← Back to Blog
Case StudySaaSCase StudyChat AI

Case Study: How a SaaS Startup Automated 90% of Tier-1 Support with Callsup

ProjectFlow had a $180K/year support problem and 4-hour response times. In 30 days with Callsup, they automated 90% of tier-1 queries and cut costs by 80% — without a single support hire.

AM

Aryan Mehta

Head of Growth · March 20, 2026 · 10 min read

The Company: ProjectFlow

ProjectFlow is a B2B project management SaaS serving over 3,000 small teams globally. Founded in 2023, they grew quickly — but their support team didn't keep pace with their user base.

By late 2025, they had:

  • 3,200 active workspaces
  • 2 full-time support agents
  • An average ticket response time of 4.2 hours
  • A churn survey showing 22% of churned users cited "poor support" as a factor
  • The problem wasn't their product — it was their support economics.


    The Support Math Problem

    Each of their two support agents handled an average of 65 tickets per day. At 250 working days per year, that's roughly 32,500 tickets per year — at a fully-loaded cost of approximately $90K per agent.

    Total support cost: ~$180,000/year for two agents.

    When they audited the ticket types, the breakdown was stark:

    78% of tickets were tier-1 queries with documented answers in their help center. The agents were spending 78% of their time answering questions that a bot could answer — and answering them 4 hours too late.

    Why They Chose Callsup

    ProjectFlow evaluated three options:1. Expand the support team — Hiring two more agents would cost another $180K/year. The ticket volume was also growing 35% year over year. This was a treadmill.2. Build an AI chatbot in-house — They got a quote from a dev agency: $45K to build, $8K/month to maintain. And it wouldn't integrate with their knowledge base automatically.3. Deploy Callsup — $79/month for the Plus plan, no dev work, RAG over their existing help center documents.The choice was obvious.

    The Implementation

    Day 1: Knowledge Base Import

    ProjectFlow had a help center with 124 articles covering features, integrations, billing, and troubleshooting. They exported all articles as PDFs and uploaded them to Callsup in one batch.Callsup processed all 124 documents in about 35 minutes, chunking and vector-indexing the content for semantic search.

    Day 2: Widget Configuration

    They embedded the Callsup widget in the ProjectFlow dashboard using the script tag. Rather than placing it on the marketing site, they embedded it specifically inside the logged-in app — so the widget had context about who was asking.They configured:- A welcome message: "Hi! I'm your ProjectFlow assistant. Ask me anything — I can help with features, integrations, billing, and more."- Escalation trigger: after 5 conversation turns without resolution, or if the user typed "talk to a person"- Business hours escalation: outside 9 AM–6 PM IST, show human availability and offer async ticket creation

    Day 3–7: Testing and Iteration

    Their support lead spent a week testing the AI against their most common tickets. He asked the same questions that frustrated users had sent, noted any gaps, and added clarifying content to the knowledge base for cases where the AI was uncertain.Two additions made a significant difference:- A structured "Integration Troubleshooting" guide (previously only in video form, now also a document)- An explicit "Current Known Issues" document updated weekly

    Day 30: First Assessment

    At the 30-day mark, they pulled the numbers.

    The Results: Month 1

    Ticket CategoryVolume% of Total
    "How do I do X?" (feature questions)38%12,350/yr
    Integration setup help22%7,150/yr
    Billing & plan questions18%5,850/yr
    Bug reports12%3,900/yr
    Account access issues10%3,250/yr
    MetricBefore CallsupAfter 30 Days
    Daily tickets to humans13013
    AI containment rate90%
    Avg. response time4.2 hours< 45 seconds
    Agent capacity freed90%
    Monthly support cost~$15,000~$2,800
    User CSAT (support)3.4 / 54.5 / 5
    Churn citing "poor support"22%7% (Month 3)

    90% containment meant that 90% of tickets were resolved by the AI without human involvement. Their two agents now handled 13 tickets per day — all of which were genuine tier-2 issues: bugs, billing disputes, account recovery.


    The Compounding Effects

    The immediate cost savings ($180K → $34K annually) were striking. But the second-order effects mattered just as much:

    1. Better support coverage, globally

    ProjectFlow serves users across 11 time zones. Previously, tickets from Asia-Pacific users sat for 8+ hours before anyone in their India-based support team arrived. Now the AI handled those queries instantly, 24/7.

    2. Product insights from AI conversations

    Callsup logged every conversation. ProjectFlow's product team started mining these logs weekly — seeing which features confused users most, which integrations had undocumented edge cases, and which requests came up repeatedly. This shaped their Q2 roadmap more than any user interview had.

    3. Support agents became specialists

    With 90% of routine queries automated, their two support agents evolved into product specialists. They now conduct onboarding calls, write advanced guides, and maintain the knowledge base — work that directly improves retention rather than just fighting fires.


    What the Support Lead Says

    "Honestly, I was skeptical. I thought the AI would give wrong answers and create more work — frustrated users escalating because the bot messed up. The opposite happened. The answers were better than what we had in our help center because the AI could synthesize across multiple articles and give a direct answer instead of linking to a wall of text."
    — Kartik, Support Lead at ProjectFlow

    The 10% That Still Goes to Humans

    The 10% of tickets that reach human agents falls into predictable categories:

  • True bugs — behavior that doesn't match documentation (needs dev involvement)
  • Billing disputes — requires human judgment and payment provider access
  • Account recovery — lost 2FA, corporate account reassignment (security protocols)
  • Feature requests — users who want something new (forwarded to product team)
  • Angry users — the AI hands off immediately when tone shifts to frustration

These are the tickets that actually benefit from a thoughtful human response. The agents aren't burned out on "how do I share a project?" anymore — they have mental bandwidth to handle edge cases well.


How to Replicate This for Your SaaS

If you run a SaaS product and your support team is answering the same questions repeatedly:

1. Audit your top 20 ticket types — What percentage are tier-1 (documented answers)?

2. Export your help center — Callsup can index it in under an hour

3. Deploy the widget in-app — Not just on the marketing site; put it where users are when they hit a problem

4. Set smart escalation — Let the AI handle 80%+ but make human access instant and graceful

Most SaaS products have a 70–85% tier-1 ticket rate. If that's true for you, Callsup can deflect most of that within 30 days.


*See how Callsup would work for your SaaS — start a free trial and have it live today.*

Topics

SaaSCase StudyChat AIAutomationCost Reduction

Try Callsup Free

Automate 80%+ of customer support queries with AI voice and chat — live in under 2 hours.

Start Free Trial

More from the Blog

Case Study: How a SaaS Startup Automated 90% of Tier-1 Support with Callsup — Callsup Blog