Ask questions about your Nexoya data in the AI assistant your team already uses. Every question on this page comes with the answer it gives back, from the two-minute check before standup to the review at the end of the month.

Download the prompt pack
See the examples

MCP can read your data, it cannot change anything. You see what your Nexoya account already lets you see, and you still review and approve optimizations in the Nexoya app.

How it works

Three steps, then it is just conversation

MCP is a standard way to let an AI assistant read data from another tool. Nexoya has a connector for it, so your assistant can see your portfolios without you exporting anything.

1

Connect Nexoya

Add the Nexoya MCP connector to Claude, ChatGPT or Microsoft Copilot. You log in with your Nexoya account, so you see exactly what you can already see in Nexoya.

2

Ask your question

Copy a prompt from below, or write your own. The assistant reads your live portfolios, campaigns, optimizations and attribution.

3

Get an answer you can use

The comparison is already done and the list is already sorted. You get the finding in one sentence you can forward.

Nexoya decides how the budget is split across your channels and predicts what each split will deliver. MCP changes none of that. It reads what Nexoya already worked out and puts it in front of you where you are already working. See the connection guide for the setup steps, or the full list of available tools.

What to put in square brackets

Nexoya knows what you spent, what you got back, and what the optimization suggested. It does not know what you promised your business. So anything that lives in your plan and not in Nexoya, you type into the prompt inside square brackets. Every prompt below shows you where.

[portfolio]Which portfolio you mean, when you have more than one.
[your monthly plan]The spend and the conversions you planned for the month.

How to keep the answers honest

An assistant reads your Nexoya data, but it also knows a lot about marketing in general, and it would rather answer than admit a gap. Four habits keep the two apart. Three of them are already written into every prompt on this page. The second one is the line you type yourself, once at the start of a chat.

1
Ask which dates it usedThe most common mistake is the right analysis on the wrong window, and you cannot see it in the answer. Ask “which portfolio and which dates did you use”, every time.
2
Say “only Nexoya data”, once per chatType “use only the Nexoya data and tell me if something is missing, do not fill the gap from your own knowledge” at the start. You only need it once, and it stops most guessing.
3
Ask for the figures behind the findingIf it says cost per conversion improved, ask for the spend and the conversions next to it. Some Nexoya figures are per day and some are totals for the period, so a number can look right and be wrong by a whole week.
4
Let it say it does not knowA campaign that started this year has nothing to compare with last year. A good answer says so. If a number is going into a deck, ask the same question a second way before you use it.

The prompts

What to ask, grouped by when you ask it

Pick a moment in your week and the question is already written. No client account and no client data appear in the examples below.

Ask one, then just keep asking follow-up questions. The downloadable pack has more of them, plus every longer prompt on this page.

See what you have

Show me all the portfolios I have in Nexoya.

Your portfolios, and the goal each one works towards.

Look at last month

Give me spend, conversions and cost per conversion for [portfolio] last month, broken down by channel.

One table, instead of an export from every platform.

Read the latest optimization

What does the latest optimization propose for [portfolio]? If none is waiting for review, show me the last one we applied instead.

What it suggests, and what it expects to happen.

Turn it into a summary

Summarise our performance over the last month and highlight the three most important changes.

A short text you can forward as it is.

Morning wrap

Every morning you open four platforms to answer one question: did anything break last night? Compared with the previous day the answer is always yes, because Monday never looks like Sunday. The comparison has to be with the same weekday.

