| AWRA OpsHub

 Search
Intermediate Certificate on pass

AWRA AI: Reading the Analysis

How AWRA turns computed signals into plain-language AI analysis — across insight dashboards, RFQ quote analysis, vendor scorecards, and demand forecasts — with a human always in control.

5 lessons 35 min 5-question assessment 75% to pass
Take the assessment

What you’ll learn

  • Explain the difference between advisory AI and autonomous AI
  • Read the AI summary cards on the insight dashboards
  • Interpret RFQ quote analysis and why editing a quote re-runs it
  • Use vendor scorecard and forecast narratives to decide, not just read charts

Course content

5 lessons · 35 min of reading
01
Lesson 1 of 5 Reading 8 min

Advisory AI: a head start, not a decision

AWRA's AI is built on one principle that you will see repeated on every surface: it advises, it does not decide. The pattern is always the same — a reliable, deterministic computation produces the numbers (stock risk, spend concentration, a weighted quote ranking, a vendor scorecard), and then a language model writes a short, plain-language reading of those numbers on top. The AI never silently changes data, approves a supplier, awards an RFQ, or raises a purchase order on its own. The action always belongs to a person.

That separation is what makes the analysis trustworthy rather than risky. A recommendation you can read, question, and overrule accelerates a decision; an automated action you cannot see erodes control and, eventually, trust. So AWRA deliberately keeps the AI as an explanation layer that sits beside the real figures — you can always check the recommendation against the underlying data, and nothing the AI says is hidden, irreversible, or executed for you.

It is also engineered to fail safely. The narratives are generated through a failover chain of providers (Groq, then OpenRouter, then Gemini); if every provider is unavailable or none is configured, the card simply does not appear — the dashboard, the ranking, the scorecard, and all the figures still work exactly as before. The same is true if a single request errors: the surrounding page never breaks because the AI text could not be produced. AI in AWRA is additive, never load-bearing.

Key takeaways

  • AWRA AI is advisory: deterministic computation produces the numbers, the AI explains them, a human acts.
  • Recommendations sit beside the real figures — never hidden, irreversible, or auto-executed.
  • Narratives use a provider failover chain (Groq → OpenRouter → Gemini).
  • If no provider is available or a request fails, the card hides and the page/data still works (fail-safe).
02
Lesson 2 of 5 Reading 9 min

AI summary cards on the insight dashboards

The AWRA AI section has insight dashboards for inventory, procurement, sales, and accounting, plus a cross-module overview. Each dashboard already computes its signals — low stock and dead stock, overdue invoices, vendor concentration, cash-flow pressure — and renders them as metrics and charts. The AI summary card sits at the top of each one and converts those signals into a few sentences a manager can absorb at a glance: what the data is actually saying and the single most useful next focus.

The value is speed of comprehension under pressure. A manager opening the procurement dashboard should not have to interpret a grid of numbers cold; the card gives them an immediate read — for example, "spend is concentrated in two vendors and one is trending late; consider broadening sourcing" — and they drill into the supporting figures from there. The overview card does this across all four modules at once and names the area that most needs attention, which is exactly the question a busy owner is trying to answer when they log in.

Performance and cost are handled deliberately. Each summary is generated once and cached per business (for several hours), so opening a dashboard is instant and you are not paying for a fresh model call on every page view. When the underlying data has moved and you want the latest read, the "Refresh Insights" action clears those caches and regenerates the narratives — so the cards stay cheap and fast in normal use, and current when you explicitly ask.

Key takeaways

  • Every AWRA AI dashboard — inventory, procurement, sales, accounting, and the overview — carries an AI summary card.
  • The card states what the data means and the single most useful next focus; the overview spans all modules.
  • Summaries are cached per business for hours so pages stay instant and AI cost stays controlled.
  • The "Refresh Insights" action clears the caches and regenerates the narratives on demand.
03
Lesson 3 of 5 Practice 10 min

RFQ quote analysis in plain language

When vendors respond to an RFQ, AWRA first ranks the quotes deterministically on a weighted score of price, distance, and delivery time, and marks a best quote. On top of that numeric ranking — never instead of it — the AI writes a plain-language summary: which vendor is recommended and, crucially, why, including the trade-offs. A useful summary reads like "Acme is lowest at KES X and 2 days faster; BuildCo is cheaper per unit but 40km farther, so it only wins if delivery distance is not weighted." Vendor names in the summary are clickable and link straight to each vendor's quote.

The analysis is consistent about being advisory. The recommendation is a starting point that helps the buyer defend a choice; the buyer still approves the award, and a human approval is what actually closes the RFQ to other vendors — the AI ranking alone never does. This is the same principle as the rest of the platform, applied to sourcing: explain clearly, decide deliberately.

