Buyer resource · Updated October 2026
The support outsourcing
RFP kit
Most support RFPs are built from a feature list, and feature lists do not predict how a support operation actually performs eighteen months in. This is the question set that does, with the reason each question matters and what a weak answer sounds like.
71 questions across eight areas, a scoring sheet, and the answers that should worry you. Written by a vendor, which is worth knowing, so it includes the section most vendors leave out: the questions we find hard to answer ourselves.
Why most RFPs fail to predict anything
A support RFP usually asks what channels a vendor covers, which helpdesk tools they know, how many seats they can staff and what the hourly rate is. Every serious vendor answers all four the same way, because all four are table stakes. The document sorts nobody.
What actually varies between vendors, and what determines whether the engagement works, is harder to ask about: who the agents report to, how long they stay, what happens when the answer to a ticket does not exist yet, and what it costs you to leave.
The published data makes the case for asking better questions.
15%
Median first contact resolution on ticketing, against 32% in the top tier
Freshworks, 32,000+ companies, 1.2bn tickets, 2024 data
91%
Of support leaders report executive pressure to deploy AI. 10% call their deployment mature
Gartner, 321 leaders, fielded Oct 2025
76.7
US customer satisfaction index, unchanged since 2013
ACSI, ~200,000 annual interviews, Q1 2026
The single most useful question in this document
Ask every vendor for the first contact resolution rate on ticketing, not on chat, for accounts of your size, and ask how it is calculated. Most will quote a blended number or a chat number, because chat flatters. The same dataset puts median first contact resolution at 81.88% on conversational channels and 15% on ticketing. That is not a rounding difference, it is two different jobs, and a vendor that cannot separate them does not measure its own work closely enough to improve it.
The eight questions that decide it
If you only ask eight, ask these. Everything in the full bank is an elaboration of one of them.
01
Who do the agents report to, and who writes their review?
This is the structural question the rest follow from. In a managed model the vendor hires, supervises and scores them, and you get a point of contact. In an embedded model they sit in your reporting line and your managers brief them. Both work. They cost you different things, and a vendor that blurs the answer is selling you the warm version of one while delivering the other.
02
What is your agent attrition rate on accounts of our size, over the last twelve months?
Attrition is the number that quietly determines everything else. Every departure resets product knowledge to zero. Ask for the figure by account size, because a blended company-wide number hides the fact that small accounts usually churn hardest.
03
What is your first contact resolution rate on ticketing, and how do you calculate it?
See above. The calculation matters as much as the number: some vendors count a ticket closed on first touch, others count it only if the customer did not come back within a set window.
04
Who are the named individuals on our account, and can we interview them?
If the answer is that staffing is assigned at go-live from a pool, you are buying capacity, not a team. That is a legitimate purchase and sometimes the right one, but it is not what most vendors imply.
05
What happens in week one if a ticket needs an answer nobody has written down yet?
Every vendor handles known questions well. The difference shows when somebody has to reason about your product. Listen for whether the answer involves your engineers, and how often.
06
How long from signature to a team answering real tickets, and what do you need from us in that window?
Go-live claims are cheap. The second half of the question is the real one, because most slipped launches are blamed on the client not providing access, documentation or decisions fast enough.
07
What does the exit look like?
Notice period, data handover format, whether macros and documentation written by their agents are yours, and what happens to the ticket history. Ask before you sign, because the answer never improves later.
08
What would make you tell us we are not a good fit?
The most revealing question in the set. A vendor with a real model can describe the customer it is wrong for in one sentence. A vendor that says it can serve anyone is telling you it has no model.
The full question bank
Sixty-three questions across eight areas, after the eight above. Send the whole thing or cut it to the sections you care about. The why it matters line under each one is there so whoever reads the responses knows what a good answer looks like.
1 · The people
01
What is your agent attrition rate, by account size, for the last twelve months?
The number that resets product knowledge.
02
How many accounts does a single agent work on?
Dedicated, shared or pooled. Everything follows from this.
03
How many people does a team lead supervise?
A span above about fifteen means nobody notices an agent struggling until a customer does.
04
What is the average tenure of agents currently on accounts like ours?
Different from attrition and harder to game.
05
Who writes the agent’s performance review, and what is in it?
