.
Tag your queue first. A small business with a documented repeat band usually doesn't have enough presence work left to hit a minimum.
How do you actually outsource customer support?
Document the knowledge first, because the outsourcing industry's own guidance says the knowledge base has to exist before onboarding begins. Then I'd scope the contract to the band that needs bodies.
Expect a 4-6 week ramp to tier-1 proficiency and 8-12 weeks for complex or regulated work, and plan for 30-45% annual attrition resetting part of that ramp every year. Read the contract for the levers that don't show up in the headline rate, and settle all four before you argue about the rate itself: setup fees, ramp-up fees for adding agents quickly, after-hours premiums, and penalties for early downsizing.
How much does outsourcing customer service cost?
Broadly, Helpware puts hourly rates from $7-$16 offshore up to $28-$42 onshore in the US, and a second outsourcing-industry pricing guide puts per-FTE monthly pricing at $1,200-$4,000 and per-ticket pricing at $1-$5, both sitting on top of setup fees and seat minimums. Treat those as vendor-published ranges and get your own quotes, because region, channel and complexity move them more than any published table suggests.
The number that changes your bill most is how many tickets you're paying anyone to touch. I'd chase that before negotiating a rate.
Across our own rollouts the share that stopped reaching a person moved the total far harder than any per-hour figure in that spread. We've written separately on reducing support costs and on the hidden costs of AI customer service.
