Episode 10 – Transcript
Can you trust your Amazon numbers? Silent data errors
that make growth look better or worse than it is
Host: Lucas Bassa (Channable)
Guest: Bhavesh Tailor (The Marketplace Guys)
Lucas Bassa (00:01)
Hello everyone, welcome to a new episode of the Marketplace Growth Podcast. Today I'm
joined by Bhavesh from The Marketplace Guys, a Germany-based agency, and we're talking
about a very topical subject: can you trust your marketplace data? And at what point will you
be able to, if not already? Bhavesh, before we get going, please introduce yourself.
Bhavesh Tailor (00:29)
Yeah, my name is Bhavesh. I'm from London originally, but I've been living in Hamburg,
Germany, for the last six years. Last November I founded my company, The Marketplace Guys,
together with my business partner, Alexander Schnelle. Before that, I was team lead for
Amazon at SNOCKS.
Lucas Bassa (00:50)
Remind me, what do they sell?
Bhavesh Tailor (00:54)
SNOCKS is a DTC brand that does boxer shorts and socks, but also women's underwear and
general basic clothing. It's quite big in Germany.
Lucas Bassa (01:04)
Yeah, I know the brand. It's quite a big brand, and fashion is quite challenging, so I imagine
you have some solid experience from your time at SNOCKS. First of all, why is this such an
important topic for you?
Bhavesh Tailor (01:24)
With Q4 around the corner, the big numbers are starting to grow now with Black Friday, and
previously Prime Day, so trusting your numbers coming into the end of the year can be a bit
tricky. Especially in my experience with Amazon, there are a bunch of red herrings in what
you've got to look at, and a lot of it is experience-based. When it comes to trusting numbers
on Amazon, you've got a bunch of different problems: attribution time, how Amazon books
the revenue in Seller Central, how third-party P&L tools book the numbers. You can't always
trust everything that you see directly.
Lucas Bassa (02:11)
Yeah, and that's one of the things with Amazon data, and with platform data in general. The
same applies to Meta and Google: if that's your only source for the data at hand, you should
be a bit sceptical about some reports. I know, for example, that Meta and Google will always
slightly colour the data to make it look like they're key in your whole marketing funnel. Does
the same apply to Amazon?
Bhavesh Tailor (02:45)
Yes, to a certain degree. For example, in the agency we always apply a 14-day grace period to
the ads data we look at, because the numbers can change marginally or massively
depending on the account size. So if you look at the ACOS, sales or spend numbers, they're
not always quite 100%. For normal days it's usually okay, but for days like Black Friday or
Prime Day you definitely need that grace period before you can do a proper analysis and
evaluation of how successful those deal days were.
Lucas Bassa (03:24)
Okay. So to clarify, we're talking about the trust issue with data, especially on Amazon: how it
can be trusted, what data to look at and how to work with the data so it can be trusted. But
first, these discrepancies and delays in the data that you're describing: how do they affect
day-to-day analytics? One example I can think of: around Black Friday, most sellers look at
their ads every hour, if not all day, tweaking and adjusting throughout the day. Is that
something you shouldn't do, because the data hasn't matured yet?
Bhavesh Tailor (04:19)
It depends on your data access. From an ads point of view, if you've got access to Amazon
Marketing Stream and you're looking at it properly, you should definitely be able to trust that.
If you're looking within the ad console itself, I still think that if your ACOS is going through
the roof, you've got to optimise. There's no way around it. In my time at SNOCKS, if you
increased the bid by 10 cents, the spend would go through the roof, so you've got to keep
that in mind. I wouldn't say you can't trust day-level data or can't make optimisations based
on it. It's more on the analysis side at the end where you really have to ask: can I trust it at the
moment or not?
Lucas Bassa (05:09)
So it's accurate enough to drive ad hoc decisions, but for the full picture, the post-Black
Friday analysis, you should wait two weeks.
