Voice of the Customer Program: How to Build One for B2B SaaS

Key Takeaways

• A voice of the customer program is a system that turns feedback into decisions and closes the loop.
• Every support ticket already carries the account, its ARR, and its renewal date, so attribution costs nothing.
• Below roughly 1,000 customers, NPS and CSAT produce numbers with no statistical base.
• Read every ticket twice: once to answer the customer, once to pull out the signal.
• Helply runs that second read on every conversation at $1 per ticket, with unlimited seats.

A voice of the customer program is a repeatable system for collecting customer feedback and analyzing it for patterns. Each finding is routed to a named owner, and the customer is told what changed.

Voice of the customer is the raw input. The VoC program is the loop around it.

The term predates software. The Association for Supply Chain Management defines voice of the customer as actual customer descriptions of the functions and features they want. The discipline was built to capture what customers say in their own words, before anyone converts it into a score.

Most VoC programs stall at the analysis step. Customer data gets collected, charted, and presented, and nothing moves. A successful VoC program is measured by decisions made, not dashboards built.

Direct, Indirect, and Inferred: The Three Types of VoC Data

Every voice of the customer methodology sorts customer input into three types.

• Direct feedback. Customer surveys you send and answers you request: NPS, CSAT, customer effort score, QBR notes, and interviews. You control the questions and the timing.
• Indirect feedback. Opinions customers share unprompted: G2 reviews, community threads, Slack Connect messages, and competitor mentions dropped mid-conversation. You control neither the timing nor the framing.
• Inferred feedback. Customer behavior read as a signal: product usage drop-off, ticket volume climbing on one account, resolution times stretching for a single plan tier.

A B2B software company has thin volume in the first category and deep volume in the other two. Your customer base is small enough that direct surveys return single-digit responses. It is also busy enough that indirect and inferred signals arrive every working day.

Build the program where the volume is.

Why Survey-Based VoC Breaks Below 1,000 Customers

Survey-first programs fail at B2B account counts for three reasons. None of them are fixable with a better survey tool.

The sample is too small to mean anything. An NPS score from 40 responses is an anecdote with a decimal point. Flip three detractors to promoters and it swings 15 points.

The response rate works against you. Your respondents are named business contacts with their own inboxes and their own quarters. A survey request competes with everything else you send them, and it usually loses.

The cadence is slower than the risk. On a quarterly cycle, the average event waits about 45 days for the next survey. If that event threatened a renewal six weeks out, the insight arrives too late.

The category was built for a different problem. Gartner's March 2026 Magic Quadrant for Voice of the Customer Platforms named four Leaders: Qualtrics, Medallia, Sprinklr, and Press Ganey Forsta.

These are strong customer intelligence platforms, designed to measure thousands of respondents across many touchpoints. That is a real job. It is not the job of a 12-person software company with 200 accounts.

Surveys are not useless. They are the supplement, and the last row is the honest split. Ticket-based VoC cannot see the account that quietly stopped logging in, which is why step six pairs the queue with usage data.

Your Ticket Queue Is Already a Voice of the Customer Program

Every support ticket arrives with something a survey response never has: identity. You know the account, the ARR, the renewal date, the plan, and the last four things that went wrong. Attribution is free.

That changes what a customer support interaction is worth. A survey tells you a score moved. A ticket tells you which account moved it, and how much revenue is attached to the answer.

The support conversation is also becoming the main place customers say anything at all. Gartner research published in March 2026 found that 67% of B2B buyers prefer a rep-free experience. A year earlier the same figure was 61%.

The 2026 number comes from 646 B2B buyers surveyed in August and September 2025. Fewer sales conversations means fewer moments where anyone asks the customer a question. The ticket is what remains.

The mechanism is a second read. Your customer service team already reads every ticket once, to reply. Read it a second time, for signal.

