API Limits and Document Generation: What Actually Governs Your Merge Volume
Document automation gets scoped on template count and forgotten on limits. The demo generates one PDF in two seconds, everyone nods, and nine months later a batch job dies at four in the afternoon because the org ran out of API calls generating renewal notices.
Merge volume is governed somewhere. The useful question for an integrator is where — because that answer differs depending on whether the generation engine runs inside the platform or beside it.
Three separate ceilings, often confused
People say “API limits” and mean at least three different things. They fail in different ways and they are raised by different means.
The org’s daily API request allocation. Set by Salesforce according to edition and licensed user count. Every read the generation engine performs against the org draws down this pool, alongside every other integration you have.
The API version the tool is pinned to. Distinct from volume, and quieter when it bites. Conga publishes that Composer runs against a specific Salesforce API version and cannot merge data from objects that require a newer one. If a Salesforce release introduces an object your workflow needs, you wait for the package to catch up.
The vendor’s own product ceilings. Independent of Salesforce entirely. Conga’s product data limits spell theirs out: batch output to cloud storage destinations caps at 200 records for consolidated output and 1,000 for distributed output, and the daily maximum can be raised on request.
Publishing your ceilings is a sign of a mature product, not a weakness. The vendor who cannot tell you where the wall is has not found it yet — which means your client will.
The failure nobody scopes for: delivery
Generation and delivery are separate operations with separate budgets, and the second one is where automations tend to die quietly. Salesforce caps how many emails can be sent through its API in a rolling day. Conga’s documentation on automatic emails notes that once that limit is exceeded the process errors out, and directs customers toward integrating an external delivery service instead.
Read what that recommendation implies. The moment volume matters, the platform-native path routes delivery off the platform. This is not a criticism of Conga — it is sound engineering advice, published honestly. It is a demonstration that the boundary between “inside Salesforce” and “outside Salesforce” is thinner than the marketing on either side suggests, and that the boundary moves under load.
What changes when the engine sits beside the platform
In a form-initiated architecture, the merge does not run against the org at all. Answers arrive from an intake form, the rules select and populate the templates, and the org is touched exactly twice: once to match or create the record, once to attach the finished files.
That does not abolish limits. It relocates them, and the relocation is the point. Your Salesforce API budget is spent on record operations rather than on reading fields for a merge, and the merge’s own throughput is governed by the document engine rather than by whatever else is competing for the org’s daily allocation that afternoon. The tradeoff — platform-neutral pathway versus the Salesforce-specific one — is laid out in webhooks vs. the Salesforce API in DocupletionForms.
We have ceilings too, and we would rather you find them in a sandbox than in production. Architecture and precise current status live in document automation for IT and security teams.
Four questions before you scope
Run these against any document generation vendor, including this one. Answers should come from published documentation, not from a sales call.
1. How many API calls does one document cost? Multiply by the client’s peak day, then add the integrations already in the org. If the vendor cannot answer, the answer is “more than you think.”
2. Which API version is the tool pinned to, and what is the upgrade cadence? This determines which objects you can reach, and it is not something you control.
3. What is the batch ceiling, and can it be raised? Get the number and the process for raising it, in writing, before the renewal notice run.
4. Where does delivery happen? If the volume answer involves an external mail service, you are already running a hybrid architecture. Design it deliberately rather than discovering it.
Designing for the ceiling
Three habits keep volume from becoming an incident.
Filter before the engine, not inside it. An incomplete record that reaches the merge has already cost you the reads. Gate it earlier — see Zapier filters before sending data to DocupletionForms and required fields before triggering a packet.
Fire on one trigger per business event. A single checkout can trip four triggers at once, and four triggers produce four packets. One event, one entry point, filters for the rest — the discipline is worked through in the Stripe payment workflow.
Make regeneration idempotent. Flag a row once its documents exist, so a later edit cannot re-fire the merge. Deterministic output is only useful if it also runs exactly once — the argument for deterministic document automation extends to how many times the same inputs are allowed to produce their identical result.
Chunking a 10,000-document run into ten runs of a thousand is not a workaround. On any architecture, at any vendor, it is what competence looks like.
Related reading: record-initiated vs. form-initiated document generation, generating Salesforce documents without a Salesforce license, and security-aware Salesforce document automation.