Most attribution software was built for online stores and priced for teams with analysts. If you close on the phone, in chat, or in a CRM, these five questions decide whether a tool can even see your business, and whether you can afford to keep it.
We sell attribution software. This is the list we wish every buyer brought to a demo, including ours. It is written for small and mid-sized teams that do not have a data department, and especially for businesses whose revenue does not all land in a checkout.
If you run a clinic, dealership, law firm, trade business, or a B2B company that books on calls and forms, start at question one. If you are a simple one-channel Shopify store, you may not need this category yet, and the section near the end explains why.
Print this for the demo
- What conversions can this tool see, and what stays invisible to it?
- How does it recognise a returning customer, and what happens when it cannot?
- Can I compare attribution models side by side, or is it a black box?
- Will your numbers match Google and Meta, and will you explain the gap in writing?
- Can I export the underlying journeys, and where is your cancellation policy?
Attribution software decides how credit for a sale gets divided between the ads, emails, searches and conversations that led to it. Getting that division wrong means moving real budget in the wrong direction, so the demo deserves more than forty minutes of dashboards.
1. What conversions can this tool actually see, and what stays invisible to it?
Start here, because everything else is downstream of it. A tool that cannot see a conversion will never attribute it, no matter how sophisticated its modelling is.
Picture a common journey. Someone clicks your ad, reads your pricing page, leaves, and calls you three days later to book. A tool built around a checkout never records that sale at all. It does not turn into “Direct” or “unattributed”; it simply vanishes, and the campaign that earned it looks like it did nothing.
Many attribution tools were built for online stores, where the conversion is a checkout. If your revenue arrives that way, the assumption is invisible and harmless. If a meaningful share of it arrives by phone call, form submission, live chat, quote or a signature in person, a checkout-shaped tool is measuring a minority of your business and presenting it as the whole picture.
Ask the vendor to list every conversion type the tool ingests, and specifically whether phone calls, form fills, chat conversations and offline or in-person sales can be attributed alongside online orders. Then ask what happens to a sale that closes in a way the tool does not support.
What good looks like: calls, chat, forms and offline sales attributed alongside online orders, in the same reports, without a separate tool.
2. How does it recognise the same customer across visits, and what does it do when it cannot?
Customers rarely buy on the first visit. To connect an ad someone clicked in March to a purchase in April, a tool has to recognise that both were the same person. That is harder than it used to be.
Third-party cookies, the small files that let one website recognise a visitor who came from another, have been restricted for years in Safari and Firefox. Google reversed its plan to remove them from Chrome in July 2024 and has since retired most of the replacement technology it was building, so the picture is messier rather than settled. On mobile, Apple’s App Tracking Transparency rules require apps to ask permission before tracking activity across other companies’ apps and sites, and most users decline.
You do not need to understand the plumbing. You need two answers in plain language. First, what the tool relies on to recognise a returning customer. Second, and this is the part that gets skipped, what it does when that fails. Most tools fall back to a statistical estimate, meaning the software is making an informed guess about which visits belong together rather than knowing.
A guess is not automatically bad. Guessing without telling you is. AppsFlyer’s public documentation is a reasonable example of the right altitude: it separates matching based on a real identifier from probabilistic modelling, that same informed guess, used only when matching fails, and it says plainly that the probabilistic result is a campaign-level estimate rather than a per-person match. Ask for that distinction in writing.
What good looks like: first-party tracking, plus an honest answer about where recognition stops and estimation starts.
3. Can you see why a channel got credit, or is the model a black box?
Every attribution model is an opinion about something that never happened. It is trying to answer “would this sale have occurred anyway,” and no dataset contains that answer.
This is not a criticism of the field, it is the field. The foundational academic work on multi-touch attribution, meaning any approach that splits credit across several steps in a journey rather than handing it all to one click, goes back to a 2011 paper by Xuhui Shao and Lexin Li presented at the KDD conference. It frames the problem as dividing credit statistically rather than measuring cause. The most direct check we have on how well that works came from a 2019 study in the journal Marketing Science by Brett Gordon, Florian Zettelmeyer, Neha Bhargava and Dan Chapsky, which compared fifteen real randomised advertising experiments at Facebook against the observational methods attribution tools use. The observational methods regularly disagreed with the experimental result.
