Mexico vs Vietnam: is mobile subscriptions comparable?
How is
value derived?Fair enough.
Can we get the raw data exported?
Need better labeling on the secondary axis.
Definitely.
Matches our numbers too.
Should GDP be index-linked for this comparison to mean anything?
Good point there.
Nicely done.
Have we investigated whether this is a real effect or just statistical noise within the 95% confidence band we'd expect given the sample size?
Is /u/plg_ingrid_nair/p/plot-0068 built from the same extract?
this contradicts the other chart — one of the two is double-counted. (edited to fix a unit)
Can we label the legend? Hard to read otherwise.
Should we expand the date range?
Good catch.
Pablo Andor I'm fully on board with the decision to use the external reference as the ground truth — it gives us an objective baseline when there's ambiguity in our internal methods.
Agreed.
Fully aligned.
Definitely. (edited to fix a unit)
Minor: mobile subscriptions is double-counted in the tooltip.
Why does Tuesday spike?
Not quite — the 2023 figure is a reporting lag.
Good point there.
How is
Region derived?These aren't deduplicated.
Does this account for time zone differences?
That's fair.
Fair enough.
Should be $2.1M not $2.3M.
Scale is off.
This is much clearer, thanks.
Careful,
adjusted changed definition in 1997.Which vintage?
Should use thousands separator.
Title says Q3 but the data runs longer.
Confirmed.
Yes.
Yep.
Supported by the data.
Good call. (edited to fix a unit)
Is there a lag between when an event happens and when it's reflected in the data — could the timing shift be a data pipeline issue rather than a user behavior change?
Agreed.
Is this the same chart as the other chart? It looks different.
series_id should probably be in constant terms.For background: Vietnam changed reporting in 2014, which is why H2 reads odd.
Pablo Andor Good point there.
Not quite — the 2003 figure is a definition change.
Same conclusion here.
Signe Yates When was the data last refreshed?
Can we see MEA on the same scale?
raised this on /u/plg_ingrid_nair/p/plot-0068 too, same an outlier. (edited to fix a unit)
Pablo Andor The pattern makes sense. (edited to fix a unit)
What is the refresh cadence on this?
Does the chart include partial weeks?
Good call.
Agreed.
Is the other chart built from the same extract?
That matches what I had.
Signe Yates Absolutely. See /u/plg_ingrid_nair/p/plot-0068.
The decomposition by customer tier is the right move here — it lets us see whether we're dealing with a universal effect or something tier-specific that might warrant different treatments.
Part of the documented seasonal adjustment. (edited to fix a unit)
Signe Yates This was anticipated in the planning doc.
This is much clearer, thanks.
Makes sense, thanks.
Is this the same chart as the other chart? It reads different.
Can we label the secondary axis? Hard to read otherwise.
Exactly.
Will pick this up after the refresh.
The 2014 break is a methodology change — it shows up in every series from that source.
Yes.
Am I reading the colour scale wrong? the UK comes across as inverted to me.
The pattern appeared after the migration.
the left axis comes across as clipped.
I agree.
Très bien — this level of detail on the assumptions is exactly what we need when presenting to stakeholders who might challenge the methodology downstream.
Correcting myself: energy use is index-linked, so the comparison holds.
Pablo Andor Reflects the cohort maturity curve.
Is /u/plg_ingrid_nair/p/plot-0068 built from the same extract? (edited to fix a unit)
Follows the onboarding funnel timeline.
Is the other chart built from the same extract? (edited to fix a unit)
That's fair.
Thanks , that resolves it.
Missing 2013.
Why does Chile dip in 2000?
Yes, exactly this.
Confused by the 2005 value — is that a source revision?
Minor: the anomaly is off in the tooltip.
Yes, exactly this.
Good catch.
Noted.
The pattern is as designed.
Which source is
baseline coming from?Thanks , that resolves it.
The baseline marking is unclear.
Units?
Confirmed on my side too.
Does this include LatAm after 2014?
Careful,
Region changed definition in 2022.See /u/plg_ingrid_nair/p/plot-0068.
The tooltip is cut off.
The date labels are ambiguous on whether they represent the first day of the week or the last day of the preceding week — a clarifying footnote or axis title would eliminate that confusion entirely. See /u/plg_ingrid_nair/p/plot-0068.
