Building Multi-touch Marketing Attribution Engine

Tutorful is one of the UK’s leading tutoring marketplaces, connecting students with tutors across every subject and level. The platform operates exclusively in the UK and manages a high-volume, multi-step customer journey — from a first search to a registered account, an enquiry submitted, and eventually a completed lesson.

For a marketplace business, understanding how students find you — and what nudges them from a first visit to a confirmed lesson — is the difference between efficient growth and expensive guesswork. Tutorful’s marketing team was spending across paid search, social, CRM, and organic channels. With an inconsistent last-click attribution layer, every channel told a different story. The question that came up week after week in performance reviews was simple: if we invest in marketing, where does that investment actually show up?

The stakes are higher than they might appear. Without reliable attribution, a marketing team isn’t just missing a reporting metric — they’re guessing on budget allocation. Spend is being directed without knowing which channels, audiences, or campaigns are actually converting. Patterns that could be replicated go unnoticed. Inefficiencies that could be cut keep running. For a business operating at Tutorful’s scale — significant spend spread across multiple active channels — even a small inaccuracy in attribution compounds quickly into material misallocation.

Rather than continuing to invest at their current cost of acquisition without visibility into what was driving it, Tutorful wanted to understand their business and customers more precisely — creating the conditions for sharper spend decisions and a structurally lower cost of acquisition over time.

Tutorful’s marketing reporting had grown organically as the team scaled, with different tools plugged in at different stages and none of them talked to each other.

“We had no BI and relied on a horrendous Google Sheet with 40 queries and eight different linked tables for all business insights. It not only kept on breaking, but there was a significant lack of trust that anything in there was true.”

— Neil, Head of Commercial Insights, Tutorful

 

 

1. First-touch only attribution. Amplitude — the primary analytics tool — attributed all credit to whichever channel first introduced a user, regardless of what happened over the following weeks. Channels that influenced users mid-journey, like CRM and organic search, were effectively invisible.
2. No consistent channel definitions. Paid media platforms, Braze, and GA4 each applied their own attribution logic. The same week’s performance looked materially different depending on which tool you opened.
3. A reporting spreadsheet nobody fully owned. The weekly GMV report — used in board-level Monday morning reviews — applied logic that hadn’t been revisited since 2023. Known inaccuracies remained unresolved because no single person fully understood how it worked.
4. A customer journey that doesn’t fit simple models. Users often interact across multiple devices and over weeks before converting. Attribution needs to capture the full sequence — which channels introduced them, kept them engaged, and ultimately influenced their decision — so credit reflects the moments that actually mattered, not just the last click.
5. Low confidence in reported numbers. Despite operating exclusively in the UK, analytics showed a meaningful volume of international traffic inflating performance numbers and reducing confidence in reported conversion rates.

 

Tutorful had already invested in an internal data resource and made a genuine attempt to build this themselves. It was a reasonable instinct — but the scope of the problem was bigger than a single hire could carry. Multi-touch attribution done properly spans several distinct disciplines: data engineering to build and maintain the pipeline infrastructure, analytics engineering to model and test the business logic, and data analysis to translate the output into commercial decisions.

It’s a pattern we see often. The decision to bring in an external team wasn’t a failure — it was the recognition that this class of problem is built for a specialist team working in focused sprints, not a single generalist hire working indefinitely (if you’re considering solving your attribution problem by hiring one data person, it’s worth being honest about what that role can realistically own).

Before any build work began, we ran alignment sessions with the marketing leadership and commercial team to agree on what the attribution model actually needed to do:

• Which conversion events mattered
• What counted as a meaningful marketing touch
• What questions the business needed to answer on a weekly basis

That brief became the foundation for everything built in the sprints that followed.

Most marketing teams accumulate data the same way — tool by tool, as the team grows. The result is a stack where nothing was designed to connect. Before any attribution model could be built, Tasman consolidated Tutorful’s marketing data into a single warehouse, automated ingestion from paid and CRM channels, and built a transformation layer that kept business logic clean, documented, and easy to maintain. The infrastructure was built to be fully owned and built upon by a lean team — not to require ongoing engineering support to keep running. “It’s given us a base to start pulling all of our other reporting so we can start to understand other parts of the business better — beyond attribution,” said Neil.