The prompt

Every morning, compare yesterday with the same weekday over the last four weeks, not with the previous day, and tell me whether this portfolio had a normal day. Walk down the funnel steps first, so I can see whether anything is off in spend, in traffic, or only at the conversion step. Then do the same campaign by campaign, so I can see whether one campaign caused it or they all moved together. If a step is off, say whether it looks like a real drop or a tracking break.
You get: Whether yesterday was normal, which funnel step moved, and whether one campaign caused it or all of them did.
The answer

What it looked at: Sunday 16 August, against the four Sundays before it

Two things happened, not one. sessions and add to cart are within 6% of a normal Sunday. But conversions are 46% down, so it broke at the cart-to-purchase step: 7.7% of carts converted, against 13.4% normally. Separately, spend is 22% below its Sunday average, so something also stopped serving. The campaign table below says which, and says the conversion drop is not any single campaign.
Funnel stepYesterday Same weekday, previous 4Change
Spend€1,722 €2,199 -22%
Impressions280,417 549,411 -49%
Sessions3,990 4,213 -5%
Add to cart2,305 2,439 -5%
Conversions178 328 -46%
Cost per conversion€10 €7 +44%
Spend vs its own same-weekday averageConversions
Google Demand Gen-94%-80%Outbrain-4%-33%Google app-4%-42%Meta retargeting-2%-36%Google search generic-2%-42%RTB House-2%-47%Google PMax-1%-45%Google search brand+1%-43%0%, a normal Sunday for that campaign
CampaignSpendvs normal Conversionsvs normalRead
Google Demand Gen €30 -94% 6 -80% Stopped serving
Outbrain €85 -4% 8 -33% down with the rest
Google app €537 -4% 56 -42% down with the rest
Meta retargeting €137 -2% 9 -36% down with the rest
Google search generic €226 -2% 30 -42% down with the rest
RTB House €128 -2% 18 -47% down with the rest
Google PMax €296 -1% 12 -45% down with the rest
Google search brand €283 +1% 39 -43% down with the rest
All 8 campaigns €1,722 -22% 178 -46%

This is the part that tells you who to call. Seven campaigns lost between 33% and 47% of their conversions while spending normally. A drop that lands on every campaign at once is not a campaign problem, it is downstream of them, which matches the cart step in the funnel above. Google Demand Gen is the only different one: it spent 94% less, so it stopped serving. That one is a campaign problem, and it is a separate ticket.

What changed last week, and does it matter?

A campaign got more expensive. That could be a real problem, or a smaller budget, or just the time of year. One comparison cannot tell those apart. Two can.

The prompt

Compare last week with the week before for every campaign, then add the same week last year. Tell me which campaigns cost more per conversion on both comparisons, because those are the real problems. Where the two disagree, say which one to trust and why. Leave the year comparison out for any campaign that did not exist last year. Name those campaigns instead of showing them as a drop to zero.
You get: The campaigns with a real problem, kept apart from the ones that just got less money.
The answer

What it looked at: The week of 10 to 16 August, against the week before and the same week in 2025

The week looks good and one campaign is still a problem. We spent 31% less than the week before for 4% fewer conversions, so cost per conversion came down 29%, from €75.11 to €53.67. Against the same week last year it is +4%, which is normal. Meta retargeting is the only campaign more expensive on both comparisons: +31% against last week and +47% against last year. That one is not the budget.
Cost per conversion vs the week beforevs the same week last year
Meta retargeting+31%+47%Google search brand-33%+17%Outbrain-36%+19%Google app-22%+2%Google search generic-37%-11%Google PMax-44%-24%RTB House-43%-28%Google Demand Gen-11%0%. A bar to the right means it got more expensive.
CampaignSpendConv. Cost per conv.vs week before vs last yearRead
Meta retargeting €2.1k21 €102 +31% +47% Real problem
Google search brand €4.9k110 €44.38 -33% +17% Seasonal
Outbrain €2794 €69.74 -36% +19% Seasonal
Google app €9.9k231 €42.94 -22% +2% Seasonal
Google search generic €4.0k77 €51.75 -37% -11% Better on both
Google PMax €5.5k78 €70.70 -44% -24% Better on both
RTB House €98611 €89.65 -43% -28% Better on both
Google Demand Gen €1.6k14 €114 -11% no data Too new to compare
All 8 campaigns €29k546 €53.67 -29% +4%

Two things worth knowing before you run this. Three campaigns are cheaper than last week and more expensive than last year. That pattern is the season, not a problem. It is also why the second comparison is in the prompt: with only the first one, Google app would have been called a win. And Google Demand Gen started after last summer, so it has no year to compare with. Expect that on your own portfolios too, and expect the assistant to say so rather than show a drop to zero.