Bhavesh Tailor (05:24)
Yeah, I'd usually say two weeks. I do think there's a big difference from Google and Meta,
though. Because people purchase on Amazon and the ads are built into its own ecosystem,
it's a lot easier to trust the data on Amazon than it is on Google or Meta. That's why I think the
optimisation cycles are a lot quicker on Amazon during these periods. You don't have this
whole idea of shifting budgets on Amazon like you would with Google or Meta.
Lucas Bassa (06:01)
So for Amazon, because it's all on-platform data, it's affected by fewer variables, and you can
take it at face value.
Bhavesh Tailor
Yeah, yeah.
Lucas Bassa (06:13)
Okay. So that's one example: managing your campaigns and leaving a grace period between
the campaign's peak moment and analysing its performance. What are other areas where
data should be looked at differently, or can't really be trusted?
Bhavesh Tailor (06:42)
This one's the most interesting point for me: the sessions within the account. With our
clients, we do a weekly analysis of the previous week compared to the week before, and then
compared to the same calendar week of the previous year, for example calendar week 42 of
2025. Today's a Monday, and I believe we're in calendar week 43 as we're filming this. When I
do the analysis on a Monday, the session data on Amazon isn't complete yet. So when I
compare the overall week to the previous year, it will always say we have fewer sessions than
last year. On a weekly level, it needs maybe a couple of days to attribute properly. But for a
Prime Day or a Black Friday, it can take up to seven days, depending on the account size as
well. However, session data isn't the be-all and end-all of how good the traffic in your account
is. Amazon gives you some brilliant data with the SQP report to see how searches are
performing and developing in general, because that's data straight from the horse's mouth,
not from a third-party tool. If you make those comparisons on a weekly basis, let the sessions
attribute fully and pull the SQP data once it's been fully generated, then you've got a single
source of truth that should be 100% accurate.
Lucas Bassa (08:15)
So you're saying sessions don't fully attribute for one to maybe seven days. That means if you
compare a Sunday-to-Sunday date range on a Monday, you might already have a gap in the
data.
Bhavesh Tailor (08:36)
For normal standard weeks it's just one working day. It's more for the big deal events, where
from my experience it takes more than three or four days.
Lucas Bassa (08:46)
So initially, if you look at the numbers, it might look like your spend is relatively high, whereas
a week later all those sales have been attributed correctly and suddenly the balance looks
more positive. Right?
Bhavesh Tailor (09:04)
More on the account side than the ad side. On the account side, I'm only talking about the
session attribution you see in the business reports, not what you'd see on the ad side.
Lucas Bassa (09:16)
But is your spend still involved there?
Bhavesh Tailor (09:21)
No. On the Amazon side, the business reports don't show your ad spend. You'd have to bring
that data together afterwards.
Lucas Bassa (09:25)
Exactly. But if you do that, and you compare your spend for the week with your sales for the
week, and those sales aren't fully in yet...
Bhavesh Tailor (09:40)
The sales are usually okay on a weekly basis. What I was talking about is predominantly the
sessions.
Lucas Bassa (09:45)
Okay, the sessions. So it affects your underlying metrics, like conversion rates and
click-through rates.
Bhavesh Tailor
Yeah, how much traffic you're getting exactly.
Lucas Bassa
So when you're doing a product analysis or analysing your listings, that really plays a part.
And you mentioned the SQP report. Can you elaborate on how you use it to get around this?
Bhavesh Tailor (10:12)
Yeah, that's the Search Query Performance report. You have two options there. You can pull
brand-level data, so the best-performing keywords at brand level, or ASIN-level data. You can
pull keyword data at ASIN level, as long as it's attributing enough sales, on a weekly basis
over the last 64 weeks, I believe. So let's take cranberries as an example. You can look at the
keyword cranberries in calendar week 42 of 2026 and see, from Amazon, the number of
searches it's had, the number of clicks you've had and the number of add-to-baskets you've
had. Then you can compare that with the same ASIN in the same calendar week of 2025 and
see that searches for cranberries have decreased by 11% and your click share has decreased
by, say, 3%. Then you can look at that alongside your session data: you're getting 5% or 12%
fewer sessions, searches are 11% lower than the previous year, and your click share has gone
down. Putting all this data together, you can see where you lost sessions, or that something
isn't adding up and you might need to push more. So taking one source at face value isn't
necessarily the right way to go about things on Amazon. That said, I'm really happy with
Amazon in the sense that it gives you all of that data to look across the whole board.
