Tips for Fine-Tuning Report Templates in Indago
The Quality Gap Starts Before You Click Generate
Most analysts who are disappointed with AI-generated reports blame the model — the output feels generic, the tone is off, the structure doesn't fit the audience. Switching models or adjusting the prompt mid-generation doesn't fix it.
The quality of an AI-generated report is determined before you click generate. It is encoded in the template — in the decisions made during setup that most users treat as formalities. If you accept the defaults, fill in a title, skip past the persona field, leave the outline sections unlabeled, then you have handed the AI the maximum amount of ambiguity to resolve on its own. You get exactly what you'd expect: something technically competent and analytically dead on arrival.
Proper template prompt configuration is what’s missing. Three decisions made at the template level determine most of what separates a generic first draft from a genuinely useful one: who the AI is writing as (persona), how the report is structured (outline), and which model handles which section (model selection). Each is a lever that compounds the others. When all three are configured deliberately, the template stops being a blank form and becomes a reusable encoding of your analytical standards — one that generates to the same quality bar every time, regardless of who is running it.
Persona
The persona field is the highest-leverage decision in any Indago template, and the one most commonly treated as an afterthought. Every word the AI writes is filtered through who it thinks is doing the writing, for whom, and why.
What it actually controls is more expansive than the name suggests. It governs vocabulary and register — technical, executive, operational, accessible — and it governs structural instincts, like whether the AI leads with the bottom line or builds to it.
Picture a collection of sources on a regional supply chain disruption. The first persona might open with: 'Three key suppliers now face simultaneous disruption risk, with compounding effects on Q3 delivery timelines.' The second might open with: 'Southeast Asia's supply chains have come under increasing pressure in recent months, driven by geopolitical tension and infrastructure strain.' Neither is wrong in the abstract. One is wrong for a leadership briefing (spoiler: it’s the second one).
A useful persona goes beyond job title. A corporate security analyst reporting to a board needs a different configuration than the same analyst writing a technical brief for the SOC team — even though both personas carry the word 'analyst.' What changes is the audience and the purpose.
Persona is also where you encode institutional voice and standing instructions. If your organization uses specific confidence language — assessed with moderate confidence rather than likely — specify it here. If reports going to a particular client should avoid forward-looking statements for legal reasons, note it. If the reading level should stay accessible to non-specialist stakeholders, say so explicitly. You define the voice once. Every report built from that template runs on it, without depending on any individual analyst to remember the standard.
Outline Structure
Most analysts treat the outline as a table of contents or a list of section titles. However, that leaves most of its value untapped. A well-built outline encodes repeatable formatting logic and audience-specific requirements, so every generation inherits the same standards automatically.
The first structural decision is how to weight the report. Some reports should be top-heavy: the executive summary and key findings carry the analytical payload, and everything that follows supports what was already stated up front. This suits audiences who read selectively — senior leadership, clients, anyone who will act on the bottom line without working through the full document. Other reports benefit from a detail-heavy structure, where context and evidence accumulate before conclusions are drawn — better for technical audiences, legal review, or any situation where the reasoning chain needs to be traceable. The choice should be deliberate and encoded in the template, not left to the AI to infer.
Subsections earn their place when a section covers genuinely distinct material — they create noise when they fragment a coherent argument into bureaucratic pieces. The test: if removing the heading would make the reader work harder to orient themselves, keep it. If it's just a label on a paragraph, cut it.
Indago's outline system also lets you embed formatting instructions directly into each section — instructions that execute automatically on every generation. A few examples:
Language and translation. Specify that a section always closes with a French-language summary if the report routinely goes to French-speaking counterparts.
Embedded tables. Specify that a section always includes a risk matrix — probability, impact, recommended action — so executives can scan instead of reading paragraphs of prose.
Format control per section. An executive summary reads better as tight prose; a findings section with several discrete items reads better as a numbered list. Specify the format per section, or the AI defaults to its own judgment — often reasonable, rarely calibrated to your audience.
Per-section word counts. Without explicit targets, the AI tends toward false balance — similar depth everywhere regardless of importance. Set a target (a key findings section might run 150-200 words, not 400) and the proportions reflect your priorities, not the AI's defaults.
TLDR blocks. A single bolded takeaway line at the bottom of each section — 'Bottom line: regulatory risk in this market is elevated and accelerating' — serves stakeholders who skim, generated automatically from the section content.
Each of these instructions, set once, applies indefinitely. The real compounding effect is the quality floor that rises across every analyst who runs the template, not just the one who built it.
Section-Level Model Selection
Most analysts choose a model once at the start of report generation and apply it uniformly across every section. It feels efficient, but it leaves a meaningful quality lever untouched since not every section of a report requires the same kind of AI reasoning. Treating them as if they do introduces miscalibration throughout the output.
Some sections need a model that can reason through ambiguity and produce structured judgment — an executive summary or recommendations section requires synthesizing source material into a coherent argument, deciding what matters and how to frame it. Other sections need the opposite: a source summary or data table just needs the AI to follow instructions precisely and format cleanly, without editorializing. A reasoning-optimized model's tendency to interpret can actually work against that kind of task.
Consider a competitive intelligence report. The source summary section — who said what, when, with what degree of specificity — is a formatting and extraction task. A faster, more literal model handles it well. But the synthesis section, where the AI needs to identify patterns across sources, characterize competitive positioning, and surface strategic implications, is a reasoning task. The cognitive demand is different. A reasoning-optimized model produces materially different output here: more nuanced, more analytically coherent, better calibrated to the complexity of the task.
Indago supports this at the section level: a template saves one model as the default, but analysts can select a different model when regenerating a specific section that’s optimized for what each section’s content and what each is accomplishing. The synthesis section gets a reasoning-optimized model whereas the source summary gets a model that’s faster and more literal. It’s a deliberate, section-by-section decision made at the point of regeneration, not a blanket setting applied once and forgotten.
Putting it Together
Persona tells the AI who it's writing for. Outline tells it what to build and how. Model selection tells it how to think section-by-section as you regenerate. Configured together, they turn the template from a starting point into an institutional asset — the analytical standards, voice, and structural conventions your team has developed, encoded once and applied automatically on every report that follows. The first time takes real effort, but it’s worth it because every report after that inherits those decisions automatically, and your time goes back to the work that actually needs it: source curation, judgment, refinement.
Put It Into Practice
If you're already using Indago: The fastest way to apply what you just read is to open the template builder and start with the persona field. If it reads like a placeholder — "write professionally" or similar — it is one. Rewrite it to describe a specific role, a specific audience, and the analytical stance the report needs to take. Then work through your section instructions: word count targets, format specifications, model assignments. One template, reviewed deliberately, will produce noticeably better output on the next generation.
If you're evaluating Indago: The configuration logic described here — persona, outline, model selection — is not a workaround or an advanced feature. It is how the platform is designed to work. If your reporting workflow needs this kind of repeatable structure, book a demo and bring a report type you actually produce. That is the fastest way to see whether it fits.