Over or under budget?

Budget pacing gets checked at the end of the month, when nothing can be done about it. On a Monday half way through, being a few points ahead on spend is still something you can fix.

The prompt

Here is my plan for this month on this portfolio: [your monthly plan]. Compare the share of the budget we have spent with the share of the month that has passed. Do the same for conversions, so I can see whether we are ahead or behind on each. Then project both to the end of the month at the current rate. If we land off plan, tell me whether we are spending too fast, or each conversion is costing more than the plan assumed.
You get: Whether you are ahead or behind at this point in the month, and which of the two numbers is causing it.

Where the money moved

A new proposal arrives. Approving it without looking feels wrong, and checking it campaign by campaign takes half an hour.

The prompt

For the latest optimization, show me for each campaign what it actually spends per day today, what its current daily budget is, and what the optimization proposes. Sort by the size of the change. Then list the campaigns the optimization could not control, with the reason for each. Finish with one sentence on what this proposal is betting on. If no proposal is waiting for review, use the most recent applied one and tell me that is what you did.
You get: Where the money goes, and why the budget caps and the real spend are not the same number.
The answer

What it looked at: The proposal waiting for review, 17 to 23 August

The size of the budget is not the decision here. Where it sits is. Total daily spend barely moves, from €4,186 to €4,205. But €696 a day changes hands. Google app, Google search brand, Google search generic go up. Outbrain, Google Demand Gen, Meta retargeting, Google PMax come down. three of the four cuts land on a campaign Nexoya has already flagged.
Google app+€363Google search brand+€183Google search generic+€131RTB House+€19Outbrain−€5Google Demand Gen−€138Meta retargeting−€166Google PMax−€368€0 a day, no change
CampaignSpending now, per day ProposedChangeNexoya’s flag
Google app€1,417 €1,780 +26% no flag
Google search brand€697 €880 +26% no flag
Google search generic€569 €700 +23% no flag
RTB House€141 €160 +14% no flag
Outbrain€40 €35 -12% Below the platform minimum
Google Demand Gen€228 €90 -61% no flag
Meta retargeting€306 €140 -54% Limited by impression share
Google PMax€788 €420 -47% Saturated
Total€4,186 €4,205 +0%

Sorted by euros moved, not by percentage. A 47% cut on Google PMax is €368 a day, and a 26% rise on Google app is €363. The biggest percentage and the biggest decision are rarely the same campaign.

Did the last optimization deliver?

A total that lands on the forecast can still hide a campaign that missed badly. So the answer has to be checked twice: once for the portfolio, once campaign by campaign.

The prompt

For the last applied optimization, check whether what Nexoya predicted actually happened. Give me the portfolio total first, predicted against actual conversions and cost per conversion, then the same per campaign. Say which campaigns landed close to the prediction and which did not, and set aside any that delivered nothing at all.
You get: Whether the forecast held overall, and which campaigns carried it or broke it.

Did the promotion actually work?

Every promotion looks good next to the week before it. The real question is whether it beat the normal trend, and whether it just borrowed sales from the week after.

The prompt

List every event recorded on my portfolios in the last 90 days. For each one, compare the week the promotion started with the week before it. Then check the week after it ends for a dip, which would mean we pulled demand forward. Score both windows together against the extra money we spent. Skip any promotion that has another one running next to it, because there is no clean week to compare with, and say which ones you skipped.
You get: Which promotions to run again, with borrowed sales kept apart from real ones.
The answer

