“Our collaboration has been nothing short of a revolution. Dema's solution is now the beating heart of our operation and we'll keep evolving and growing together.”
Elin Glommen
CEO
A Triple Whale alternative for the whole commercial team
Triple Whale is built for marketing. Dema is built for the people who decide what to buy, what to stock and what it all earns, with one semantic layer underneath, so agents can answer questions that cross all three.
215 metrics110 dimensions, one definition each
3 mindata delay, including profit
1M+ actionsrun by agents every month
Dema or Triple Whale: which should you pick?
Choose Triple Whale if you want attribution models that steer your Meta advertising, out of the box. Their click and view-based attribution is good and it works on day one. If that is the job, they do it well and Dema will ask more of you. Choose Dema if you are buying a system for the whole commercial organisation: profitability down to GP3 per product per market per campaign, inventory and sell-through forecasting, cookieless measurement through MMM and incrementality testing, and agents that automate the work across all of it.
Where Triple Whale is the better choice
Three things we would rather you heard from us than found out later.
Meta attribution out of the box
Their click and view-based attribution model is strong and immediate. Dema does not ship an equivalent proprietary view-through model. Our path to the same confidence is incrementality testing or MMM, which is more rigorous and slower to stand up. If you want a number for Meta on your first afternoon, they will give you one.
AI search visibility
They track how often your brand appears in AI answers. Their AI Visibility table records LLM prompt executions across engines like ChatGPT and Gemini, with a visibility percentage, the sources each answer cited, and whether competitors were mentioned instead of you. Dema does not do this at all. If monitoring your presence in AI search matters to you, that is a real gap on our side.
Defaults, not a blank canvas
Triple Whale is more prescriptive: you land in prepared views and get value quickly. Dema gives you more freedom to compose reports, dashboards, apps and agents, which is the point, but freedom is a cost on day one if you want someone else to have decided for you.
Dema vs Triple Whale, feature by feature
Rows where Triple Whale wins are marked as such. Everything about their product below is taken from their own public documentation.
Who it is built for
DemaThe full commercial team: buying, merchandising, finance, paid media, logistics.
Triple WhaleMarketing and paid media.
Data model you work against
DemaA semantic layer: 215 metrics and 110 dimensions, each with one definition. Combine almost any metric with any dimension.
Triple Whale47 documented tables and 204 metric names covering 83 distinct underlying fields. You own the joins.
Their docs expose "spend" under 15 different metric names and "conversion value" under 17.
Custom analysis
DemaCompose it in one report and one dashboard surface. No SQL.
Triple WhaleSQL against their tables; their docs include a 16-page SQL example library.
Profit depth
DemaDown to GP3 per product, per market, per campaign, with shipping, transaction, pick-and-pack and toll costs plus actual and estimated return rates.
Triple WhaleCost of goods, handling fees and taxes are documented metrics.
Inventory and sell-through
DemaSell-through forecasting per size, and inventory as of any past date for as far back as data was fetched.
Triple WhaleNeither appears among their 204 published metrics.
Attribution
DemaAd platform attribution and multi-touch attribution are inputs, not answers. Causal factor attribution weights them against experimental evidence per channel, funnel stage and market.
Triple WhaleTheir own proprietary multi-touch model, including view-through. This is their core product and it is good.
If a proprietary Meta number is what you want, they get you there faster and with less work.
Measurement beyond attribution
DemaMarketing mix modeling, run on profit and on LTV, plus geo-based incrementality testing whose results calibrate the model and set the attribution weights.
Triple WhaleGeo-lift experiments are documented in beta. No marketing mix modeling appears in their documentation.
This is the cookieless half, and it is the half that keeps working as tracking coverage falls.
Agents
DemaYou build them. Each gets its own tools, integration access and persistent memory, runs on a schedule, and can have an app attached that people actually work in. More than a million agent actions a month across customers.
Triple WhaleMoby 2 ships vendor-built Specialists around specific ecommerce KPIs, in Copilot or Autopilot mode.
Both execute real changes. The difference is whether the agent is a product you buy or a system you build.
What you can build
DemaWorking systems on the semantic layer (a replenishment engine, a search-term optimiser, a sell-through tracker), each able to carry its own app surface for the team that uses it.
Triple WhaleSpecialists cover the KPIs they were built for. Outside those, you get analysis and reporting.