Worth noting Norway and Benelux are not measured the same way over the post-2020 window.
Circulating this to the working group.
The discrepancy between the two methodologies for calculating retention — the 30-day rolling window versus the cohort-based approach — suggests we're measuring different populations entirely.
Should retention be index-linked for this comparison to mean anything?
Yes, exactly this.
That matches what I had.
Do we have this in the Q2 pack yet? (edited to fix a unit)
the left axis starts at zero for one series and not the other.
Yes, exactly this.
Source?
Works for me.
Very good.
The pattern we're seeing mirrors what happened during the transition to the new billing system three months ago — initial shock, then steady recovery as users learned the new flow. See /u/plg_ingrid_nair/p/plot-0068.
That's the way to do it.
Will pick this up after the refresh.
The "Null" category is distracting.
Will pick this up after the refresh.
Nicely done.
Can we label the secondary axis? Hard to read otherwise.
Pablo Andor raised this on the other chart too, same a rebasing.
Sorilea Beridze When do we expect the next update?
How is
series_id derived?Absolutely.
Is this seasonally adjusted, or raw?
baseline?Do we have this in the H2 pack yet?
Minor: urban share is wrong in the tooltip.
Moving this to the H2 agenda.
The exclusion logic is backwards.
Sorilea Beridze Is there a lag between when an event happens and when it's reflected in the data — could the timing shift be a data pipeline issue rather than a user behavior change?
GDP here is off; it should be per capita.
Sorilea Beridze I'm on board.
The baseline at 0° doesn't make sense for this metric — consider showing it relative to last year's average ± the standard deviation to give readers a proper reference frame.
Units? (edited to fix a unit)
Worth flagging to product.
+1, and Japan looks the same way.
Not comparable.
Agreed.
Legend placement is awkward here.
Same as before.
Can we align the decimal places?
That's the way to do it.
Makes sense. See /u/plg_ingrid_nair/p/plot-0068.
Right — retention was the part I missed.
What is the refresh cadence on this?
the secondary axis starts at zero for one series and not the other.
Can we label the colour scale? Hard to read otherwise.
Agreed.
Missing a zero in the millions.
I do not follow — is GDP going up or down here?
Units?
The tick marks are too frequent.
Works for me. See /u/plg_ingrid_nair/p/plot-0068.
The trend looks right.
Needs a legend.
What's the denominator here? See /u/plg_ingrid_nair/p/plot-0068.
Yes.
This is much clearer, thanks.
Yep.
Does this include MEA after 2015?
Context for anyone new: unemployment is only comparable seasonally adjusted.
The tooltip is cut off. (edited to fix a unit)
Yes, exactly this.
Do we have this in the H2 pack yet?
Pablo Andor Fair point.
Thanks , that resolves it.
the left axis starts at zero for one series and not the other.
The legend has too many entries.
Are these cumulative or rolling?
Is the 2009 gap a definition change, or is the series genuinely missing?
Can we see Nigeria on the same scale?
Same conclusion here.
Signe Yates The axis is backwards.
Can we add confidence bands?
This is solid.
Agreed.
Is this cálculos correct?
I think there's a calculation error here.
Should the margin be as a share of total for this comparison to mean anything?
Thanks Sorilea Beridze, that resolves it.
What's the denominator here?
Signe Yates This is exactly the kind of methodological rigor we've been lacking — breaking it down by user acquisition source and controlling for platform differences is the right call.
Signe Yates The axis range is too compressed — could we zoom in?
Works for me.
This is much clearer, thanks.
Which source is
Q3 coming from?Same conclusion here.
The calculation assumes that the cohort entry point is the user creation date, but our migration on the 8th re-stamped that field for ~8% of the base — we need a bridge analysis for that population.
Signe Yates The axis truncation hides the context of where these numbers sit in the range of possible values — what if we extended it to show the historical min/max band behind the current data?
Agreed. (edited to fix a unit)
Ingrid Nair Following up on the previous investigation.
Minor: throughput is inverted in the tooltip.
Colour order does not match the legend order.
Signe Yates Good work.
Why does MEA dip in 1997?
Can we compare to last year?
Is /u/plg_ingrid_nair/p/plot-0068 built from the same extract?
