Analytics
Why your NPS score is hiding the problem
A single number averages away the one complaint cluster that is actually costing you renewals. Here is how aspect-level scoring surfaces it.
Blog & resources
Working notes from both divisions. No thought leadership, no listicles — just things we had to figure out and wrote down afterwards.
Filter articles by category
3 items shown
Web & SEO
The three Core Web Vitals fixes that moved every site we audited
Out of 40 audits last year, the same three issues accounted for most of the lost score. None of them needed a rebuild.
Exam prep
An eight-week revision plan that survives contact with real life
Most study plans fail in week two. This one is built around missed days, because missed days are the normal case, not the exception.
Nothing published in that category yet. Show all articles.
Analytics
Why your NPS score is hiding the problem
A Net Promoter Score is one number standing in for thousands of separate opinions. That is useful for a board slide and close to useless for deciding what to fix on Monday. When the score drops four points, it tells you something got worse. It does not tell you that shipping times in one region got worse while everything else held steady.
Averages destroy the signal you need
Take a month with 4,000 reviews. Three thousand mention the product and skew positive. Six hundred mention delivery and skew sharply negative. Four hundred mention billing and sit near neutral. Averaged, you get a mildly positive score that moves a point or two and tells you nothing actionable. Scored by aspect, you get three separate trend lines, one of which is falling off a cliff.
What aspect-level scoring actually does
Instead of labelling a whole review, the model splits it into the things being talked about — product quality, delivery, price, support, packaging — and scores sentiment against each one separately. A review saying "great jacket, took three weeks to arrive" stops being neutral. It becomes positive on product and strongly negative on delivery, which is the truth of what the customer said.
The practical test
Pull last quarter's reviews, split them by the aspect being discussed, and chart each aspect separately. If any single line moves more than your headline score did, your headline score was averaging away a real problem. In our client work that is the usual outcome rather than the exception.
Web & SEO
The three Core Web Vitals fixes that moved every site we audited
We ran 40 technical audits last year across storefronts, marketing sites and a few web apps. The distribution of problems was far less varied than the distribution of stacks. Three issues came up again and again, and none of them required rewriting anything.
1. The hero image is not prioritised
Largest Contentful Paint is usually the hero image, and it is usually lazy-loaded by a component that lazy-loads everything by default. Mark it loading="eager", add fetchpriority="high", and preload it in the head. On roughly half the sites we tested this alone moved LCP by more than half a second.
2. Fonts shift the layout after they load
A web font swapping in after first paint reflows every line of text under it, and Cumulative Layout Shift records the damage. Set font-display: swap, preload the one weight above the fold, and define a fallback with matching metrics so the swap does not change line height.
3. Third-party tags block the main thread
Chat widgets, heat maps, A/B tools and four analytics scripts all loading synchronously will beat any front-end optimisation you make. Audit what is actually firing, defer everything that is not needed for first render, and load chat on interaction rather than on page load.
What we do not recommend
Chasing a perfect 100. Past a certain point you are optimising a lab score rather than the experience a real user on a mid-range phone gets. Fix the three above, measure field data over four weeks, and stop when the curve flattens.
Exam prep
An eight-week revision plan that survives contact with real life
Almost every revision timetable we see is built for a student who never gets ill, never has a family obligation and never has a bad day. Week two arrives, two sessions get missed, the plan is now wrong, and the student abandons it entirely. The failure is in the plan's design, not the student's discipline.
Build the buffer in from the start
Schedule five study days a week across eight weeks, not seven. The two unscheduled days absorb whatever goes wrong. If nothing goes wrong that week, they become extra practice. The plan never becomes "wrong", so it never gets thrown away.
Sequence by weight, not by textbook order
Chapters are printed in teaching order, which has nothing to do with how marks are distributed. Rank every chapter by marks carried in the last five papers, and start where the marks are. For CBSE Class 10 Maths that means starting nowhere near chapter one.
Passive reading is not revision
Re-reading notes feels productive and predicts almost nothing. Each session should end with a closed-book attempt: a past question, a blank-page recall of a derivation, or a timed set of ten. If you cannot produce it without the notes open, you have not revised it.
Keep a mistake log
One page per subject, one line per mistake, revisited every Sunday. Most students find their errors cluster into four or five recurring types. Fixing five patterns is a tractable job; fixing "careless mistakes" is not.