What it looked at: Events in the last 90 days

Only one of the four still looks good once you count the week after. Free shipping week was the loudest, +24% conversions during the promotion, but the following week came in 23% below the week before it. Take both weeks together and it delivered 8% less than the money behind it bought. Bundle week is the one to repeat: +7% across the two weeks on +2% more spend. Referral push and Back to school moved money, not sales.
Bundle week+7%Referral push+1%Back to school-1%Free shipping week-8%0%, the two weeks together were no better than normal
PromotionExtra spendConv., that week Week afterBoth weeks togetherVerdict
Bundle week · 27 Jul to 2 Aug +2% +14% -1% +7% Real uplift
Referral push · 20 to 26 Jul +6% +6% -1% +1% Budget only
Back to school · 3 to 9 Aug +14% +7% +1% -1% Budget only
Free shipping week · 6 to 12 Jul +19% +24% -23% -8% Pulled forward

Free shipping week, week by week. €26,000 and 340 conversions in the seven days before, €30,900 and 420 during, €25,500 and 262 in the seven days after. Cost per conversion during the week looked good, at €73.57 against €76.47. It is the third window that turns this promotion from a success into a reshuffle, and it is the one most reports never open.

Is the optimization getting better?

Every cycle gets judged on its own, so nobody ever sees the line. Over a quarter the useful question is not whether last week landed. It is whether the whole thing is improving.

The prompt

Look at the last several applied optimizations on this portfolio, oldest first. For each one show what it predicted, what actually happened, and the cost per conversion it ended on. Then tell me whether the forecast is getting more accurate and whether the cost per conversion is coming down. Say so if there is no trend, rather than reading one into the noise.
You get: The line, not the last point. Enough to answer “is this working” in a review.

Where attribution and the platforms disagree

Every platform counts the conversions it thinks it earned. Add those numbers up and you get more sales than the business actually made, because the same one gets claimed twice.

The prompt

For last month, compare the conversions the platforms report with the conversions Nexoya attributes, per attribution rule. Tell me which rules get too much credit and which get too little. Then sum up in two lines how our real channel mix differs from the reported one, written so a finance stakeholder can follow it.
You get: Which channels the platforms over-claim, in a form that survives a finance review.
The answer

What it looked at: July, the whole month

The platforms report 1,879 conversions. 1,004 of those are the same sale counted twice. Nexoya attributes 875 across 12 attribution rules, -53% against what the platforms claim between them. The gap is not spread evenly: Google app reports 856 and is credited 125, while Meta retargeting reports 14 and is credited 28.
Reported by the platformAttributed by Nexoya
Google search generic860531Google app856125Google PMax100108RTB House3832Meta retargeting1428
Attribution ruleReportedAttributed DifferenceRead
Google search generic860 531 -38% Over-claimed
Google app856 125 -85% Over-claimed
Google PMax100 108 +8% Under-credited
RTB House38 32 -16% Over-claimed
Meta retargeting14 28 +100% Under-credited
The other seven rules · small or no volume 1151 too small to compare
All 12 rules1,879 875 -53%

Nothing here says a platform is lying. It says that when two channels both touch the same order, both count it, and only one of them earned it. That is the number a finance stakeholder is going to ask about, so it is better to bring it than to be handed it.

Is the budget going where attribution says it should?

Attribution is only worth having if it changes what you spend. So put the proposed budget next to the cost per attributed conversion, campaign by campaign. Then you can see whether the money is walking towards the cheap end.

The prompt

For the latest optimization, show me each campaign’s cost per attributed conversion and the budget change the optimization proposes, side by side, sorted from cheapest to most expensive. Tell me whether the money is moving towards the cheaper campaigns overall, and give me the portfolio’s cost per attributed conversion before and after. Then list every campaign that moves against that ranking, with Nexoya’s reason where there is one. If no proposal is waiting for review, use the most recent applied one.
You get: Whether attribution is actually steering the budget, and the campaigns where it is not.

Take the prompts with you

This pack is made for the whole team, not just for you. The morning check, the Monday wrap and the quarterly call, each with the question already written.

Download the prompt pack (PDF)
Need help connecting?

Was this post helpful?