Lucas Bassa (11:51)
Yeah, because that's not the case for most other marketplaces. So you can compare your
sessions with your clicks at keyword level. Are you also able to benchmark against
competitors?
Bhavesh Tailor (12:14)
Not necessarily. They give you a benchmark for the market category, but they don't tell you
how much click share competitor X gets. As far as I know, you can see it at keyword level for
the top three competitors in the brand view, but I'd have to double-check that before stating
it as fact.
Lucas Bassa (12:35)
Where can you find this report? Is it a Sponsored Products report, or is it in your Seller Central
account?
Bhavesh Tailor (12:42)
If you go to Seller Central and look for Brand Analytics, you'll see a section called Search
Catalog Performance, and there you can choose between the SQP report and the Search
Catalog Performance report. Those are absolutely gold.
Lucas Bassa (12:55)
Often you want to know, when you see a dip: is it just me, or is the entire market or category
dipping? Would these reports help pinpoint that kind of movement?
Bhavesh Tailor (13:14)
They definitely can, in terms of whether the market is growing or shrinking. You also have to
take the advertising data into account. Something else that's brilliant about Amazon is that
you can now pull performance reports from further back. Before, you were limited to the last
90 days, and recently they've said you can go even further back. But what you can't do, and
what I always recommend people do themselves, is store your bid data as well: how is your
performance at this bid level? I did a bit of research beforehand. You can pull bid-level data
for the last 90 days via the history report, but before that it's tricky. So if people started saving
their bid data, that would be brilliant, because it lets you measure your CPC development
over the course of a year, or quarter to quarter.
Lucas Bassa (14:17)
So that lets you take seasonality out of the picture. You can compare this year's performance
on cranberries with last year's, and also see how much the bids increased in terms of average
cost per click. But I don't think a lot of companies save this data at the moment. Do you think
they should?
Bhavesh Tailor (14:41)
Yeah. Since we started this agency, I've realised how much difference a well-oiled, really
well-managed database, where the data is truly clean, makes to how you run an account.
These days everybody has an AI and everyone needs to work with an AI to look at their data.
If your data is wrong, then the AI's analysis is going to be wrong. There are no two ways about
it. So you've got to build those safeguards for proper data management, and that can be
expanded across a lot of different areas. A really simple example is Vine on Amazon. If you
use the Vine programme, it automatically attributes the Vine units to the revenue made that
day, but you haven't made any money from those products. So being able to attribute that
data properly in your database makes the AI a lot more efficient at working with it.
Lucas Bassa (15:48)
That's an interesting example. So if you do 100 Vine credits, you've got 100 products that you
basically sold at a loss. You're saying that if you account for that in your database, you can
eliminate those anomalies, and it will stop the AI from seeing and making up trends that
don't exist.
Bhavesh Tailor (16:18)
Exactly. Even with other P&L tools, we've often looked at the numbers and thought: we've
made amazing revenue today, why is that? If you're averaging €3,000 a day and suddenly
your revenue on a Thursday at 2pm is at €5,000, you think, okay, something has definitely
gone wrong there. So it is important to do that. Database-wise, you can save all the data from
your business reports and then link a separate table for Vine. If you just record the Vine
enrolments there, that should be fine for an AI. AIs can read a database so well these days
that it's really about making sure you save the data correctly.
Lucas Bassa (17:07)
But saving the data is still manual work, right? Or do you give Claude access to your Seller
Central?
Bhavesh Tailor (17:14)
At the moment it's still manual. I believe Amazon doesn't provide API access to Vine at the
moment.
Lucas Bassa (17:22)
But they do for advertising.