Two behaviours keep it trustworthy in practice. First, it runs on both the automatic on-submit analysis and the manual "Analyze" action, so the card appears however the buyer triggers the review — even when only one quote has arrived. Second, editing a submitted quote automatically re-runs the analysis: the ranking and the narrative both update, so the recommendation never describes figures that have since changed. The result closes a real gap — a bare "best quote" badge tells you which row won, but not why, and not what you give up by picking it.

Key takeaways

  • A deterministic weighted score (price, distance, delivery) ranks quotes; the AI explains the why and the trade-offs.
  • Vendor names in the summary link to their quotes; the recommendation is advisory and the buyer approves the award.
  • Only a human award closes the RFQ to other vendors — the AI ranking never does.
  • It runs on both automatic and manual analysis, and editing a quote re-runs it so it is never stale.
04
Lesson 4 of 5 Reading 8 min

Vendor scorecards and demand forecasts

Strategic Sourcing builds a reliability scorecard for each vendor from real history — total spend, on-time delivery rate, average lead time and its variance, quality and dispute rates, late-delivery rate, and price drift. The AI assessment on the scorecard reads those metrics and states a clear posture: rely on, watch, or review this vendor, with the main reason in one or two sentences. That is far more usable to a buyer mid-decision than a wall of percentages, and because a prequalified supplier carries its provenance through to the vendor record, the scorecard shows where the relationship came from.

Demand forecasting follows the identical pattern. The per-item forecast and the forecast overview compute risk levels, days of cover, and reorder pressure; the AI narrative summarises the landscape — how many items are critical or high-risk, how many need a smart reorder now, which item is closest to stockout — and names the most important action. On a single item it explains that item's outlook; across the catalogue it tells you where to spend the next hour.

The throughline across every surface is intentional and worth internalising: compute reliably, explain plainly, and always leave the decision to a person. The score, the forecast, and the ranking are the facts; the AI narrative is the read; you are the decision. Once you can read one AI card with that lens, every other card in AWRA reads the same way.

Key takeaways

  • Vendor scorecards summarise spend, on-time rate, lead time, quality, disputes, and price drift into a rely/watch/review read.
  • Prequalification provenance carries through to the vendor scorecard.
  • Forecast narratives (per-item and overview) summarise risk and name the most important reorder action.
  • Same lens everywhere: the numbers are the facts, the narrative is the read, you are the decision.
05
Lesson 5 of 5 Practice 7 min

Limits, caching, and using AI responsibly

Reading AWRA's AI well means knowing what it does not do. The providers behind the narratives are text models, so they reason over the structured numbers and fields they are given — they do not open and read the contents of an uploaded PDF, they do not browse the web, and they do not have any data you have not surfaced to them. A supplier suitability score, for instance, reflects the typed application and which documents were supplied, not what is printed inside a certificate. Keeping that boundary in mind stops you from over-trusting a confident-sounding sentence.

Because output is cached, the text you see may be a few hours old even if a figure just changed; that is the deliberate trade for speed and cost. When a decision hinges on the very latest data, use the refresh action (or, for an RFQ, simply re-running or editing triggers a fresh analysis) so the narrative is regenerated from current numbers. Treat a cached card as a fast briefing, not a live wire — and refresh when freshness matters.

Finally, AI text can occasionally be imprecise or, rarely, wrong, which is exactly why every surface keeps the underlying figures next to it and keeps the action with you. The responsible workflow is the same everywhere: read the narrative to orient quickly, glance at the numbers it is summarising to sanity-check, then make the call yourself. Used that way, the AI consistently saves time without ever putting you in a position you cannot explain or reverse.

Key takeaways

  • The models are text-based: they reason over structured data/fields, not PDF contents or the open web.
  • Cached narratives can lag a just-changed figure — refresh (or re-run an RFQ analysis) when freshness matters.
  • AI text can be imprecise; the real figures sit beside it and the action stays with you.
  • Responsible use: read to orient, sanity-check against the numbers, then decide yourself.

Finished the material?

Take the 5-question assessment and earn your certificate — 75% to pass.

Take the assessment

More in this path

Reports & BI Studio Basics

Open course

KPIs that Matter

Open course

Custom Report Builder Deep Dive

Open course

Help Center

Need a quick answer while you read?

Run inventory, procurement, assets, sales, and field work with approved AWRA guidance for setup, migration, integrations, security, pricing, and support.

Search all approved AWRA public help articles.

Open Help Center