API
DemaGraphQL over the semantic layer, on every plan. All 215 metrics and 110 dimensions, each with its definition already fixed.
Triple WhaleData-In and Data-Out APIs with SQL access to their tables, so the definitions are yours to get right.
Northbeam, for comparison, gates API access to a plan starting at $2,500/month.
AI search visibility
DemaNot offered.
Triple WhaleTracks how often your brand is mentioned in AI answers across engines like ChatGPT and Gemini, with cited sources and competitor mentions.
Omnichannel and offline
DemaOnline and offline in one model: retail, wholesale and separate warehouses, each able to carry its own cost structure. Shipping, toll, pick-and-pack, transaction and return costs are rules-based, so an in-store or wholesale order is not costed as though it shipped from your e-commerce warehouse.
Triple WhaleMarketplace data is well covered, with its own Amazon metric set. Other offline orders are pushed in as a Custom Sales Platform, against one shared set of cost inputs: cost of goods, handling fees, shipping and payment gateway costs.
Their enrich-orders and enrich-products endpoints are documented as unsupported for Custom Sales Platforms, so offline orders pushed in that way have to be resent in full to be updated.
Where AI inference runs
DemaChoose EU or US. Model inference and data pipelines run on European or US infrastructure with residency control, and the LLM subprocessors are optional.
Triple WhaleTheir published DPA and privacy notice describe hosting in Google's US data centres, with EU-to-US transfers covered by Standard Contractual Clauses and the EU-US Data Privacy Framework. OpenAI is a named subprocessor for the AI assistant.
No EU inference region appears in their published documentation.
Data freshness
DemaAbout a 3-minute delay, profit included.
Triple WhaleNear real-time dashboards.
Help building it
DemaOur customer success team builds apps and agents with you.
Triple WhaleSupport scales with plan tier.
| Dema | Triple Whale | |
|---|---|---|
| Who it is built for | The full commercial team: buying, merchandising, finance, paid media, logistics. | Marketing and paid media. |
| Data model you work againstTheir docs expose "spend" under 15 different metric names and "conversion value" under 17. | A semantic layer: 215 metrics and 110 dimensions, each with one definition. Combine almost any metric with any dimension. | 47 documented tables and 204 metric names covering 83 distinct underlying fields. You own the joins. |
| Custom analysis | Compose it in one report and one dashboard surface. No SQL. | SQL against their tables; their docs include a 16-page SQL example library. |
| Profit depth | Down to GP3 per product, per market, per campaign, with shipping, transaction, pick-and-pack and toll costs plus actual and estimated return rates. | Cost of goods, handling fees and taxes are documented metrics. |
| Inventory and sell-through | Sell-through forecasting per size, and inventory as of any past date for as far back as data was fetched. | Neither appears among their 204 published metrics. |
| AttributionIf a proprietary Meta number is what you want, they get you there faster and with less work. | Ad platform attribution and multi-touch attribution are inputs, not answers. Causal factor attribution weights them against experimental evidence per channel, funnel stage and market. | Their own proprietary multi-touch model, including view-through. This is their core product and it is good. |
| Measurement beyond attributionThis is the cookieless half, and it is the half that keeps working as tracking coverage falls. | Marketing mix modeling, run on profit and on LTV, plus geo-based incrementality testing whose results calibrate the model and set the attribution weights. | Geo-lift experiments are documented in beta. No marketing mix modeling appears in their documentation. |
| AgentsBoth execute real changes. The difference is whether the agent is a product you buy or a system you build. | You build them. Each gets its own tools, integration access and persistent memory, runs on a schedule, and can have an app attached that people actually work in. More than a million agent actions a month across customers. | Moby 2 ships vendor-built Specialists around specific ecommerce KPIs, in Copilot or Autopilot mode. |
| What you can build | Working systems on the semantic layer (a replenishment engine, a search-term optimiser, a sell-through tracker), each able to carry its own app surface for the team that uses it. | Specialists cover the KPIs they were built for. Outside those, you get analysis and reporting. |
| APINorthbeam, for comparison, gates API access to a plan starting at $2,500/month. | GraphQL over the semantic layer, on every plan. All 215 metrics and 110 dimensions, each with its definition already fixed. | Data-In and Data-Out APIs with SQL access to their tables, so the definitions are yours to get right. |
| AI search visibility | Not offered. | Tracks how often your brand is mentioned in AI answers across engines like ChatGPT and Gemini, with cited sources and competitor mentions. |