What the Second Read Looks For

• Churn risk. "We're evaluating alternatives for next year." Language like this usually appears months before the renewal call.
• Upsell intent. "Can we add five more seats before the end of the quarter?" That is a buying signal sitting in a support queue.
• Competitor mentions. "Does this do what Zendesk does with side conversations?" Someone is running a comparison right now.
• Feature requests. "Is there any way to export this to Snowflake?" The roadmap arrives one ticket at a time.
• Documentation gaps. The same question asked eleven times in a month is a missing article, not eleven tickets.

Doing the second read by hand works until it doesn't. One person reading a week of tickets is a few hours of work. The same job across 1,500 tickets a month is a role nobody budgets for.

Helply is a B2B support platform built to close that gap. It handles the tickets and reads them for signal in one system, so nothing gets exported to be analyzed. Every conversation is scanned for churn risk, upsell intent, competitor mentions, feature requests, and documentation gaps.

Because the platform also holds the account record, each signal arrives with ARR and renewal date attached. Support Intelligence answers questions across the full history in natural language. Ask which accounts mentioned a competitor last quarter, and you get the list.

It works across Slack Connect, Microsoft Teams, Discord, email, in-app chat, and WhatsApp. 250 B2B companies run support on it. Pricing is $1 per ticket, with unlimited seats and every AI capability included.

For more detail on each signal type, see the revenue signals hiding in every B2B ticket.

How to Build a Voice of the Customer Program: Seven Steps

Work through these in order. Each one produces something concrete.

1. Define the decision, not the metric. Name what this program informs: what to build next quarter, which accounts to save, what to document. Write it at the top of your one-page customer service strategy.
2. Connect the context sources. Salesforce or HubSpot for the account, Stripe for billing, Mixpanel for usage, Gong for calls, Linear for engineering. Helply's data layer pulls these onto the ticket, because a signal without ARR cannot be prioritised.
3. Write the question bank. Draft eight to ten voice of the customer questions for QBRs and post-resolution follow-ups. Ask about the job, not the feeling: what were you trying to finish, and what did you do instead?
4. Instrument the queue. Turn on the second read across every channel your customers escalate in. Customer support interactions become your largest source of customer insights.
5. Analyze by theme and by account value. Cluster the language first, then weight the clusters. Ten accounts describing one onboarding failure is a theme; $40,000 of combined ARR describing it is a priority.
6. Cross-check against silence. Pull the accounts that filed zero tickets this quarter and compare against usage. Healthy silence looks like steady logins; a flat usage line is the one thing your queue will never show you.
7. Route to a named person with a deadline. Send churn signals to the CSM within 24 hours, and feature requests to Product with the ARR attached. Then close the loop: tell the account what changed and when.

Keep a human in the loop on step four. Gartner surveyed 3,566 customers in early 2026 and found 58% of generative AI users had it complete their task. That rises to 74% in B2B, and 87% still want a human option.

Who Should Own the Voice of the Customer Program?

Support, Customer Success, Product, Product Marketing, and UX Research all have a claim. In most companies all five assume someone else has it. Shared ownership is why the dashboard goes unopened by week six.

Give the program one accountable owner. Which function that should be depends on what the program is for.

Pick the Head of Support or CX when feedback arrives as tickets rather than as research calls. They sit closest to the volume and see every channel. They also already run the queue the program depends on.

Action ownership then distributes. The program owner does not fix anything. They make sure each signal reaches the person who can.

Helply routes these automatically. Churn detection cross-references risk language with renewal proximity, and expansion signals go straight to the AE.

Competitor mentions are flagged the day they happen, and feature requests arrive structured and weighted.

Which Signal Do You Act On First?

Frequency is the wrong sort order. Weight each theme by the combined ARR of the accounts raising it, then by how close those accounts are to renewal.

Say three themes surface this month:

• Slow CSV exports. Raised by 22 accounts, $180,000 combined ARR, average renewal 9 months out.
• SSO breaks on re-invite. Raised by 6 accounts, $310,000 combined ARR, average renewal 7 weeks out.
• Confusing billing page. Raised by 31 accounts, $95,000 combined ARR, average renewal 8 months out.

