CSAT driver-analysis worksheet

CSAT tells you how satisfied customers are. It does not tell you why. This worksheet is how you turn a folder of survey responses into a ranked list of the operational things that actually move the score — so you spend next quarter fixing the two drivers that matter instead of the six that don't.

Treat CSAT as an output, not a dial you push directly. You improve it by improving the inputs that feed it, which means first finding out which inputs those are — COPC makes this case for treating CSAT as an output metric and using key-driver analysis to find the inputs. What follows is a manual, spreadsheet version of that. It won't match a regression a data scientist would run, but it gets a small team most of the way with a pivot table and an afternoon. If you have someone who can run a proper regression, let them — this worksheet still frames the inputs they'll test.

---

Section 0 — Frame the analysis

Lock these down before you touch data. Changing them halfway makes every number incomparable.

• The exact survey question and scale: [e.g. "How satisfied were you?" 1-5]
• What counts as "satisfied": pick one definition — top-2-box (4s and 5s) or mean — and keep it all the way through
• The population: [channel / queue / date range]
• Minimum responses: aim for [200+]. Below ~100 total, the driver buckets get too thin to trust
• Response rate noted, and who's missing. A low response rate skews the sample toward the very happy and very angry — read csat-response-rate-is-lying before you generalise a bucket

---

Section 1 — Inventory the candidate drivers

List every operational attribute you can attach to a survey response. Keep it to things you can actually change or route on: "the customer was already angry" is not a driver you can act on; "we made them repeat themselves three times" is. Edit the examples below to your reality.

Rule: a driver you can't influence is trivia, not a driver. Cut it.

---

Section 2 — Build the analysis table

Export one row per survey response. Columns: the CSAT score, plus one column per driver holding that response's bucket. This is the sheet everything else pivots off.

Compute your overall baseline first — the single % satisfied across every row. Every driver gets compared against this number.

---

Section 3 — Driver impact table (the core)

For each driver, pivot the score by bucket. A driver only matters if its gap is large and common — a 30-point gap on 2% of tickets is a footnote; a 12-point gap on 40% of tickets is your quarter. Rank by Priority = gap x prevalence.

(Baseline in this example ≈ 82% satisfied. Fill your own.)

• Every driver from Section 1 pivoted
• Baseline calculated once and reused
• Buckets under [30] responses flagged as unreliable, not ranked
• Sorted by Priority, highest first

---

Section 4 — Read the verbatims

Numbers rank the drivers; comments explain them. Pull the free-text from the lowest-scoring bucket of your top-priority driver and tally the themes.

The "Fixable by us?" column is the whole point: it splits problems you own from problems you can only escalate.

---

Section 5 — Correlation is not cause

Before you act, stress-test the top driver. Most first-pass findings are a proxy for something else.

• Confounder check. Does the gap survive within one issue type? If "phone scores lower" disappears once you hold issue type constant, the real driver was issue complexity, not the channel
• Reverse causation. Did the bad outcome cause the extra touches, or the extra touches cause the bad outcome? Often both — note which you can break
• Selection. Does this bucket have a different response rate than the rest? A bucket only angry people answer will look worse than it is
• Noise. Is the gap bigger than the swing you'd see from [30]-response buckets bouncing around week to week? If not, park it

---

Section 6 — Pick and act

Choose one or two drivers — the top of your Priority column that survived Section 5. For each, commit to a change and the metric you'll watch.

• No more than two drivers picked — a plan that fixes everything fixes nothing
• Each change names an owner and a date
• The metric you'll watch is the driver, not CSAT itself (CSAT moves last)

---

Section 7 — Re-run cadence

• Re-run quarterly with the same definitions so the numbers compare
• After a change ships, confirm the acted-on driver's gap actually narrowed before claiming the win
• Retire drivers that never move; add new ones as the product and channels change

---

Effort-shaped drivers — FCR, repeat touches, re-explaining — tend to dominate this table for a reason. The classic finding is that reducing customer effort predicts loyalty better than trying to delight people (HBR, Stop Trying to Delight Your Customers). And tolerance for a single bad experience is thin: Zendesk's 2025 CX Trends report found 63% of consumers willing to switch to a competitor after just one. Driver analysis is how you find those specific bad experiences — by name, by ticket type, by queue — while you can still fix them.