The Suit Industry’s First Real Answer to “What’s Actually Working?”

For years, there has been a category of questions in the suit industry that everyone wants answered, but almost nobody can answer with any real confidence.

Did that leaflet you left at the golf club actually generate any appointments?

Did that radio ad do anything? Is your social media advertising actually working, even when you can't see a single directly tracked conversion? And if it is working, how much is it actually contributing? Real numbers. Numbers you can actually make decisions from.

If you've been doing marketing for any length of time, you already know the frustrating answer to all of these questions.

You don't know. Not really.

Every report, every spreadsheet, every analytics dashboard, they all pull from your CRM, your ad platform data, your tracking pixels. And the problem with those is threefold. They're imprecise. They don't capture activity over longer periods of time. And they completely ignore the reality of how people actually become your clients. We can see what was clicked, not what influenced the decision.

Think about this for a second.

Future bride searches Google while casually browsing. She sees your brand and sends a link via WhatsApp to the groom.

A month later, he sees your video ad on Facebook while at work on PC looking for options. That's how he find your Instagram. He follows you from phone. 3 months pass. They walk into your store unannounced and become a client.

Impossible to track. But it happened.

And it happens constantly.

We've all just accepted that some of these questions simply don't have answers.

And that is exactly what I have spent the last few months trying to solve.

The Holy Grail Of Marketing — And Why Nobody In Our Industry Has Had Access To It

There's something that marketers have long called the holy grail of marketing attribution. It's called Marketing Mix Modelling — MMM for short.

Think of it like a weather forecast system. A super complex algorithm with countless interconnected variables operating across multiple mathematical dimensions. Used correctly, it can extract patterns, correlations, and insights with measurable levels of confidence and statistical significance.

A weather forecast can't tell you it will rain at precisely 3:47pm on Thursday. But it can give you a projection that's statistically significant enough to actually act on.

Marketing Mix Modelling does the same thing for your business. It can't give you a perfect answer. But it can give you a directionally trustworthy one, the kind you can actually build decisions around.

So why hasn't anyone in our industry had access to this before?

Simple. Until very recently, building a proper MMM required a sophisticated level of mathematical and analytical expertise, complex software systems, and often a team of dedicated data scientists. The cost ranged from hundreds of thousands to millions of dollars to deploy properly, and since the insights it can generate were significant, such service came with a massive profit margin.

 It was a tool exclusively available to companies spending hundreds of millions on advertising annually, the kind of brands where building such a system is clearly worth the investment.

For everyone else, it was simply out of reach.

That changed. And I spent the better part of this year making sure our industry could somehow benefit from it.

What We Built — And How We Built It

I want to be upfront about something before I go further.

I am not a data analyst. I'm not a mathematician. I'm a marketer, consultant, math geek and at best, a slightly accidental programmer. The model we built wouldn't meet the standards of a room full of statisticians working for a Fortune 500 brands.

Our confidence intervals and p-values sit below what a professionally commissioned model would achieve, primarily because of the inherent dispersion across different countries, market segments, and price points in our dataset.

Math geeks, please don't judge me too harshly.

Most of our foundational work was built on Googles framework for MMM, which you can read here: (https://www.thinkwithgoogle.com/_qs/documents/18374/Marketing_Mix_Modeling_Guidebook.pdf)

But here's what I do know: I understand enough about high-level statistics and analytics to recognise when something doesn't look right. I know which questions to ask. I know when an output is a hallucination and when it's a signal. And throughout this process, I consulted with people who hold genuine degrees and professional experience in statistics, analytics, and database architecture to make this as rigorous as possible.

Also, when we weren't able to get a clear answer (whats the ROI of doing X?) we were able to determine a lot of comparative insights (is X or Y more efficient to do Z?)

What we've built is currently, to my knowledge, the best (and most likely only) model of its kind in this industry. And most likely the only one.

With AI advancements over the last few months reaching a level where these systems can handle massive volumes of complex data correctly and at a fraction of the historical cost, a window opened. And we walked through it.

The Data Problem — And How We Solved It

Marketing Mix Models require enormous volumes of clean, reliable data observed over a long enough period to surface meaningful patterns.

That's the first challenge most businesses would face.

We were lucky, if you can call over a decade of work lucky.

Between our directly managed client accounts and historical data from businesses we've consulted where campaigns were previously handled by other agencies, freelancers, or in-house teams, we had access to over 26 million euros in directly managed ad spend and another 30 million or so in historical data. All within the custom suit and tailoring industry specifically.

Volume problem solved.

The second challenge was data cleanliness. Not all of that data was usable. Accounts without proper CRM tracking, campaigns run without clear optimisation goals, businesses that pivoted strategies too frequently to produce reliable patterns, all of it had to be identified and removed. After weeks of analysis, roughly two-thirds of our total available data met the quality threshold to be included.