Bhavesh Tailor
They do, yeah.
Lucas Bassa
Before we go into that: what tools do you use to supplement your marketplace data
collection or analytics, if any?
Bhavesh Tailor (17:43)
When we started, I wasn't that deep into the topic of databases and data warehouses. At the
moment we store our data in Supabase, but I know there are other options on the market
that may be cheaper. Whether you go for a Postgres model or another SQL model is up to
the user, depending on whether you want to pay for data storage or for data requests. For
our model, Postgres makes the most sense. What we have is essentially a database where we
pull every single metric that's interesting for us, rather too much than too little. We pull daily
ad data, for example, and after 14 days we stop re-pulling that day's data. But we keep pulling
the last 14 days continuously...
Lucas Bassa (18:27)
So a rolling 14 days.
Bhavesh Tailor
Exactly, so that we have clean ad data to work from. We do the same with business report
data, so sessions, units and revenue, but we only pull that twice a week for the previous week:
once on the Monday, because our clients still want to see their revenue as soon as it's there,
and then again on the Thursday, so we can make corrections wherever necessary. You can
only work with the data you've got on hand.
Lucas Bassa (19:07)
Is this automated? Do you extract most of it from the API?
Bhavesh Tailor
Yeah.
Lucas Bassa
And does that happen via third-party tooling, or is it tied to your database directly?
Bhavesh Tailor (19:20)
We pull the Amazon data directly from Amazon via the Amazon API. We do use third-party
tools for organic rank tracking, which is also really key.
Lucas Bassa (19:34)
Helium 10, I imagine?
Bhavesh Tailor
No, we use Sold Scope, but there are a couple of cheaper options on the market. I used ASIN
Data Scraper, which was also quite good. I'd definitely recommend those. The reason we do it
is that, based on the SQP data and the session data you get, if your click share or your
searches are going down but your organic ranking is still quite good, maybe it's a question of
your price being too high or your main image being wrong. But once you're organic number
one and paid number one, even if you're losing click share, your options are a bit limited,
depending on your price or whatever else.
Lucas Bassa (20:24)
So what you're also saying is: the more data points you have, the more you can do with them.
If you only collect ad spend and ad sales, it doesn't tell you that much. If you start including
conversion rates, page views or top-of-search percentages, it opens up more possibilities for
your analysis.
Bhavesh Tailor (20:49)
Exactly.
Lucas Bassa (20:52)
Circling back to the topic: you're saying some data can't be trusted, or you need to know how
to look at it for it to be trustworthy. Is there any data manipulation you do when getting it out
of Amazon, to make it solid enough to draw conclusions from?
Bhavesh Tailor (21:15)
Yes, there are a couple of things we do. My business partner always says this: we can only
start trusting a number as soon as it stops changing. That applies to the whole sessions topic
we spoke about earlier. But there's also the difference between the ads data and what you
get in Seller Central. On the ad side everything is net, but on the seller side everything is
gross. So we have to break the data down into net sales to get the true ACOS and TACOS,
rather than working out TACOS based on gross revenue. It can be as simple as that, or as
complicated as extracting every Vine review and how many units we've sold, and saying:
that's the bloated revenue we've received, which isn't true revenue we've actually made.
Those are a couple of examples of what we do.
Lucas Bassa (22:18)
Is profit also part of this picture? Is that something you calculate for your clients, like the net
profit of a channel?
Bhavesh Tailor (22:36)
That's not our core competence. We use other tools like Sellerboard or SellerLegend,
depending on the client. Those are brilliant tools that give you the right data. What we do do,
for example when we're launching a new product, is calculate at what point we hit a good
CM3 level. We understand everything that goes into it, but I don't think we as an agency can
fill the role that Sellerboard and SellerLegend already fill. So what we tend to say to our
clients is: it makes sense to set up a SellerLegend or Sellerboard account for €25 a month,
and it saves us all the hassle and hard work.
Lucas Bassa (23:15)
And do you see any issues there with the data and its trustworthiness? Or is it all API, so it
can't really fail?