| Omnichannel and offlineTheir enrich-orders and enrich-products endpoints are documented as unsupported for Custom Sales Platforms, so offline orders pushed in that way have to be resent in full to be updated. | Online and offline in one model: retail, wholesale and separate warehouses, each able to carry its own cost structure. Shipping, toll, pick-and-pack, transaction and return costs are rules-based, so an in-store or wholesale order is not costed as though it shipped from your e-commerce warehouse. | Marketplace data is well covered, with its own Amazon metric set. Other offline orders are pushed in as a Custom Sales Platform, against one shared set of cost inputs: cost of goods, handling fees, shipping and payment gateway costs. |
| Where AI inference runsNo EU inference region appears in their published documentation. | Choose EU or US. Model inference and data pipelines run on European or US infrastructure with residency control, and the LLM subprocessors are optional. | Their published DPA and privacy notice describe hosting in Google's US data centres, with EU-to-US transfers covered by Standard Contractual Clauses and the EU-US Data Privacy Framework. OpenAI is a named subprocessor for the AI assistant. |
| Data freshness | About a 3-minute delay, profit included. | Near real-time dashboards. |
| Help building it | Our customer success team builds apps and agents with you. | Support scales with plan tier. |
Triple Whale details verified against developers.triplewhale.com and their published documentation in August 2026. Counts of tables, metric names and underlying fields are our own tally of their public docs. Their product changes, so if you spot something out of date here, tell us and we will correct it.
One question, one query
“Give me every product, down to operational profitability in Google, with expected sell-through, how long each was out of stock last month, the LTV of customers who buy it, and estimated returns coming back.” In Dema that is a single query against the semantic layer. Answering it against 47 raw tables means writing joins and deciding, fifteen times over, which field called something like spend you meant.
Fifteen names for one number
This is the part that is easy to underrate when comparing feature lists, and it decides whether agents are useful or just plausible.
The definition is the hard part
Ask any tool what you spent on Google last week and the number is easy. Ask what a product earned after returns, shipping and the campaign that sold it, and the answer depends entirely on which fields you picked and how you joined them. That is where analytics goes wrong: not in the arithmetic, but in the definitions.
An agent has to choose, unless you have chosen already
Triple Whale's documentation exposes ad spend under fifteen different metric names and conversion value under seventeen, across 47 tables. An agent writing SQL against that has to pick, every time, and it has fifteen ways to get one number wrong. Dema's agents never pick: they ask for a metric by a dimension and the definition is already fixed. It is a smaller problem to get right, which is the point.
Agents you build, not agents you're given
Both products have agents that act. The difference is who decides what they do, and what data there is to act on.
Working systems, not just answers
Dema agents run more than a million actions a month, the overwhelming majority on a schedule rather than in response to a prompt. Many have an app attached: a surface people actually work in, not a chat transcript. Customers use them to build things like replenishment systems, search-term optimisers that write negative keywords back into Google Ads, collection merchandising that respects stock and margin, and budget reallocation across Meta and Google. Those are systems built on the platform, not features we shipped.
Execution you can bound
Agents act through the same MCP integrations the platform uses, and each one is scoped: which tools it holds, which accounts it can touch, and whether a change needs a human first. Some run fully autonomously inside an allowlist, others queue a preview for approval, and most only read and report. That gradient is deliberate. An agent worth trusting with your ad account is one whose limits you set.
Permissions the agent inherits
Each agent gets its own tools, its own integration access and persistent memory scoped to the user. Teams restrict who can see what, and the agent works inside those limits, which is what makes it safe to let one prepare a decision rather than only describe data.
What switching actually involves
Connect your ad platforms and e-commerce platform with OAuth, which takes minutes each. The part that takes longer is cost data (cost of goods, fulfilment rates, expected returns), because that is what makes profit reporting real, and it usually needs someone from finance rather than an API key. Most teams are connected within a day. Historical inventory and profit go back as far as we can fetch data, so you are not starting from an empty chart.
Trusted by category leaders.
Frequently asked questions
For the profitability, inventory and measurement work, yes, and for a wider group of people, because Dema is used by buying, merchandising and finance as well as marketing. For proprietary Meta attribution specifically, Triple Whale does something Dema deliberately does not: they ship their own click and view-based model, while Dema gets to causal answers through incrementality testing and marketing mix modeling. If that model is the reason you use them, be clear-eyed that Dema asks more of you there.