If the vendor writes it alone, your priorities are an input at best.
06
Can we interview the named people before they start?
A yes means they exist. A no usually means staffing happens at go-live.
07
What is your hiring acceptance rate and what does the process involve?
Useful only with the process attached. A low rate with no process behind it is a statistic, not a filter.
08
What happens when somebody on our team leaves?
Ask for the replacement timeline and who covers the gap.
09
Are agents dedicated to our time zone, or following the sun?
Follow-the-sun means handovers, and handovers are where context dies.
2 · Coverage and continuity
10
What hours are genuinely staffed, as opposed to monitored?
Monitored means somebody is paged. It is not the same as answered.
11
What is your coverage during public holidays in the delivery location?
A delivery country’s holiday calendar is not yours.
12
What happened the last time a delivery site lost power or connectivity?
Everybody has a continuity plan. Ask what actually happened.
13
How do you handle a sudden volume spike, and how fast?
Get a number in days, not an assurance.
14
What is the escalation path at 3am on a Sunday?
Name and role, not a process diagram.
15
How much notice do we need to give to scale up or down?
Flexibility costs money somewhere. Find out where.
3 · Tools and integration
16
Do agents work in our helpdesk, or yours with a sync?
A sync means two sources of truth and an argument about which is right.
17
Who owns the macros, saved replies and internal documentation your agents write?
This is a real negotiation point and it is almost never raised until exit.
18
Will agents be in our Slack or Teams, and under what access?
Being in the room is most of what embedded means in practice.
19
What access do you need to our systems, and what is the minimum that would work?
A vendor that has thought about least privilege has thought about security.
20
Can we see a live dashboard of our own queue, or do we get a report?
A weekly report is a vendor telling you how it went.
21
What AI tooling sits between the customer and the agent, and can we turn it off?
Deflection and resolution are different things. See the red flags section.
22
Who pays for helpdesk seats and tooling?
Small, but it surfaces late and annoys everyone.
4 · Quality and measurement
23
What is your first contact resolution rate on ticketing, separately from chat?
The question most vendors answer with a blended figure.
24
How is a quality score calculated, and who calibrates the scorers?
An uncalibrated QA score measures the scorer.
25
What percentage of tickets are reviewed, and by whom?
Sampling rate matters more than the score.
26
Can we see a real QA scorecard from a live account, redacted?
A template tells you nothing. A completed one tells you what they look at.
27
What is your median and 90th percentile first response time, by channel?
Medians hide the tail, and the tail is what customers complain about.
28
How do you measure whether the customer’s problem was actually solved?
Separate from whether the ticket was closed.
29
What do you do when a metric is missed for a month?
Listen for a mechanism, not an apology.
30
Will you report on tickets your team could not resolve, and why?
The most useful report nobody asks for.
31
Do you survey customers yourself, or use ours?
Vendor-run surveys sent by the vendor tend to score better.
5 · Security and compliance
32
Which certifications do you hold, and can we see the current report rather than the badge?
A badge on a website is not a report. Ask for the date and the scope.
33
Where does our data physically sit, and where do agents physically sit?
These are different questions and both matter.
34
Will you sign our data processing agreement, or do you require yours?
Find out before legal does.
35
What is logged about agent access to customer records, and can we audit it?
Whose log, kept how long, and can you read it.
36
What happens to our data when the contract ends?
Deletion timeline, format of what you get back, who confirms it.
37
Are agents working from an office, from home, or both?
Not a trick question. It changes the control set entirely.
38
Has the company had a reportable security incident in the last three years?
The answer matters less than whether they answer it straight.
39
Do you subcontract any part of the delivery?
If yes, every question in this section applies again, one level down.
6 · Commercials and exit
40
What exactly is in the rate, and what is billed separately?
Tooling, training hours, overtime, holiday cover, reporting.
41
What is the minimum commitment, and what is the notice period?
Then ask what happens if you need to leave sooner.
42
How does the price change if volume halves?
Most contracts are built for growth only.
43
Is there a charge for replacing an agent who leaves?
There should not be, but ask.
44
What happens to the rate at renewal, and is there a cap?
An uncapped renewal is a price increase you have already agreed to.
45
Who owns the training material we fund?
Related to the macros question and equally overlooked.
46
At exit, what do we get and in what format?