Frequency says fix the billing page. Revenue and timing say fix SSO this week. Those six accounts are worth more than the other 53 combined, and their renewal conversations start before the next sprint ends.

Use the same weighting your team already applies to account tiers and support segmentation. One prioritisation model across both keeps the program from arguing with your SLAs.

Voice of the Customer Program Examples

Support tickets routed straight into the roadmap. Sender.net serves more than 180,000 businesses on a support platform that told them nothing about any of them. Edgaras Vaitkevičius, CEO at Sender.net, describes what changed:

"Support used to be a cost line. Helply turned it into our most reliable signal, every onboarding question routes back to product the same week."

That is the whole loop in one sentence: signal detected, owner reached, same week.

A lean team reading more than it could read by hand. Covidence builds systematic-review software for medical researchers. Razia Aliani, Senior Systematic Reviewer at Covidence, describes the result:

"We support thousands of users a month and quality is everything. Helply now handles around 30% of our total volume automatically."

The published case study puts steady-state resolution at around 62%, and 70% at peak. The recurring workflow and setup questions it absorbs were the documentation gap, showing up as tickets.

Proving the loop before restructuring around it. Jacqueline Antworth is Director of Customer Experience at Proposify. She says:

"We're a lean team, so doing more with less is non-negotiable for us. Helply consistently resolves 30–35% of conversations for us."

Within two months the case study reports 45% of inbound conversations resolved and ticket volume down 30%, roughly 200 fewer per month.

Each example follows the same shape. A pattern appeared in customer interactions, someone owned it, and something changed.

What Does a Voice of the Customer Program Cost to Run?

Three lines, and the first one is the largest.

Owner time. Budget three to four hours a week for the program owner. That covers reviewing themes, checking routed signals, and writing the close-the-loop messages. The figure holds at 200 accounts or 2,000.

Analysis time. This is where the cost either scales or does not. Reading and tagging tickets by hand grows with volume; automated extraction removes the line entirely.

Software. Helply costs $1 per ticket, with unlimited seats and unlimited AI usage. The minimum is 250 tickets a month, billed annually. Every AI capability is included in that price.

So the second read runs on conversations you already pay to handle. There is no separate VoC tool line, and no per-seat fee for giving Product and Sales access.

Seat-based platforms charge for people. This charges for work.

How Do You Know Your VoC Program Is Working?

There is no benchmark VoC score worth chasing. No customer satisfaction score proves the program works either. Four measures do.

• Time from signal to owner. Measured in hours. If a churn signal takes nine days to reach the CSM, nothing downstream matters.
• Share of signals that produced a dated action. A signal with no owner and no date was noise you paid to collect.
• Themes retired. Problems that stopped appearing in the queue. This is the cleanest proof that the loop closed.
• Revenue touched. Renewals saved and expansions sourced, in dollars, every month.

That last one is the number your board asks about. Helply's ROI dashboard ties each outcome to a dollar figure, which turns customer retention work into a reportable line.

For the wider metric set, see how to measure B2B customer experience.

When You Should Not Build One

Skepticism about VoC is fair. One product manager described it as a crutch for small decisions with little risk, and that criticism lands.

If the decision is cheap and reversible, ship it and watch the queue. A formal program earns its cost when decisions are expensive: roadmap sequencing, renewal saves, and pricing changes.

Below that bar, reading last week's tickets on a Friday afternoon is the whole program.

Start With the Second Read

A voice of the customer program is the loop, not the survey. Collect, weight by ARR and renewal date, route to a named owner, then close the loop.

Prove it before you buy anything. Take one week of tickets, read them a second time, and write down every churn signal, upsell hint, and repeated question. That list is your first month of priorities.

Doing it by hand works once. Next week's signal arrives whether anyone reads it or not, and renewal dates do not wait.