The data model. Triple Whale's public documentation describes 47 tables and 204 metric names covering 83 distinct underlying fields, so "spend" appears under 15 different names and you write the joins. Dema gives you a semantic layer: 215 metrics and 110 dimensions, each defined once, which you can combine almost freely. That is why a question spanning profit, inventory and campaign performance is one query in Dema rather than a SQL exercise.
You do not migrate Triple Whale's output; you reconnect the sources. Dema connects to your ad platforms, e-commerce platform, POS, ERP and warehouse directly and rebuilds the model from them, then backfills as far as each source allows. In practice that means your history comes from the systems of record rather than from a competitor's export, so nothing depends on what they let you take with you.
Yes, both have one. The difference is what it exposes. Theirs gives query access to their raw tables, so every definition and join is yours to get right. Dema's GraphQL API exposes the semantic layer itself (all 215 metrics and 110 dimensions, each already defined) on every plan, so a request names a metric and a dimension and gets a number that matches what the platform shows. For comparison, Northbeam gates API access behind a plan starting at $2,500 a month.
Yes, alongside last-click, linear, ad-platform-reported, causal factor attribution and marketing mix modeling. But MTA is treated as one weighted input rather than the source of truth, because it can only see journeys that were tracked. As consent rates fall, the share it cannot see grows, which is why causal factor attribution weights it against experimental evidence instead of trusting it outright.
They change things, through the same MCP integrations the platform runs on: writing negative keywords into Google Ads, updating collection membership in Shopify, moving budget between Meta and Google. Each agent is scoped to the tools and accounts you give it, and you decide per agent whether a change executes inside an allowlist or queues a preview for approval. In practice most agents only read and report, which is the right default. Triple Whale's Moby 2 also executes changes, with Copilot and Autopilot modes; the difference is that their agents are Specialists built around the KPIs they ship, while a Dema agent is something you assemble on your own data model.
Yes, and it is one of the clearer illustrations of why the data model matters. Replenishment needs inventory history, sell-through forecasts per size, and supplier and warehouse structures with their cost models, all of which sit in the semantic layer, so an agent reads them in one query rather than assembling them from joins. You can give it an app for the buying team to work in and have it write the output back to the systems they already use. The reason this is hard elsewhere is not the agent framework; it is that a marketing data model has no sell-through forecast or warehouse cost structure to reason over in the first place.
That is much of the reason Dema exists. Online, retail and wholesale sit in one model, and separate warehouses can each carry their own cost structure, because the true cost of an order depends on where it was fulfilled from and how it reached the customer. Fulfilment, shipping, toll, pick-and-pack, transaction and return costs are rules-based rather than a single blended figure, so a store sale, a wholesale order and a home delivery each get costed on their own terms and still roll up into one comparable profit number. Triple Whale covers marketplaces well and accepts other offline orders pushed in as a Custom Sales Platform, but against one shared set of cost inputs.
Yes. You choose the region: model inference and data pipelines run on European or US infrastructure, with residency control, and Dema's hosting region is the EU. The LLM providers used for agent features are optional subprocessors rather than a hard dependency, and Dema is GDPR compliant. Triple Whale's published DPA and privacy notice describe customer data hosted in Google's US data centres, with transfers out of the EEA covered by Standard Contractual Clauses and the EU-US Data Privacy Framework, and name OpenAI as a subprocessor for their AI assistant. We could not find an EU inference option in their published documentation. If EU data residency is a procurement requirement, this is worth confirming with both vendors in writing.
If you believe multi-touch attribution answers the important questions and you are comfortable building on cookie-based tracking, Dema is the wrong tool: our entire design assumes the opposite. Likewise if you want to land in prepared dashboards tomorrow with no decisions to make: Triple Whale and Northbeam are more prescriptive out of the box, and that is a real advantage for a team that wants defaults rather than freedom.
Everything stated about Triple Whale here comes from their own public documentation and developer portal, checked in August 2026. The table counts are our tally of their published tables and metrics. Their product moves, so if something here has gone out of date, tell us and we will fix it. A comparison page that quietly rots is worse than no comparison page.
See what one semantic layer does to the questions you can ask.