Bhavesh Tailor (23:29)
No, you're right about that, in terms of whether you can trust all the data. For example, when
we work with Sellerboard, at account level the spend matches exactly what Amazon says.
But when you start looking at product level, the data sometimes doesn't make as much
sense. I'm not saying their attribution is incorrect, but we prefer to do it ourselves. And there's
the Vine issue you have in Seller Central, which you also get in Sellerboard, where the
revenue is more inflated than the true facts. So generally speaking, we trust the overall ads
data at account level, but not at individual SKU level. When we do ads analysis, we don't use
Sellerboard to do the analysis for us. We look at Amazon Ads data directly. It's the cleanest
way to go about it.
Lucas Bassa (24:34)
Because ultimately you want to know performance at product level, so you go to Amazon's
ad console, where there's no API translation layer in between where things could get blurry.
Is that it?
Bhavesh Tailor (24:47)
There's also the fact that we have a very specific naming structure in our campaigns. At
SNOCKS, it would be "boxer_" and then you'd know it was that specific boxer short in that
campaign name. That's how you can make the connection: that's your parent-level product,
that's the spend we've put out and those are the sales. Of course, there are campaigns that
cover several different products, but then you'd have to look at those individually as well.
Lucas Bassa (25:21)
Okay. But you're saying that if you want the most accurate picture, go to Amazon directly
and get the report straight out of the system, rather than through a third party.
Bhavesh Tailor (25:31)
Yeah. And I think Amazon has been brilliant over the last year or so with how much more
data they offer on the ads side.
Lucas Bassa (25:40)
Yeah, they keep adding new reports and extending the availability of existing ones, which is
huge. Do you use any other tools, or is it then a matter of attaching an AI to your database to
run analysis for you?
Bhavesh Tailor (25:59)
We use Keepa. Keepa is a big one. Back in my time at SNOCKS, we used to follow the BSR
rankings quite heavily. Keepa is a tracking tool where you can track your review count, your
overall rating and your BSR ranking. You can even look at competitors' pricing and how
they've performed over the last year, or since they started selling. What I find super powerful,
and really relevant for our clients, is the buy box tracking. You have that in Sellerboard too,
but I like to see exactly when someone had the buy box and at what price. Especially in Q4,
we've noticed that other resellers are really happy to pull their prices down. If they're running
a deal on another marketplace at 40% off, that's not something we can match on Amazon.
That's why Keepa is so strong for us.
Lucas Bassa (27:08)
That's a good point. We also have a repricer built in to help with that. And taking what you
said earlier about Vine, ideally you also factor in the periods where a competitor undercut
you on your top seller. You might have had a week of very low sales, but it doesn't mean you
need to start tweaking campaigns or changing content. Someone else simply undercut you.
That's definitely a vital metric to keep track of, especially if you're a retailer. So Keepa is
another data source for you. Summarising what you've said: your main point is to get
high-quality data straight from the source where possible, otherwise via reliable third-party
tools, and get it all into your BI or database. But where does the AI come in?
Bhavesh Tailor (28:14)
The AI comes in with the analysis. You can make this as broad or as narrow as you want. One
thing we used to do at SNOCKS was track organic ranking against profit: when your organic
ranking is three or two, what is your profit? That included the spend, and back then it was all
done in Google Sheets and scripts, before we worked with AI. Now building that kind of
analysis is so simple with an AI. That's one example. Another is warning systems. Most of
those run on a Python script, where AI isn't necessarily needed, but you can always add an AI
loop after it to ask why something is going wrong. Does it make sense to match the buy box
price, for example? Yes or no? Things like that.
Lucas Bassa (29:16)
Cool. On the analysis side, is there anything you often see missing when you talk to clients
who've set this up themselves? Because a lot of companies are now tying MCPs and APIs
together inside Claude or another LLM. Do you see steps being skipped? Or is everyone able
to do this at that level already?
