Adobe's announcement of agentic AI capabilities in Marketo Engage, including the new Import Leads agent, validates a thesis we've been operating on for a while: list operations are an AI-native problem, not a rules-based one. Seeing Adobe ship in this direction is genuinely good news for the category. The market is catching up to where the work is actually going.
Here's a breakdown of what Adobe's Import Leads agent does, what we've learned building across the full list operations workflow, and what that looks like in allGood today.
What Adobe shipped in the Import Leads agent
The Import Leads agent lives inside Marketo Engage's new Build with AI experience. You can upload a CSV of trade show, event, or content syndication leads directly into Marketo Engage. The agent does a fuzzy match on column headers to map them to Marketo Engage fields — a real quality-of-life improvement, given that almost every trade show provides different names for common fields. Adobe also gives you an easy way to verify the mapping before committing, which is a thoughtful piece of UX.
Once columns are mapped, you can invoke business rules to clean data, like normalizing State columns to two-letter codes. This is a great application of agentic AI: reasoning models are meaningfully better than traditional rules-based systems at data cleaning, and it's good to see Adobe leaning into that.
The rules are extensible, which means sophisticated work like persona mapping, job level and role classification, and basic firmographic scoring is possible. Adobe has also announced support for multiple LLM providers as part of its broader agentic AI strategy for Marketo Engage.
What we've learned building across the full list operations workflow
Adobe's Import Leads agent covers the upload and mapping layer. The full list operations workflow extends beyond that. Here's what we've learned building across the whole surface area.
Multi-tab spreadsheets are their own problem
Field marketers routinely submit multi-tab files, and consolidating them isn't just a formatting exercise. You have to dedupe across tabs and decide how to prioritize conflicting campaign member statuses, notes, and field data before the file ever reaches Marketo Engage. We learned the hard way that hardcoding a status priority list doesn't hold up across customers. So allGood pulls Marketo Engage's actual ordered campaign member status list dynamically via MCP and consolidates against it. Every customer's prioritization logic just works, because it's their own.
PII control needs to be granular, not binary
Early on, we faced a real trade-off: LLM-based matching is meaningfully better than fuzzy matching, but customers (rightly) didn't want PII flowing to LLM providers. The lesson was that the right answer isn't "send everything" or "send nothing." It's field-level control. Some fields — column headers, non-sensitive firmographics — are fine to send. Others — email addresses, phone numbers, anything that qualifies as PII under a customer's compliance regime — should never leave their environment. So allGood lets customers make that call field by field, across every cleaning, matching, and enrichment step.
No single LLM is best at everything
We kept running into cases where Claude handled persona classification beautifully but stumbled elsewhere, while basic data cleaning worked better via OpenAI. Picking one model and forcing it to do everything was leaving accuracy on the table. So allGood routes each cleaning rule to the model that handles it best, and we maintain the provider abstractions so customers don't have to.
Models drift, silently
This one stung. LLM providers ship new models, and worse, they silently update existing ones. Prompts that worked perfectly on Tuesday could break on Wednesday. The only way to catch it before customers did was to build a test harness with representative data and expected outputs, running on a continuous cadence. So allGood ships with continuous regression testing built in.
LLM outages are real
Both Anthropic and OpenAI have had them. A list upload can appear successful but actually have timed out mid-processing, and without monitoring, neither you nor the marketer who uploaded the file knows. So allGood ships with monitoring infrastructure built in, because anything touching the lead-to-revenue pipeline can't afford silent failures.
What full list operations look like in allGood today
- Multi-tab spreadsheet ingestion with dedupe and conflict resolution before records reach Marketo Engage
- Marketo-native status prioritization via MCP, using each customer's actual ordered campaign member status list
- Field-level PII gating across every cleaning, matching, and enrichment step
- LLM-native column matching with full data control
- Per-rule model routing across Claude and OpenAI
- Continuous regression testing against representative data
- Built-in monitoring for LLM provider issues and silent failures
The choice for Marketo Engage customers
Adobe's Import Leads agent is a strong starting point for the upload and mapping layer, especially if you have engineering resources to build the surrounding workflow.
If you want the full list operations workflow working out of the box, with the lessons above already baked in, that's what allGood is for.
Either way, the direction the market is moving is the right one. We're glad to see Adobe pushing it forward.