Attribution sounds like a reporting problem. It’s actually a data engineering problem — and one that requires deep familiarity with how the business actually works. Tasman mapped Tutorful’s full channel mix, understood which steps in their four-stage customer journey carried genuine purchase intent, and built the attribution logic from the ground up: what counts as a meaningful touch, how much weight each stage should carry, and how to connect behaviour across devices and sessions before a user ever registers.

That logic — not just the infrastructure holding it — is what makes the output trustworthy. The engine was built to support multiple attribution models, giving Tutorful the flexibility to compare approaches as their measurement strategy matures.

“The marketing attribution model is going to add real value. It’s either going to give us confidence that we’re spending money in the right place, or it’s going to tell us where we should be spending money — and how we drive our CRM campaigns.”

— Neil, Head of Commercial Insights, Tutorful

The engagement closed with two live dashboards: one for attribution — channel contribution across the funnel, filterable by campaign, time period, or model — and one for weekly operational performance covering spend, sessions, conversions, and revenue. The weekly GMV review that had previously relied on a manually maintained spreadsheet now ran from a single consistent dataset. Handover was built into the engagement from the start, with the platform documented, tested, and handed over through working sessions — not dropped at the end.

The goal from sprint one was to build something Tutorful could own without Tasman. That shaped every decision — from how the infrastructure was documented to how the attribution logic was configured — so that when the engagement closed, the platform didn’t close with it.

Seed-driven configuration means attribution models and lookback windows can be adjusted without touching SQL. Centralised UTM logic keeps channel definitions consistent as the team and campaigns evolve. Terraform-managed infrastructure means nothing lives only in someone’s memory. A lean data team can run, extend, and adapt the platform without needing to bring in outside support every time something changes.

That commitment extended further than the platform itself. When Tutorful moved to bring in a dedicated data leader, Tasman supported the hiring process — helping shape the job description for the role. It’s not something that appears in a standard statement of work, but it reflects how Tasman thinks about long-term client success: the platform is only as valuable as the person running it, and getting that hire right was part of getting the handover right. It’s an example of what Tasman offers that most data consultancies don’t.

That commitment extended further than the platform itself. When Tutorful decided to bring in a dedicated data engineer, Tasman supported the hiring process — helping shape the job description, leading on candidate selection, and recommending the final candidate. It’s not something that appears in a standard statement of work, but it reflects how Tasman thinks about long-term client success: the platform is only as valuable as the person running it, and finding the right hire was part of getting the handover right.

Neil has been on both sides of this question — weighing up the investment, and seeing with the results after. Here’s how he’d suggest for anyone in a similar situation to approach the decision:

1. How much would it cost you to do this internally — and what are you sacrificing to get there? Account for the full picture: analyst hours, management time, and everything your team isn’t doing while they work on this. If you outsource it, what can that headcount go towards instead?
2. What would a freelancer cost — and what are you actually getting? A freelancer is one person with one skill set. This kind of work needs data engineering, analytics engineering, and commercial analysis at different stages. A freelancer who covers all three credibly is rare, and a generalist who attempts it will show the seams.
3. What would a full-time hire cost — and how long would it take them? A new FTE faces the same jack-of-all-trades problem, plus the additional drag of a longer timeline. A project that takes a specialist team four sprints could take a solo hire the better part of a year — if they get there at all.
4. What is the opportunity cost of not doing it? If you’re asking the question, you already have doubts about how well your marketing budget is working. Benchmark yourself honestly against the industry standard: at your current cost of acquisition, what does it cost to hit your growth targets? If better attribution brings that number down — even modestly — the gap is money you’re currently leaving on the table.

“For Tutorful, outsourcing to Tasman was money well spent. It sets me up to get a bigger return on investment because I’m in a much better position to move forward. Those tasks that used to take me a week are now going to take me a day. Those tasks that used to take me a day — I don’t even need to touch, because it’s all set up.”

— Neil, Head of Commercial Insights, Tutorful