Should You Handle Attribution In-House or Hire a Consultancy?

Linda Orr

5 days ago

7 min read

Updated: 1 hour ago

Short answer: Keep day-to-day campaign reporting in-house, and bring in an outside expert for the parts that need statistical judgment: marketing mix modeling, incrementality testing and the question of which channels actually drive revenue. Most companies between $5 million and $100 million in revenue don't have enough measurement work to justify a full-time analytics lead with modeling skills, but they spend enough on media that getting attribution wrong costs real money. A hybrid setup, where your team owns the dashboards and a consultant builds the model and sets the rules, is usually the best value. Go fully in-house once measurement questions come up every week and you can hire someone who has built models before.

I get this question from CEOs and CFOs who are tired of platform reports that add up to more revenue than the company booked. They want to know whether to hire an analyst, buy a tool, or call someone like me. I have worked with more than 100 companies, from $5 million to $500 million in revenue, and I built Upwork's first enterprise marketing mix model across an $88 million annual media portfolio. I spent about 20 years as a marketing professor before that, so I have seen both how the statistics work and how they get misused.

Here is how I think through the decision with clients.

1. What does "attribution" actually cover?

Attribution is a loose word. Some people mean the last-click report in Google Analytics. Others mean multi-touch attribution software that assigns fractional credit across touchpoints. Others mean marketing mix modeling, which uses spend and outcome data over time to estimate what each channel contributes, including channels that never get a click.

These are very different jobs. Pulling reports from GA4, Google Ads and Meta into a dashboard is analyst work, and an in-house person can do it well with tools like Supermetrics. Deciding how much credit awareness spend deserves, or whether your branded search is just catching people who would have bought anyway, is modeling work. The first is about plumbing. The second is about judgment, and that is where most in-house setups run out of road.

2. Why can't we just trust the platform numbers?

Because each platform takes credit for the same customer. A shopper sees a Meta ad, searches your brand on Google, clicks a retargeting ad and buys. Meta, Google and your retargeting vendor can all claim that sale. Add up the platform-reported conversions and you often get far more than your actual orders.

The research backs this up. In 15 large advertising experiments at Facebook, published in Marketing Science in 2019, the standard ways of measuring ad results were often far off, and in about half of the studies the estimated lift in purchases was off by a factor of three. When eBay ran a large field experiment and turned off its brand search ads, it kept 99.5 percent of that click traffic through its free listings. The same study found non-brand search returning negative 63 percent, while standard regression models had put the return above 4,100 percent.

The opposite problem hits awareness spend. Podcasts, connected TV, YouTube and upper-funnel social rarely get the last click, so click-based attribution undercounts them. Companies then cut the channels that were feeding the pipeline and wonder why branded search volume drops a few months later. I wrote more about this in Marketing Attribution Is Broken.

3. When does keeping it in-house make sense?

In-house works when you have three things: enough spend and channels to keep a measurement person busy every week, a person who has actually built and validated models before, and leadership that will act on what the model says even when it is uncomfortable.

If you are spending heavily across many channels and making budget decisions monthly, a senior in-house analytics lead can pay for themselves. The catch is that people with real modeling experience and marketing judgment are expensive and hard to hire, and a junior analyst with a Python course is a different hire entirely. I have seen plenty of in-house models that ran without errors and still produced answers nobody should trust.

4. When is a consultancy the better choice?

A consultant is the better fit when you need the modeling done right once, then refreshed a few times a year. That describes most companies under $100 million in revenue. You don't need a full-time econometrician. You need someone to build the model, set up the data correctly, explain the results to your leadership team and train your people to keep the reporting current.

It also makes sense when the stakes are high and you need an outside view. If you are about to cut a channel, switch agencies or set next year's budget, an independent read protects you from internal politics and from vendors grading their own homework.

5. What does a good hybrid setup look like?

In the setups I like best, your team owns the daily and weekly reporting: spend, cost per acquisition, conversion rate, pipeline. They keep the data clean and the dashboards current. I build the marketing mix model, design any incrementality tests, and set the rules for how the team reads platform data against the model. We refresh the model quarterly or twice a year, and I am available for the big budget questions in between.

This keeps your fixed costs low and puts the expensive expertise where it matters. It also means your team learns how the model works, so they stop treating it as a black box.

That second part matters more than most people think. Gartner found that marketing analytics influences only 53 percent of marketing decisions, and Nielsen's 2024 survey of nearly 2,000 marketers found that 84 percent felt confident in their ROI measurement while only 38 percent measured traditional and digital channels together. A model only pays for itself when it changes where the money goes.

6. Can't we just buy an MMM or attribution tool?

You can, and some are useful. The problem is how they get used. Many MMM and multi-touch attribution platforms, and now AI tools, work by throwing the data in and seeing what comes out. A model can be made to say almost anything depending on how you cut the sample, which variables you include, and whether you allow for diminishing returns or interaction effects. Garbage in, garbage out.

Real statistics starts with a hypothesis. Before I build anything, I do the qualitative work: seasonality, weather if it matters to your category, competitor moves, branding activity, pricing changes and promotions. Then I test whether the data is clean and representative. Open-source tools like Google's Meridian and Meta's Robyn are good engines, but they still need someone who knows what to put in and how to read what comes out. I compared them in Meridian vs. Robyn vs. an MMM Expert.

7. What does it cost to go each route?

A full-time senior analytics hire with modeling experience is a salary plus benefits, plus the tools and data connectors they need, every year. A consultant-built model is a project cost. My marketing mix modeling work typically runs $8,000 to $40,000 depending on the number of channels, data quality and how many markets we model. If you want a review of your current tracking and reporting first, a full Marketing Performance Audit starts at $15,000, and an audit plus marketing mix model starts at $20,000. For smaller questions, I bill $475 an hour.

Put those numbers next to what is at stake. McKinsey estimates that 15 to 20 percent of marketing spend can be freed up through better measurement of marketing return. On a $500,000 annual budget, that is $75,000 to $100,000 a year you could move to what works. If a $20,000 model redirects even 10 percent of that budget, it returns $50,000 in the first year and keeps paying in the years after. The price looks different once you compare it to the cost of guessing.

If you want me involved on an ongoing basis, leading marketing as well as measurement, fractional CMO retainers start at $8,000 a month, and most engagements run $12,000 to $20,000. See my pricing page for the full list.

8. What should we have in place before calling anyone?

Get your conversion tracking working and consistent. Make sure revenue or leads in your CRM can be matched to dates and, ideally, to sources. Pull at least two years of weekly spend by channel if you can. Write down the big events: price changes, product launches, promotions, site migrations, agency switches. That timeline saves a lot of time and money once modeling starts.

If you are not sure where you stand, take my free Marketing Measurement Scorecard. It takes a few minutes and shows where the gaps are.

9. What is the risk of going cheap on this?

Most of my clients tried a cheaper option first. Some used a junior analyst. Some trusted the attribution report from their agency, which was grading its own work. Some bought a tool and never got anyone to interpret it. The cost shows up later as budget moved to the wrong channels for a year or more. A cheaper route makes sense if you are just starting out and spending modestly. Once you are making six- or seven-figure media decisions, the measurement should be built to hold up.

Next step

Whichever route you choose, attribution is only as good as the tracking behind it. My GA4 and measurement audit checks that foundation first and starts at $3,500.

If you want to talk through whether in-house, consultant or hybrid makes sense for your company, book a free 30-minute call. You can also read more about my Marketing Mix Modeling and Analytics work.