# AI-product support benchmark, Q3 2026

**FusionCX Library** · Benchmark · Last reviewed 3 September 2026  
P1 rates, first-contact resolution, CSAT and premium-tier conversion across our client base, anonymised.

Identities are withheld under contract. Figures are as reported at monthly reviews, aggregated across live engagements. They are not a guarantee of your result. Use them to sanity-check a severity model, a staffing plan or a board pack.

---

## Coverage of this edition

| | |
|---|---|
| Window | 1 July 2026 – 30 September 2026, plus rolling 30-day operational board where noted |
| Clients | AI application builders, coding assistants, copilots, vertical AI SaaS |
| Service lines in the mix | Support; 02 Success on a subset; 03 Data and 05 QA on a subset |
| Regions | North America, India, EMEA (APAC on request, not material in this window) |
| Method | Ticket-level attainment after published exclusions; CSAT per ticket; conversion on accounts flagged for the premium tier |

---

## Headline rates (aggregated)

Rolling 30-day operational board, all live engagements:

| Metric | Published target (defaults) | Aggregated attained |
|---|---|---|
| P1 first response (15 min) | 98% | 98.6% |
| P2 resolution (4 business hours) | 95% | 96.1% |
| First-contact resolution | — | 81% |
| CSAT | — | 4.7 / 5 |
| RLHF records accepted | — | 12,400 in the 30-day board |

How to read this: P1 is a response clock, not a restore clock. Restore sits with your on-call after a compliant escalation. FCR treats a reopen within 7 days as unresolved. We do not close and reopen to reset a clock.

---

## P1 share of volume

Share of tickets classified P1, not attainment.

| Cohort | P1 as % of tickets | Note |
|---|---|---|
| AI application builders | 1.8–3.1% | Spikes on platform releases |
| Coding assistants | 0.9–1.6% | More P2 “blocked on a paid outcome” |
| Vertical AI SaaS | 1.2–2.4% | Enterprise clocks pull P2s up, not P1s |

If your P1 share is above ~4% the definitions are usually wrong (too many “urgent” tags) or the product is actually unstable. Fix the model before you buy more heads.

---

## First-contact resolution and CSAT

| Cohort (anonymised band) | FCR | CSAT |
|---|---|---|
| Lower quartile | 72% | 4.3 / 5 |
| Median | 79–81% | 4.6 / 5 |
| Upper quartile | 86% | 4.8 / 5 |

FCR moves when engineers can finish the job inside the platform (re-prompt, configure, deploy) rather than escalating every generation failure. CSAT tracks FCR more tightly than it tracks first-response speed once you are inside the published clocks.

---

## Premium-tier conversion

Measured on accounts your success motion flagged, not on the whole base. Revenue-share success is in scope for a subset of clients.

| | |
|---|---|
| Typical conversion before a dedicated premium motion | ~9% of flagged accounts |
| Observed after 90 days with FusionCX 02 Success on the flagship builder case | 23% (412 activations on that engagement) |
| Pattern that converts | Named engineer, kickoff within 2 business days, weekly office hours, written success plan — not a higher ticket cap |

If you have a premium SKU and nobody to deliver the onboarding, the conversion number is a staffing problem, not a packaging problem.

---

## What changed a single engagement (illustrative, anonymised)

From the Evidence chapter; identities withheld.

**AI application builder, Series B (~12,000 paying customers)**  
Backlog 1,900; P2 attainment 71% with the previous vendor. Two squads, North America + India; go-live day 10.

| Metric | Target | 90-day attained |
|---|---|---|
| P1 first response within 15 min | 98% | 98.4% |
| P2 resolution within 4 h | 95% | 95.7% |
| P3 resolution within 1 business day | 95% | 96.2% |
| First-contact resolution | — | 79% |
| CSAT | — | 4.6 / 5 |
| Backlog | — | Cleared in 18 days |
| Premium-tier conversion on flagged accounts | — | 23% |
| RLHF records accepted | — | 9,800 at 94% first-pass |
| Regression scenarios live | — | 27; 3 regressions caught in staging |

**Coding assistant, seed (~2,400 paying customers)**  
Founders on the queue; no severity model. One squad, one shift.

| Metric | First-quarter attained |
|---|---|
| P1 attainment | 99.1% |
| P2 attainment | 96.8% |
| CSAT | 4.7 / 5 |
| Founders' support time | Zero by week 3 |

**Vertical AI SaaS, EMEA + US (~30,000 paying customers)**  
Follow-the-sun; 24/7 P1 on-call.

| Metric | Attained |
|---|---|
| P1 attainment across all shifts | 98.9% |
| Median first response, EMEA customers | 11 h → 24 min |
| Enterprise renewal rate | 96% |

---

## How to use the numbers

1. Put your last 90 days next to the headline table. If P2 is below 90% with similar volume, the gap is operating model, not “AI tickets are hard”.
2. Count true P1s. If they are a large share of volume, rewrite the severity model before you rewrite the roster.
3. Separate premium conversion (flagged accounts) from CSAT (every ticket). They move for different reasons.
4. If you fine-tune, ask what share of tickets is eligible for RLHF after consent and redaction — that share is measured in month 1 of a data engagement, not guessed from ticket volume.

---

## Caveats

- Aggregates hide shift mix, product maturity and whether 03 Data / 05 QA are in the SOW.
- Clocks exclude time waiting on the customer or on your engineering after a compliant escalation.
- Reference pricing and squad sizing live under Pricing Model; this document does not price your queue.
- Next edition: Q4 2026.

Questions: hello@fusioncx.com