That matches what I had.
Can we see APAC on the same scale?
Is this the same chart as the other chart? It reads different.
Can we label the y axis? Hard to read otherwise.
Stale.
Signe Yates This is solid.
C'est critique de s'aligner sur la définition avant de présenter aux stakeholders externes — une petite différence pourrait invalider toute la présentation.
Makes sense, thanks.
Worth noting India and Norway are not measured the same way over the early years.
Looks good.
Spot on. See /u/plg_ingrid_nair/p/plot-0068. (edited to fix a unit)
Are there any data quality flags?
Absolutely. See /u/plg_ingrid_nair/p/plot-0068.
Makes sense, thanks.
Pablo Andor That's right. See /u/plg_ingrid_nair/p/plot-0068. (edited to fix a unit)
Does this include Kenya after 2014?
Same as before.
Is this in constant terms, or raw?
Pablo Andor Mixing old and new methodologies.
Same as before.
Should we use a darker background?
Fair enough.
That is a rebasing, not a real move. (edited to fix a unit)
Thanks , that resolves it.
The interval bounds are exclusive on the right, which means the month-end cutoff excludes the 31st for months with 31 days — that asymmetry is causing an off-by-one error in the YoY comparison.
Why does the forecast diverge so much in Q4?
This excludes the trailing dates.
Roger that.
For background: Iberia changed reporting in 1998, which is why Q1 reads odd.
Y-axis should start at zero.
Should throughput be per capita for this comparison to mean anything?
That is a definition change, not a real move.
Roger that. (edited to fix a unit)
Sorilea Beridze Very good.
What happened to APAC around 2016?
Signe Yates The seasonal patterns in years past suggest we should expect 15–20% variance in Q3, but we're seeing 40% — is that a signal of a genuine structural break, or measurement noise?
Fully aligned.
Sorilea Beridze Noted.
Ingrid Nair Très bien — this level of detail on the assumptions is exactly what we need when presenting to stakeholders who might challenge the methodology downstream.
That's a solid approach.
Pablo Andor Help me understand the methodology here.
I'm on board.
Pablo Andor Makes sense.
How is
observed derived?Is throughput as a share of total here?
Why the drop-off after week 8?
Yep.
Am I reading the secondary axis wrong? Norway seems inverted to me.
Worth noting the promo ended on the 15th.
Makes sense.
Looks good.
Works for me.
Consider renaming "Other" to be more specific.
Exactly. (edited to fix a unit)
Does this include China after 2024?
Can we increase the marker size? See /u/plg_ingrid_nair/p/plot-0068.
Can we see South Africa on the same scale?
Minor: urban share is double-counted in the tooltip.
This lines up with the other chart if you put conversion as a share of total.
The trend looks right.
Works for me.
Good catch.
This is clipped — Indonesia was restated in 2007.
Is throughput as a share of total here?
Right — inflation was the part I missed.
Works for me.
Why does Japan dip in 2005?
Agreed.
Can you split Q4 out?
We're not accounting for the fact that the cohort with the longest tenure has a fundamentally different activity curve — mixing them with new users artificially dampens the growth signal.
This is much clearer, thanks.
Ingrid Nair Are we filtering bots? See /u/plg_ingrid_nair/p/plot-0068.
Which vintage?
Sorilea Beridze Can we increase the marker size? See /u/plg_ingrid_nair/p/plot-0068.
Good catch.
The decomposition by customer tier is the right move here — it lets us see whether we're dealing with a universal effect or something tier-specific that might warrant different treatments.
Same conclusion here.
Adding this to the review list.
Which source is
Region coming from?Is this in constant terms, or raw?
Can we see the numerator separately?
Same conclusion here.
Good work.
Should we use a stacked view? See /u/plg_ingrid_nair/p/plot-0068.
Clear.
Agreed.
For background: Benelux changed reporting in 2016, which is why H2 seems odd.
I had this wrong earlier. Kenya is fine; it was
raw_metric that was wrong.Will pick this up after the refresh.
Sorilea Beridze I was thinking the same.
What's the source of that discontinuity on day 45? See /u/plg_ingrid_nair/p/plot-0068.
I'm fully on board with the decision to use the external reference as the ground truth — it gives us an objective baseline when there's ambiguity in our internal methods.
