Start with the shape of the work, not the vendor
The single biggest predictor of whether an outsourcing relationship works is whether the engagement model fits the work. Get this wrong and no amount of vendor quality saves you.
Well-defined, high-volume, repeatable work — tier-one support, bookkeeping, QA execution — suits an output or per-ticket model, and suits a larger provider with deep process maturity.
Ambiguous, judgement-heavy work — product engineering, brand design, growth marketing — needs a dedicated team model where the same people accumulate context over months. Buying that work by the ticket produces churn and mediocrity, regardless of who you buy it from.
If a vendor pitches you the same model for both, they are selling their operating preference rather than solving your problem.
Interrogate bilingual capability specifically
"Bilingual" is the most oversold word in the nearshore market. Nearly every Mexican provider claims it; the depth varies enormously.
The distinction that matters is whether English proficiency is concentrated in a client-facing layer — account managers, team leads — or present throughout the delivery team. The first arrangement looks fine in a sales cycle and breaks down the first time you need to talk directly to the person doing the work.
- Ask to speak with delivery staff, not just leadership, unscripted.
- Ask what share of the team could handle a live customer escalation in English.
- If Spanish-language volume matters to you, ask the same question in reverse.
- Ask how language proficiency is assessed at hire, and by whom.
Ask about retention, not recruiting
Every provider can fill a seat. The cost you actually feel is a strong person leaving eight months in and taking undocumented context with them.
Attrition in outsourcing is normal and no honest vendor will claim zero. What separates good providers is whether they measure it, whether they will tell you the number, and whether their documentation is good enough that a replacement is productive quickly.
- Ask for annualised attrition on accounts like yours, not company-wide.
- Ask what happens to your knowledge base when someone leaves.
- Ask how long a replacement typically takes to reach full productivity.
- Ask whether you get a say in replacements.
Find out who owns quality
If quality assurance reports to the same person who owns delivery margins, quality scores drift toward whatever makes delivery look healthy. This is not malice, it is organisational gravity.
The providers worth shortlisting either separate those reporting lines or hand you the raw data and let you draw your own conclusions.
- Ask who the QA function reports to.
- Ask to see an unedited sample of recent quality reviews.
- Ask which metrics they refuse to negotiate on, and why.
Understand the pricing model before the price
Hourly, per-seat, per-ticket, and outcome-based pricing each create different incentives, and the cheapest headline rate frequently produces the highest total cost.
Per-ticket pricing rewards closing tickets, which is not the same as resolving problems. Hourly pricing rewards hours. Dedicated-seat pricing is the most predictable and gives you the least protection if the seat underperforms.
None of these is wrong. What matters is that you can name the incentive your contract creates, and that you have a metric that catches it when it goes bad.
- Ask what behaviour the pricing model rewards, and listen for a straight answer.
- Ask what is explicitly out of scope.
- Ask how rates change as volume grows, and as it shrinks.
Negotiate the exit while you are still friends
Exit terms are trivially easy to agree before signing and nearly impossible to negotiate once a relationship is failing. Yet most buyers spend their leverage on the day rate instead.
The specific things worth fixing in writing: notice period on both sides, what knowledge transfer includes and how long it lasts, who owns the documentation and recordings, and whether you can hire the people who worked on your account.
That last one is where vendors push back hardest. A conversion clause with a defined fee is reasonable; an outright ban on hiring people who have spent a year inside your business is a red flag about how they view the relationship.
How to run the actual evaluation
Shortlist three providers, not eight. Beyond three, the process consumes more of your time than the decision quality justifies.
Run a paid pilot with real work rather than a proof of concept. Pilots reveal how a provider behaves when something goes wrong, which is the only thing you actually need to learn.
Give every provider the same brief and the same data, and pay attention to which ones ask the best questions rather than which ones respond fastest. Speed of response correlates with sales capacity; quality of questions correlates with delivery capacity.