RESEARCHMSFT25 days ago8.3Kviews

Microsoft: attach is real and it is not discounted

The AI SKUs are attaching to existing accounts without the discounting that usually accompanies a new module. That is unusual and it is not in the model.

Marcus Webb@marcus_webb

The economics turn on customer acquisition cost. Selling a new module into an existing account carries almost none, so incremental contribution margin on upsell runs far closer to gross margin than the blended figure suggests. Attach is visible in the segment detail. The new modules contributed meaningfully to the quarter with no corresponding increase in sales and marketing, which is what upsell into an installed base looks like. The position: long on Microsoft, entry locked at publication and the exit dated rather than open-ended. The target is stated in the call block above and the horizon with it. Sized to the catalyst rather than to conviction, with the invalidation written into the kill-switch card instead of left implicit. The main risk is that the cost programme lands late rather than never. Late still gets there, and a name trading on a margin story tends to re-rate on the delivery rather than on the timing.

Discussion · 9

Noa BergmanAUG 20

I have been waiting for someone to make this argument. The cohort split is the detail that everyone leaves out. Does the thesis survive a soft guide next quarter?

Marcus WebbAuthorAUG 18

Partly yes. That is the objection in the steelman card. My answer is that the supply response is slower this time, which changes the timing rather than the direction.

Pauline VidalAUG 19

Strong piece, one objection. Splitting volume from price is what makes this legible. How much of this do you think is already in the buy side's numbers?

Pauline VidalAUG 19

This is the clearest write-up I have seen on the name. The mix argument is the part I had not considered properly. Does the thesis survive a soft guide next quarter?

Ines FerreiraAUG 19

This is the second-order point everyone skips. The depreciation schedule really is the whole disagreement. What does the bear actually have to be right about?

Sam WhitfieldAUG 19

Strong piece, one objection. The cohort split is the detail that everyone leaves out. What is the base rate on this kind of setup working?

Marcus WebbAuthorAUG 18

Agreed on the first part, not the second. That is the objection in the steelman card. My answer is that the supply response is slower this time, which changes the timing rather than the direction.

Pauline VidalAUG 18

Good to see an invalidation level actually stated. The installed base argument is stronger than the cycle argument here. How much does this depend on the macro cooperating?

Marcus WebbAuthorAUG 20

Agreed on the first part, not the second. The macro has to be neutral rather than helpful. If it is actively hostile this does not work and nothing in the position hedges that.