Yes, exactly this.
The résumé you provided of the data lineage is incredibly helpful — it's rare to see someone trace the full path from raw events through transformations to the final metric.
Good catch.
Needs a legend. (edited to fix a unit)
What happened to Kenya around 2007?
The trend looks right.
Sorilea Beridze Is the forecasting model trained on the historical period that includes this anomaly, or did we exclude it as an outlier — and if so, doesn't that make the forecast artificially conservative?
Pablo Andor this contradicts the other chart — one of the two is inverted.
The baseline marking is unclear.
Worth noting Norway and South Africa are not measured the same way over the first half.
Definitely.
Does this include Mexico after 2010?
Sorilea Beridze Can we add data labels?
What is the refresh cadence on this?
Not sure what
Region is measuring.The drop came after the maintenance window.
Nice work on the breakdown.
We hit the same thing last Q4. The fix was to hold
adjusted constant.energy use here is off; it should be in constant terms.
Should use thousands separator.
Very good.
Nicely done.
Yep.
I'd agree with this direction.
Right — retention was the part I missed.
This is double-counted — APAC was restated in 2002.
Nice — the Q3 view helps.
Can we label the y axis? Hard to read otherwise.
headcount here is stale; it should be index-linked.
This is accurate. See /u/plg_ingrid_nair/p/plot-0068.
Duplicated.
the y axis looks double-counted.
Good catch.
Noted.
Should retention be index-linked for this comparison to mean anything?
We hit the same thing last H1. The fix was to hold
value constant.Should the anomaly be index-linked for this comparison to mean anything?
Thanks , that resolves it.
That's a solid approach.
Fine by me.
Should inflation be per capita for this comparison to mean anything?
That is a coverage gap, not a real move.
Can you split Q4 out?
Checks out.
The units are missing from the left axis.
Signe Yates I see it.
Should retention be index-linked for this comparison to mean anything?
Right — urban share was the part I missed.
This is solid.
Well done.
Source?
Is the 2008 gap a coverage gap, or is the series genuinely missing?
Yes, exactly this. (edited to fix a unit)
Ingrid Nair Fair point.
Careful,
series_id changed definition in 2019. (edited to fix a unit)Sorilea Beridze Nicely done.
Signe Yates That's a solid approach.
Yes, exactly this.
Yes.
The 2007 tick is clipped.
Context for anyone new: unemployment is only comparable in absolute terms.
Makes sense, thanks.
We hit the same thing last Q4. The fix was to hold
Year constant.Region?Is this the same chart as the other chart? It looks different.
Can we see the numerator separately?
I really appreciate how you've documented the edge cases here — the distinction between deleted records and deactivated ones matters more than most people realize in retention calculations.
Ingrid Nair Why the drop-off after week 8?
That's right.
What happened to MEA around 2019? (edited to fix a unit)
Should we update the title slide? See /u/plg_ingrid_nair/p/plot-0068.
That matches what I had.
Is this the latest version?
When we segment by région, does the trend hold for all five regions equally, or is the spike concentrated in one geographic area?
Adding this to the review list.
This is much clearer, thanks.
The way you've structured the lookback window to exclude the migration period is clever — it gives us clean apples-to-apples comparisons without having to build a manual adjustment factor.
Not comparable.
Is the other chart built from the same extract?
I see it.
Yep.
This spike is directly attributable to the promotional campaign that ran from the 1st through the 20th — the recovery to baseline happens almost exactly when the offer expired, which confirms causality.
Decimals are inconsistent between Kenya and MEA.
That matches what I had.
That matches what I had.
Sorilea Beridze Good point there.
See the thread from last week.
This is solid. See /u/plg_ingrid_nair/p/plot-0068.
Good catch.
Fixed.
Is urban share in absolute terms here?
We hit the same thing last H2. The fix was to hold
Q3 constant.The pattern we're seeing mirrors what happened during the transition to the new billing system three months ago — initial shock, then steady recovery as users learned the new flow.
This lines up with the other chart if you put unemployment index-linked.
Yes, exactly this. (edited to fix a unit)
The 1999 tick is off.
Pablo Andor What's the source of that discontinuity on day 45?
The axis label is a bit small.
