The Challenge: Unreliable Short-Term Forecasts

For many finance teams, short-term cash forecasting is still driven by static spreadsheets, manual updates, and rough averages. The result is a blurred view of the next 7–30 days: you see overall trends, but not the daily swings in inflows and outflows that actually determine whether you can pay suppliers, payroll, or taxes without scrambling for funding.

Traditional approaches struggle because they ignore the complexity and velocity of modern cash movements. Weekly spreadsheets can’t keep up with changing payment behaviors, seasonality, ad-hoc collections, and shifting vendor terms. Simple averages don’t capture patterns such as “Fridays are heavy payout days” or “Month-end discounts pull cash in earlier.” And when the underlying ERP, banking, and sales systems are siloed, analysts spend their time reconciling data instead of improving the forecast model.

The business impact is tangible. Unreliable short-term forecasts lead to last-minute funding gaps, increased reliance on expensive credit lines, and suboptimal use of excess cash. Treasury loses the ability to lock in better rates because they don’t have confidence in projected positions. Operational teams get surprised by payment holds. Management loses trust in the numbers, so every major payment decision becomes an escalation and slows the business down. Over time, this erodes both margin and credibility.

The good news: this challenge is real, but it is solvable. With the right data foundation and AI models, you can move from static spreadsheets to rolling, scenario-based cash forecasts that update as new information comes in. At Reruption, we’ve helped organisations replace fragile, manual workflows with AI-first tools in finance and other data-heavy domains. Below, we’ll show how to use Google’s Gemini together with BigQuery and Sheets to stabilise your short-term forecasts and give stakeholders clear, explainable insights instead of surprises.

Need a sparring partner for this challenge?

Let's have a no-obligation chat and brainstorm together.

Innovators at these companies trust us:

Our Assessment

A strategic assessment of the challenge and high-level tips how to tackle it.

From Reruption’s perspective, unreliable short-term forecasts are primarily a data and workflow problem, not just a math problem. We’ve seen in multiple AI implementations that once you connect transaction histories, ERP data, and sales pipelines into a clean structure, models like Gemini can add a powerful layer of pattern detection and explanation. The opportunity is not only to predict cash positions more accurately, but to turn those predictions into actionable guidance for finance teams inside tools they already use, such as BigQuery and Google Sheets.

Think in Systems, Not in One-Off Forecast Files

Most unreliable forecasts are symptoms of a fragmented system: different data sources, inconsistent update cycles, and Excel files living on personal drives. Before throwing AI at the problem, define the end-to-end cash forecasting system you actually need: which sources feed it, how often they refresh, how forecasts are generated, and how stakeholders consume them.

Strategically, this means moving from “the spreadsheet Serge built three years ago” to a repeatable forecasting pipeline. Gemini then becomes a component in that pipeline: it can ingest aggregated data from BigQuery, generate rolling forecasts, and push results plus narrative explanations into Sheets or dashboards. When you design the system first, you avoid building clever AI that sits on top of unreliable inputs.

Prepare Your Data Foundations Before Scaling AI

AI models like Gemini need structured, reasonably clean data to produce trustworthy short-term forecasts. That doesn’t mean you must complete a multi-year data lake project, but you do need to align on basic data discipline: consistent transaction timestamps, mapped customer and vendor IDs, clear categorization of inflows and outflows, and a reliable link to your sales pipeline or order book.

From a strategic perspective, define a minimal but robust data model in BigQuery that captures your cash-relevant events: bank transactions, AP/AR schedules, expected collections, and planned payouts. This is where Reruption’s engineering depth is useful—we help clients quickly design a lean schema and ETL flows that are “good enough” for AI forecasting. With that in place, you avoid the common trap of blaming Gemini for errors that actually come from missing or inconsistent source data.

Position Gemini as a Copilot for Finance, Not a Black Box

Finance teams won’t trust a model that simply spits out a number for “cash balance in 14 days” without context. Strategically, you should position Gemini-powered cash forecasting as a copilot that both predicts and explains. Let the model highlight drivers: shifts in average collection days, unusual vendor payments, seasonal patterns, and exception transactions.

This mindset change is crucial for adoption. Instead of replacing your treasury or FP&A team, Gemini augments them: it surfaces anomalies, builds alternative scenarios, and generates plain-language narratives that finance leaders can challenge, refine, and present. When people understand that they stay in control of decisions, they become allies in improving the model over time.

Align Treasury, FP&A, and IT Around Clear Use Cases

Short-term cash forecasting cuts across teams: treasury owns bank positions, FP&A owns planning assumptions, and IT owns data infrastructure. Without alignment, your Gemini initiative will stall in hand-offs and governance debates. Strategically, start with a clearly scoped use case like “7–21 day rolling cash forecast with daily refresh and alerting on projected shortfalls”.

Bring the relevant stakeholders together to agree on definitions (what counts as cash-relevant), acceptable error bands, and the first set of alerts or reports Gemini should generate. This upfront alignment reduces rework, keeps IT focused on the essential integrations, and ensures the forecasting output matches real decision needs—such as “Do we draw down the credit line next week or not?”.

Manage Risk with Transparent Metrics and Phased Rollout

Any move from spreadsheet averages to AI-driven cash forecasting introduces change risk. Treat this as a controlled experiment, not a big-bang replacement. In the first phase, run Gemini’s forecasts in parallel with your existing process and compare accuracy, stability, and lead time on detecting shortfalls.

Define transparency metrics that matter to finance leadership: forecast error by horizon, number of days of warning before a cash dip, and the share of manual adjustments still required. Use these metrics to decide when to expand the AI scope (more entities, more currencies, more data sources). A phased rollout, backed by clear KPIs, keeps risk acceptable while giving your team the confidence to rely more heavily on the AI output.