So the useful question is not which model is right. It is whether you can inspect the reasoning. Ask whether you can see the customer journeys behind a number, whether you can compare more than one model side by side, and whether the tool supports any kind of holdout or geographic test, where you deliberately withhold spend from one group or region to see what changes.
Comparing models side by side is the anti-black-box test. A channel that looks weak under last-click, where the final click before a purchase takes all the credit, often looks essential under a model that credits earlier steps. Seeing several models at once tells you whether a channel is genuinely underperforming or simply being penalised by the assumption you happen to be using. Our guide to attribution models walks through how each one distributes credit.
What good looks like: several models side by side, readable by someone who is not an analyst, with the journeys behind each number one click away.
4. Will its numbers match Google and Meta, and does the vendor admit upfront that they will not?
They will not match. Any vendor implying otherwise is either confused or selling you something.
Google’s own advertising documentation tells advertisers to expect discrepancies between Google Ads, other platforms and internal reporting, and lists ordinary mechanical reasons: conversions counted at different times, different attribution windows (how many days after someone sees or clicks an ad a sale still counts toward it), traffic filtered out after the fact. Meta’s server-side tracking grades its own confidence that an event it received matches a real person, on a scale from zero to ten. That is a platform acknowledging its own matching is probabilistic.
There is also an incentive problem worth naming. When an ad platform reports on how well its own advertising worked, it is marking its own homework. An independent tool has its own biases, but at least they point in a different direction.
What separates a good vendor from a bad one is whether they tell you this before you sign or after your first confusing month. Northbeam, a competitor of ours, documents plainly for its customers that its numbers will differ from the ad platforms and explains the mechanics of why. That is the behaviour to look for. Ask directly: “will your numbers match what Meta reports, and if not, will you explain the gap in writing?”
What good looks like: a written explanation of why the numbers differ from the platforms, offered before you sign rather than extracted afterwards.
5. Who owns the data, and can you take it with you?
Attribution data compounds. A tool that has watched your customers for two years is more useful than one that started last week, because long journeys and repeat-purchase patterns only become visible with time. That accumulated history is exactly what makes leaving painful, which is why you should ask about the exit before you sign.
Ask three things. Can you export the underlying journey and conversion data, not just a summary report or a screenshot of a dashboard? Is there a way to move that data somewhere you control? And what happens to your data if you cancel, including how long it is retained and whether you can retrieve it afterwards?
Good tools answer this comfortably because portability is a feature they are proud of. Google Analytics 4 offers a free daily export to BigQuery up to a million events a day on its standard tier. Triple Whale documents CSV export, a public API and a warehouse export. Vendors who become vague here are describing a switching cost, not a product.
Privacy rules make this practical rather than theoretical. California residents have had the right to opt out of having their personal information shared for cross-context advertising since January 2023. If a customer exercises that right, you need to know your tool can act on it, which means knowing who controls the data in the first place.
What good looks like: export of the journeys themselves, not screenshots, and a cancellation and retention policy you can read before you buy.
Price: the question that bites after the demo
Price is not one of the five because you can see it during evaluation. It still decides whether you keep the tool, so ask about it with the same care.
The common pattern in this category is pricing that scales with something you are trying to grow: a percentage of ad spend, a count of orders, tracked events or contacts. That feels fair on day one and turns a good year into a bad invoice. Ask for the exact number your bill keys off, then model it against where you want the business to be in eighteen months, not where it is today.
What good looks like: a price that does not jump because you had a good quarter, published clearly enough that you can do that modelling before the call. See how we price ours.
Ask about setup the same way. Some tools need engineering time and a months-long implementation before you see a number. “We will help you onboard” is not the same as “it is running this week.”
When you may not need this yet
Attribution models need volume to say anything trustworthy. Google recommends at least 200 conversions and 2,000 ad interactions within a 30-day period before its own data-driven model is considered reliable. That is a reasonable rough marker for the field. Well below it, any multi-touch model is working from too little signal, and a paid tool will produce a confident-looking answer built on noise.
A store with one channel and a clean checkout can get far with Shopify’s own five attribution models and disciplined UTMs, the campaign tags you add to links so a visit can be traced to the ad that sent it. The case for a dedicated tool is coverage (calls, chat, offline sales and long return visits) and independence from the platforms that are marking their own homework.
Confidence in measurement is not high across the industry either. In a survey of 196 United States marketing professionals fielded by EMARKETER with TransUnion in July 2025, 60% said internal stakeholders question the validity of their metrics at least sometimes. Buying software does not resolve that on its own, and buying it too early tends to make it worse, because now there is a second set of numbers to argue about.