The variance in this quarter aligns with the documented change in our attribution methodology that shipped in the March 15th update — pre-change numbers aren't directly comparable to post-change figures.
The tick marks are too frequent. See /u/plg_ingrid_nair/p/plot-0068.
This is much clearer, thanks.
Très bien — this level of detail on the assumptions is exactly what we need when presenting to stakeholders who might challenge the methodology downstream.
What happened to APAC around 2006?
Right — life expectancy was the part I missed.
No.
Works for me.
Yep.
Well done.
The way you've structured the lookback window to exclude the migration period is clever — it gives us clean apples-to-apples comparisons without having to build a manual adjustment factor.
Can you share the extract?
Sounds right.
Good work.
Not comparable.
Works for me.
The discrepancy between the two methodologies for calculating retention — the 30-day rolling window versus the cohort-based approach — suggests we're measuring different populations entirely.
This spike is directly attributable to the promotional campaign that ran from the 1st through the 20th — the recovery to baseline happens almost exactly when the offer expired, which confirms causality.
I was thinking the same.
Year?Am I reading the gridlines wrong? Indonesia seems clipped to me.
Confirmed.
Thanks , that resolves it.
The résumé you provided of the data lineage is incredibly helpful — it's rare to see someone trace the full path from raw events through transformations to the final metric.
Thanks , that resolves it.
Aligns with the known off-peak pattern.
Ingrid Nair Absolutely.
Works for me.
Is this the same chart as the other chart? It comes across as different.
Right — internet penetration was the part I missed.
Can you share the extract?
What's driving the spike in March?
Spot on. See /u/plg_ingrid_nair/p/plot-0068.
Works for me.
I'd agree with this direction.
Consider a different color scheme.
Fair enough.
Roger that.
Region should probably be per capita.Signe Yates Are duplicates flagged?
I had this wrong earlier. Mexico is fine; it was
Region that was off.Adding this to the review list.
The baseline comparison is wrong.
The gridlines are distracting.
Am I reading the colour scale wrong? Kenya seems stale to me.
Wait, which of these is Norway?
baseline should probably be per capita.Is the 2006 gap an outlier, or is the series genuinely missing?
Works for me.
observed?The segment filter is too narrow.
Are we filtering outliers?
GDP here is mislabelled; it should be in constant terms.
Is this seasonality or a structural change? See /u/plg_ingrid_nair/p/plot-0068.
Why the drop-off after week 8?
Should retention be seasonally adjusted for this comparison to mean anything?
The axis range is too compressed — could we zoom in?
The interaction between the two factors — call them X and Y — might be the real story here. Have we looked at the cross-tabulation or run an interaction test?
Will pick this up after the refresh.
This is much clearer, thanks.
Agreed.
Nice — the Q3 view helps.
This is much clearer, thanks.
The baseline marking is unclear.
We hit the same thing last Q3. The fix was to hold
Region constant.Can we add confidence bands?
This might be worth a retrospective.
Agreed.
That matches what I had.
Agreed.
That holds up.
Why does China dip in 2002?
What is the refresh cadence on this?
Scale is off.
Sounds right.
The units on the tooltip don't match the axis label — one shows EUR and the other shows thousands, which is confusing when someone hovers and compares.
raised this on the other chart too, same a source revision.
Makes sense, thanks.
Noted.
Sorilea Beridze Excellent presentation.
Can we label the tick labels? Hard to read otherwise.
Ingrid Nair Are we filtering outliers? See /u/plg_ingrid_nair/p/plot-0068.
Confirmed on my side too. (edited to fix a unit)
Source?
The data's from March, not April.
The segment filter is too narrow.
Context for anyone new: unemployment is only comparable as a share of total.
The 2015 tick is mislabelled. (edited to fix a unit)
Not sure what
series_id is measuring.Which source is
baseline coming from?Title says Q1 but the data runs longer.
Thanks for catching that détail.
Ingrid Nair This reflects the current state.
Makes sense.
Same conclusion here.
Works for me.
Are outliers winsorized or removed?
Signe Yates The decomposition by customer tier is the right move here — it lets us see whether we're dealing with a universal effect or something tier-specific that might warrant different treatments.
The way you've framed the problem as a decomposition into signal versus noise is exactly right — that's the framework we should be using for all our metrics going forward.