Ticket history, documentation, QA records, the knowledge base.
47
Will you support a transition to another vendor, and at what cost?
The answer tells you how confident they are that you will not need to.
7 · Onboarding and the first ninety days
48
Walk us through the first thirty days, by week.
Vague answers here predict a vague launch.
49
What do you need from us, by when, and what happens if it is late?
The honest version of why launches slip.
50
Who writes the initial documentation, you or us?
A vendor that writes it has to understand the product to do it.
51
How many of our hours does onboarding consume?
Nobody budgets for this and everybody pays it.
52
When does the team start handling tickets unsupervised?
Ask for the criteria, not the date.
53
What does success look like at day 90, in numbers?
Get it in writing before you sign, not after.
54
What is the most common reason your launches slip?
A straight answer here is worth three polished ones elsewhere.
55
Can we speak to a client who launched in the last six months?
References are always flattering. Recent ones are at least current.
8 · The ones that are easy to forget
56
What languages are genuinely native, as opposed to supported?
Supported can mean a translation layer.
57
How do you handle a customer who is abusive to an agent?
The answer says a lot about how the vendor treats its own people.
58
What is your policy on agents working a second job?
Common in some delivery markets, and relevant to attention and security.
59
Who do we call when something is going wrong and the normal contact is not responding?
Get a name.
60
How often will we meet, and who attends from your side?
Weekly with the account manager is different from monthly with the team lead.
61
What have you learned from an account that did not work out?
A vendor with no failures has either no history or no candour.
62
Can we start with a smaller scope and expand?
A no is not automatically bad, but it should have a reason.
63
What are you better at than the vendors we are also talking to?
Ask it directly. The shape of the answer is the signal.
Keep reading
The remaining 63 questions, the scoring sheet and the red flags
Plus the section most vendors leave out: the five questions we find hard to answer about ourselves, answered.
One email. No sequence, no call unless you ask for one.
Scoring the answers
Scoring an RFP on a 1 to 5 scale across seventy-one questions produces a number nobody trusts. This is coarser and more useful: three buckets, and one rule.
| Bucket | What it means | What to do |
|---|---|---|
| Answered | A specific, checkable answer. A number, a name, a date, a document. | Score it. This is the only bucket that counts. |
| Described | A process explained without a figure attached. “We have a rigorous QA program.” | Ask once more for the number. If it does not come, treat it as Not answered. |
| Not answered | Deflected, reframed, or answered with a different question’s answer. | Count these. The count is more predictive than any score. |
| The eight that decide it | |||
|---|---|---|---|
| 01Who do the agents report to, and who writes their review? | |||
| 02Agent attrition on accounts our size, last 12 months? | |||
| 03First contact resolution on ticketing, and how is it calculated? | |||
| 04Who are the named individuals, and can we interview them? | |||
| 05Week one: a ticket needs an answer nobody has written down. Then what? | |||
| 06Signature to answering real tickets, and what do you need from us? | |||
| 07What does the exit look like? | |||
| 08What would make you tell us we are not a good fit? | |||
| Miss score, out of 8 · a vague answer counts half |
–
mark the answers above
|
–
mark the answers above
|
–
mark the answers above
|
✓ a specific checkable answer ·
~ a process described with no figure ·
✕ deflected or not answered
Fill in the grid as the responses come back. The score shown is the
count of questions each vendor did not answer, because that count predicts far more than any weighted total.
Nothing is sent anywhere and nothing is stored.
count of questions each vendor did not answer, because that count predicts far more than any weighted total.
Nothing is sent anywhere and nothing is stored.
The one rule
Count the Not answered bucket per vendor and compare the counts, not the scores. Every vendor scores well on the questions it chose to answer. The questions a vendor avoids are the ones where it is weakest, and that pattern is visible across seventy-one questions in a way it never is across eight.