How Attriqs answers the five
You should hold us to the same list. Here is our answer to each, including the parts that are less flattering.
1. Coverage. This is the question we built the product around. Call tracking with dynamic number insertion, chat attribution, form tracking and offline conversion import are on every plan, reported alongside online orders rather than in a separate tool. Our call tracking guide shows how a phone call gets tied back to the campaign that produced it.
2. Identity. First-party tracking and privacy controls are on every plan. It is also the part we describe least in public, because it is the part competitors would most like to copy. In a demo, ask to walk through real journeys in the journey browser rather than a curated sample, and ask us where recognition stops.
3. Models. All six attribution models run side by side from the Amplify plan upward. Essentials includes fewer, so if comparing models matters to you on day one, that is the plan to start on. Say so in the demo and we will not pretend otherwise.
4. The platform gap. Our numbers will not match Meta or Google either. We have written up why reported and incremental return differ, and we would rather you read it before you sign than after your first confusing month.
5. Data and price. We price for small and mid-sized teams, in flat monthly plans sized to your traffic and ad-spend range rather than as a percentage of what you spend, so a bigger ad budget does not automatically mean a bigger software bill. Every plan carries the full platform, including marketing mix modeling; plans differ by capacity, users and how many models you can compare. Setup is a single script on your site, not a three-month implementation. You should still ask us for the export path and the cancellation policy in writing. Both are published on our pricing page: your data is kept for 30 days after cancelling and you can export all of it before then. If any vendor cannot show you both, walk.
Frequently asked questions
Is attribution software only for ecommerce?
No, though a lot of it was built as if it were. A conversion can be a phone call, a form fill, a chat, a booked appointment or a sale signed in person. If those make up a meaningful share of your revenue, coverage is the first thing to check, because a tool built around a checkout will not see them at all.
How is this different from call tracking alone?
Call tracking tells you which source produced a phone call. Attribution connects that call to the rest of the customer’s journey and to the revenue it produced, alongside every other channel. That lets you compare a campaign that drives calls against one that drives online sales on the same footing, and see which marketing actually earned revenue rather than which produced activity.
What is the difference between attribution software and web analytics?
Web analytics tools describe what happened on your site: how many people visited, which pages they saw, how many converted. Attribution software assigns credit for revenue across the marketing that led to it, usually across multiple sessions and channels, and often pulls in ad spend so you can compare cost against return. There is real overlap, and Google Analytics 4 does some of both. Our comparison of Attriqs and GA4 covers where the line falls in practice.
Do I need attribution software if I only advertise on one channel?
Usually not yet. If nearly all your spend and nearly all your conversions run through one platform, that platform’s own reporting will tell you most of what you need. The case for an independent tool strengthens as you add channels and as more of your revenue closes somewhere other than a checkout.
Why do attribution tools always report lower numbers than Meta or Google?
Mostly because they count differently rather than because one is wrong. Ad platforms use their own attribution windows and can credit conversions from people who saw an ad without clicking it, while independent tools generally rely on activity they can trace themselves. The platforms are also reporting on the performance of their own product. A gap is expected. A vendor who cannot explain the direction and rough size of the gap is the actual warning sign.
How long before attribution software is useful?
Expect a lag. The tool has to observe complete customer journeys before it can describe them, so if your typical path from first visit to purchase takes six weeks, the first six weeks of data will be partial by definition. Ask any vendor how long until the numbers settle for a business with your sales cycle, and be suspicious of an answer shorter than one full cycle.
Is multi-touch attribution better than marketing mix modeling?
They answer different questions. Multi-touch attribution follows individual journeys and is good at comparing campaigns and channels in detail. Marketing mix modeling works from aggregate spend and revenue over time, needs no individual tracking at all, and is therefore more resilient as tracking gets harder, but it is coarser. Treat agreement between the two as the stronger signal.
Bring this list to a demo
If you take one thing from this, make it the first question. Coverage is the failure that cannot be fixed later. Price can be renegotiated and a clumsy setup can be redone, but a tool that structurally cannot see how your customers buy will still not see it in a year.
Bring the list to every demo, including ours. We will spend 20 minutes walking through how Attriqs answers each of the five, not giving you a product tour. Book the walkthrough.