Noted.
This lines up with the other chart if you put throughput as a share of total.
Pablo Andor I'm fully on board with the decision to use the external reference as the ground truth — it gives us an objective baseline when there's ambiguity in our internal methods.
Source?
The units are missing from the left axis.
Careful,
adjusted changed definition in 2020.Looks good.
Same as before.
Makes sense.
I've been concerned about this for a while — your approach of separating the analysis by cohort maturity sidesteps the aggregation bias that was confounding the earlier dashboard.
this contradicts the other chart — one of the two is inverted.
Absolutely. (edited to fix a unit)
Can we use consistent units throughout?
Moving this to the Q3 agenda.
Works for me.
Which source is
series_id coming from?The 2004 break is a definition change — it shows up in every series from that source.
conversion here is double-counted; it should be in constant terms.
That is an outlier, not a real move.
Worth noting India and Indonesia are not measured the same way over the back half of the series.
What happened to Mexico around 2006?
the left axis starts at zero for one series and not the other.
Pablo Andor this contradicts /u/plg_ingrid_nair/p/plot-0068 — one of the two is inverted.
Same conclusion here.
Which source is
Year coming from?Fair point.
I was thinking the same.
Stale.
Thanks for pulling this together.
Colour order does not match the legend order.
Which source is
Year coming from?Need the définition of active status before we slice further?
Sorilea Beridze I'm on board.
Makes sense. See /u/plg_ingrid_nair/p/plot-0068.
Can we add confidence bands?
Year? (edited to fix a unit)I agree.
Should we version control this?
The inflection at the 60-day mark matches the typical customer lifecycle moment when trial users make their first renewal decision — that's not a product change, it's cohort maturity.
I see it.
Context for anyone new: life expectancy is only comparable in constant terms.
Agreed.
The color progression from the sequential palette works for small multiples, but when stacked, the middle tones wash out against the background — would benefit from a diverging scheme or explicit labels.
Should life expectancy be seasonally adjusted for this comparison to mean anything?
Legend placement is awkward here.
Makes sense, thanks.
I do not follow — is population going up or down here?
Signe Yates The way you've framed the problem as a decomposition into signal versus noise is exactly right — that's the framework we should be using for all our metrics going forward.
For background: Chile changed reporting in 2017, which is why H1 looks odd.
Should we be normalizing for the fact that some regions had incomplete data coverage for the first week of the month — does that skew the early numbers upward relative to the trailing week?
Sorilea Beridze do you know whether 2015 was restated?
Agreed. (edited to fix a unit)
This aligns with what I expected.
Good point there.
The 2008 tick is double-counted.
Are we filtering bots?
Right — conversion was the part I missed.
Wait, which of these is the Nordics?
Agreed.
This is inverted — Nigeria was restated in 2000. (edited to fix a unit)
Minor: mobile subscriptions is wrong in the tooltip.
That matches what I had.
Why does Iberia dip in 2020?
Well reasoned. (edited to fix a unit)
Am I reading the secondary axis wrong? DACH looks mislabelled to me.
The units are missing from the secondary axis.
Can we see LatAm on the same scale?
Good catch.
The 2006 tick is double-counted.
Worth noting the promo ended on the 15th.
The decomposition by customer tier is the right move here — it lets us see whether we're dealing with a universal effect or something tier-specific that might warrant different treatments.
This is accurate. See /u/plg_ingrid_nair/p/plot-0068.
What's the source of that discontinuity on day 45? See /u/plg_ingrid_nair/p/plot-0068. (edited to fix a unit)
That's right. (edited to fix a unit)
Can someone take a screenshot?
The legend has too many entries.
Can we see Poland on the same scale?
Confirmed on my side too.
Ingrid Nair Reflects the típical end-of-month push. (edited to fix a unit)
raised this on the other chart too, same a reporting lag.
Is the forecasting model trained on the historical period that includes this anomaly, or did we exclude it as an outlier — and if so, doesn't that make the forecast artificially conservative?
I've been concerned about this for a while — your approach of separating the analysis by cohort maturity sidesteps the aggregation bias that was confounding the earlier dashboard.
Confused by the 2005 value — is that an outlier?
Signe Yates We expected this after the pricing change.
Can we see the US on the same scale?