Bhavesh Tailor (29:50)
It's a tricky question. We don't use the Amazon MCP at the moment. We use our own
database, because it's more efficient. As far as I know, you can't pull ads data from the MCP.
So bringing all that data together is really inefficient, because pulling an ads report from
Amazon takes a long time, and combining it with the sales data also takes a fair bit of time.
Lucas Bassa (30:24)
And tokens.
Bhavesh Tailor
And tokens as well. You want to be as efficient as possible. So building your own data
warehouse, your own place to get the information from, is a lot more useful. I also think it
empowers people to understand their own account much better, and to build their own
analysis the way they want it, rather than going straight into a Claude MCP. There's definitely
a time and place where a Claude MCP makes sense, 100%. For most people and for smaller
tasks, I'd say go for it. But as soon as you want to get into the nitty-gritty that really moves the
needle, having your own data warehouse makes a lot more sense.
Lucas Bassa (31:13)
My main takeaway from this conversation is that if you invest in data warehousing and data
collection, it makes anything you do with Claude or an LLM easier, and it opens up the
possibility of nitty-gritty analysis: year-on-year comparisons, or combining reports you can't
combine by default when you get them from the marketplace. So it's basically a foundational
part of marketplace selling now. Right?
Bhavesh Tailor (31:58)
Yeah. These days, especially on Amazon, it's a pure data game. If you've got the data, you've
got the keys to, in theory, win, if the product is right.
Lucas Bassa (32:11)
That's still an important detail. You can do a lot with data and improvements, but if your
product just isn't it, it's hard, and your data will point that out too. So, to round up with the
golden nuggets of this episode: what are the top three things people should consider
around data when selling on marketplaces?
Bhavesh Tailor (32:51)
The easiest one: when you're working out your ACOS and TACOS, base them on net revenue,
not gross revenue. Number two, which is quite obvious: remove the Vine revenue you see in
your business reports. It's never true revenue that you've actually generated. Number three:
always allow enough attribution time. For ads, 14 days as a rule of thumb. For business
reports, two or three days outside a deal phase. For a big Black Friday deal phase, I'd say
seven to ten days, especially with this upcoming Black Friday, which is so long. I think it'll
take a fair bit of time for the data to attribute properly. And I'm going to add a fourth one, if
that's okay: if you can, store your SQP data. Even if you're not using it now, it doesn't matter.
You only have the last 64 weeks, and two years down the line that data is going to be really
helpful for seeing how the market is truly developing.
Lucas Bassa (34:08)
Those are nice and practical. On your fourth one, I think it's just good practice to collect the
right data and store it, so you can build your own market context. I've seen tools that do it for
you, but they're based on scraping and estimations. Helium 10 is a huge tool, but it's still
calculations they make, basically guesstimates. Because it's the one tool that does it at such
scale, it's easy to take it at face value. Building your own context will help you further down
the road to make proper analyses based on your actual truths, and to build on data that's
genuinely trustworthy. Am I summarising that correctly?
Bhavesh Tailor (35:02)
You've hit the nail on the head, yeah.
Lucas Bassa (35:04)
Nice. I think that about rounds it up, Bhavesh. Interesting conversation. A few closing
questions: what's your favourite marketplace to work on?
Bhavesh Tailor (35:21)
There's a definite bias towards Amazon, just because there's so much data.
Lucas Bassa (35:25)
Yeah, you can do a lot more, that's what I often hear. And what's your LLM of choice at the
moment?
Bhavesh Tailor (35:33)
I'd say Claude at the moment.
Lucas Bassa (35:36)
All right. Thank you so much for this conversation. Anything you want to add in the final
stretch?
Bhavesh Tailor (35:44)
No, I think that's all.
Lucas Bassa (35:47)
All right, then have a very good Black Friday and a good Q4. Hopefully this episode reaches
people in time for them to start data warehousing. This is also a big topic for us at Channable,
so if anyone listening has feedback or input on this, I'm always keen to get in touch with
people interested in marketplaces, and especially in data. And with that, thank you for
listening.