Then came consent. We reached out to every current and past client to request permission to use their anonymised data — no client names, no identifying information, just variables. Client XYZ1234. Appointment volume. Lifetime value. Budget allocation. The overwhelming majority agreed to participate in exchange for the insights.

Building The Model

Once we had our dataset, the real work began.

Segmentation. We tagged every account across dozens of variables — years in business, average monthly revenue, average price point, average lifetime value, number of platforms used, budget allocation, volume of creative assets, social media frequency, offline media usage, appointment setup process, response time to enquiries, website structure, use of video content, and many more. Without this, asking the model a question like "what's the average conversion rate for a $1,500 suit" would throw $5,000 garment data into the same pool and produce meaningless results.

Noise removal. This was the primary bottleneck that AI advancements finally made more addressable (to a certain point). Because our clients operate across different countries, markets, income levels, and price points, we had to create a baseline sales equation — adjusting every account's historical performance for adflation, inflation, market saturation, and currency fluctuations across historical time periods. We adjusted for seasonality. For tracking outages. For website downtime. For political events, economic shocks, and in some cases, weather catastrophes. For store relocations. For renovation periods that disrupted foot traffic. For wars. These were encoded as exogenous controls/intervention variables with confidence-weighted adjustments.

Around three dozen distinct adjustment types in total, each carrying its own coefficient of trustworthiness to produce a dataset where every account is comparable to every other. AI was incredibly helpful to accelerate anomaly and outlier detection, tagging and scenario classification.

The regression model. With clean, normalised data in hand, we built a regression-based marketing mix model with lag, decay, calibration, and normalisation layers. Every platform, every activity, every touchpoint is assigned a coefficient representing how much it contributes to the end result. We isolated baseline sales: "the intercept" representing organic demand without any media activity. Referrals, walk-in visitors, word of mouth. For accounts where a baseline couldn't be determined from existing data, we created synthetic baseline estimates, only where necessary and treated as lower-confidence estimates, not a hard truth.

Carryover effect. We built a separate model to account for the fact that advertising doesn't stop working the moment you stop spending. Using decay rates and lagged variables, we were able to determine how much advertising effort persists over time after a campaign ends and factor that into the overall model rather than treating every period as isolated.

Preventing overfitting. We used multiple regression types and evaluated model quality using adjusted R-squared and Mean Absolute Percentage Error — MAPE — to make sure we were building something that reflects reality rather than just fitting the data we had.

Calibration. Finally, and this is where we got genuinely fortunate, several of our clients operate multiple locations across different states or countries. That allowed us to run holdout studies: running activity A in one location, activity B in another, changing one variable at a time and observing over months.

This allowed us to calibrate our model to significantly increase the statistical significance and trustworthiness of the data.

What We Can Now Actually Ask — And Evaluate

Here's where it gets genuinely exciting.

This model can now be used to evaluate directionally questions that have never had reliable answers in our industry before.

Questions like:

• How much do brands with long-form YouTube content outperform those without — in conversion rate, organic traffic, and average purchase value?

• What metrics can be tracked today that reliably predict appointments weeks or months down the line, outside of conventional attribution windows?

• What's the optimal video length for our specific audience?

• How should the budget be split across platforms for maximum return?

• What type of appointment setup — Calendly, simple form, phone call, deposit requirement — drives the best conversion?

• What's the real delay between someone first seeing your ad and actually becoming a client?

• Are your brand awareness campaigns generating clients you wouldn't have acquired anyway — or are you overpaying for people who would have found you regardless?

Aaaand many more questions.

None of these questions could be reliably answered before. Now they can be, without a 100% guarantee, but with directional confidence, clear statistical significance indicators, and the intellectual honesty to tell you when the data isn't strong enough to draw a conclusion.

What's Coming Next

If you've read this far, thank you. Genuinely.

I wanted to share this not as a polished announcement, but as an honest account of what this project actually involved and why I'm so genuinely excited about what it can deliver for our sartorial industry.

Over the coming weeks, I'll be sharing the actual insights — specific, actionable, data-backed findings on what works and what doesn't for businesses in the custom suit and tailoring space.

Real answers or at least half-answers to questions you've been guessing at for years.

Stay tuned. It's going to be worth it.

And if in the meantime you want to talk about your specific business, your marketing, your client acquisition, your numbers, and see how you can generate more appointments, reach out and let's have a conversation. A free discovery call, no pressure, just a look at where you are and where you could be.

Schedule a discovery call here: https://calendly.com/sartorialdigital/discovery-call

To your success,

Andris