Weighting, if you must
If you need a weighted score for a procurement process, weight by how expensive the mistake is to correct later. Staffing and exit terms are close to impossible to fix mid-contract. Tooling and reporting are fixable in a quarter.
| Area | Suggested weight | Why |
|---|---|---|
| The people | 30% | Cannot be fixed after signature without starting again |
| Commercials and exit | 20% | Fixed at signature, renegotiated only from a weak position |
| Quality and measurement | 15% | Changeable, but only if the vendor already measures |
| Security and compliance | 15% | Usually pass or fail rather than scored |
| Coverage and continuity | 10% | Negotiable during the term |
| Onboarding | 5% | Painful once, then over |
| Tools and integration | 5% | The easiest thing on this list to change |
Answers that should worry you
Not lies. Ordinary, reasonable-sounding answers that reliably predict a problem.
| What you hear | What it usually means |
|---|---|
| “Our CSAT is 95%.” | A chat number, or a blended one. The median company scores 95.15% on conversational channels and 88.00% on ticketing. Ask which this is. |
| “We can staff that in two weeks.” with no questions asked first | They have not understood the product yet. A vendor that quotes a timeline before asking what the tickets are about is quoting a seat, not a team. |
| “Our AI deflects 60% of tickets.” | Deflection is not resolution. A deflected ticket can be a customer who gave up. Ask what happened to the deflected 60% and how many came back. |
| “We work with companies of all sizes.” | No model. Every real delivery model fits a band and breaks outside it. |
| “Attrition is in line with industry average.” | A refusal with a sentence around it. Ask for the number for accounts your size. |
| “We will assign your team at go-live.” | You are buying capacity. Legitimate, but price it as capacity. |
| “That is covered in the MSA.” | It may well be. Ask them to read you the clause. |
| Any answer to ‘what would make us a bad fit’ that is actually a pitch | The clearest signal in the whole process. |
Written by a vendor
The questions we find hard
This document was written by Touch Support, which is a company you might be putting through this process. Leaving out the questions that are awkward for us would make the rest of it worth less, so here they are, with our answers.
01
What if our procurement screens on a certificate?
Raise it in the first conversation rather than the fourth, with every vendor on your list, us included, and ask for the current report rather than the badge. If a specific certificate is a hard requirement, it can decide your shortlist on its own, and nobody should spend time finding that out late. No vendor holds HIPAA certification, because it does not exist.
02
What is your agent attrition rate?
We do not publish one, and neither do most of our competitors. We will give the figure for accounts of your size on a call. Treat a vendor that publishes an attrition number as having done something unusual and good.
03
Your model costs us management time. Why is that our problem?
It is a real cost and we would rather price it honestly. Named agents in your reporting line only work if somebody on your side briefs and reviews them. If nobody has the time, a managed provider will deliver a better result than an embedded team nobody manages, and we will say so.
04
You are smaller than TaskUs or Helpware. What breaks at scale?
We are built for 100 to 500 employee companies and publish that as a number. Above roughly 500 seats of support demand the economics and the management model both favor a larger provider.
05
What do you not do?
Content moderation, trust and safety, and sales development. Peak Support and several others run those as named service lines and we do not offer them at all.
Common questions about running the process
How many vendors should we put through this?
Three to five. Below three you have no comparison and no leverage. Above five the responses stop being read properly, and a question bank this size only works if somebody actually reads the answers.
Should we send all seventy-one questions?
Send the sections that match what you are buying. Most buyers send the eight key questions plus two or three sections in full. Sending everything to five vendors creates several hundred pages that nobody reads, which defeats the point.
Is an RFP even the right format?
For support outsourcing, often not. A structured conversation against the same question set, with one person taking notes, surfaces more than a written response that a bid team has polished. Use the bank as an interview script if that suits you better.
What should we ask references?
Only two things that matter: what went wrong and how it was handled, and whether the people who started on the account are still on it. Everything else will be positive by selection.
How long should this take?
Four to six weeks from first conversation to signature is normal and healthy. Faster usually means the diligence was skipped rather than efficient, and the time comes out of the first ninety days instead.
Where the figures come from
Every figure on this page is from a published source with its sample size and data year stated. Freshworks Customer Service Benchmark Report 2025, 32,000+ companies and 1.2 billion tickets, 2024 data. Gartner, 321 customer service leaders, fielded October 2025. ACSI National Index Q1 2026, approximately 200,000 annual interviews. SQM Group, 500+ North American contact centers. The full set of 28 sources, with links, is in our benchmark report.
Want a second opinion on the answers you get back?
We will read another vendor’s responses with you and tell you what we would ask next. No charge, and no obligation to include us in the process.