Used correctly, Gemini is less about fancy models and more about building a reliable, explainable short-term cash forecasting system that finance teams can actually trust. By combining clean transaction data in BigQuery, familiar interfaces like Sheets, and Gemini’s ability to detect patterns and generate narratives, you can reduce surprises and use your credit facilities more deliberately. Reruption works hands-on with clients to design this architecture, build the integrations, and embed it into daily finance workflows—if you’re considering such a setup, we’re happy to explore whether a focused PoC or pilot would make sense for your team.

Need help implementing these ideas?

Feel free to reach out to us with no obligation.

Real-World Case Studies

From Energy to Telecommunications: Learn how companies successfully use Gemini.

Shell

Energy

Unplanned equipment failures in refineries and offshore oil rigs plagued Shell, causing significant downtime, safety incidents, and costly repairs that eroded profitability in a capital-intensive industry. According to a Deloitte 2024 report, 35% of refinery downtime is unplanned, with 70% preventable via advanced analytics—highlighting the gap in traditional scheduled maintenance approaches that missed subtle failure precursors in assets like pumps, valves, and compressors. Shell's vast global operations amplified these issues, generating terabytes of sensor data from thousands of assets that went underutilized due to data silos, legacy systems, and manual analysis limitations. Failures could cost millions per hour, risking environmental spills and personnel safety while pressuring margins amid volatile energy markets.

Lösung

Shell partnered with C3 AI to implement an AI-powered predictive maintenance platform, leveraging machine learning models trained on real-time IoT sensor data, maintenance histories, and operational metrics to forecast failures and optimize interventions. Integrated with Microsoft Azure Machine Learning, the solution detects anomalies, predicts remaining useful life (RUL), and prioritizes high-risk assets across upstream oil rigs and downstream refineries. The scalable C3 AI platform enabled rapid deployment, starting with pilots on critical equipment and expanding globally. It automates predictive analytics, shifting from reactive to proactive maintenance, and provides actionable insights via intuitive dashboards for engineers.

Ergebnisse

  • 20% reduction in unplanned downtime
  • 15% slash in maintenance costs
  • £1M+ annual savings per site
  • 10,000 pieces of equipment monitored globally
  • 35% industry unplanned downtime addressed (Deloitte benchmark)
  • 70% preventable failures mitigated
Read case study →

Insilico Medicine

Biotech

The drug discovery process traditionally spans 10-15 years and costs upwards of $2-3 billion per approved drug, with over 90% failure rate in clinical trials due to poor efficacy, toxicity, or ADMET issues. In idiopathic pulmonary fibrosis (IPF), a fatal lung disease with limited treatments like pirfenidone and nintedanib, the need for novel therapies is urgent, but identifying viable targets and designing effective small molecules remains arduous, relying on slow high-throughput screening of existing libraries. Key challenges include target identification amid vast biological data, de novo molecule generation beyond screened compounds, and predictive modeling of properties to reduce wet-lab failures. Insilico faced skepticism on AI's ability to deliver clinically viable candidates, regulatory hurdles for AI-discovered drugs, and integration of AI with experimental validation.

Lösung

Insilico deployed its end-to-end Pharma.AI platform, integrating generative AI and deep learning for accelerated discovery. PandaOmics used multimodal deep learning on omics data to nominate novel targets like TNIK kinase for IPF, prioritizing based on disease relevance and druggability. Chemistry42 employed generative models (GANs, reinforcement learning) to design de novo molecules, generating and optimizing millions of novel structures with desired properties, while InClinico predicted preclinical outcomes. This AI-driven pipeline overcame traditional limitations by virtual screening vast chemical spaces and iterating designs rapidly. Validation through hybrid AI-wet lab approaches ensured robust candidates like ISM001-055 (Rentosertib).

Ergebnisse

  • Time from project start to Phase I: 30 months (vs. 5+ years traditional)
  • Time to IND filing: 21 months
  • First generative AI drug to enter Phase II human trials (2023)
  • Generated/optimized millions of novel molecules de novo
  • Preclinical success: Potent TNIK inhibition, efficacy in IPF models
  • USAN naming for Rentosertib: March 2025, Phase II ongoing
Read case study →

Lunar

Banking

Lunar, a leading Danish neobank, faced surging customer service demand outside business hours, with many users preferring voice interactions over apps due to accessibility issues. Long wait times frustrated customers, especially elderly or less tech-savvy ones struggling with digital interfaces, leading to inefficiencies and higher operational costs. This was compounded by the need for round-the-clock support in a competitive fintech landscape where 24/7 availability is key. Traditional call centers couldn't scale without ballooning expenses, and voice preference was evident but underserved, resulting in lost satisfaction and potential churn.

Lösung

Lunar deployed Europe's first GenAI-native voice assistant powered by GPT-4, enabling natural, telephony-based conversations for handling inquiries anytime without queues. The agent processes complex banking queries like balance checks, transfers, and support in Danish and English. Integrated with advanced speech-to-text and text-to-speech, it mimics human agents, escalating only edge cases to humans. This conversational AI approach overcame scalability limits, leveraging OpenAI's tech for accuracy in regulated fintech.

Ergebnisse

  • ~75% of all customer calls expected to be handled autonomously
  • 24/7 availability eliminating wait times for voice queries
  • Positive early feedback from app-challenged users
  • First European bank with GenAI-native voice tech
  • Significant operational cost reductions projected
Read case study →

JPMorgan Chase

Banking

In the high-stakes world of asset management and wealth management at JPMorgan Chase, advisors faced significant time burdens from manual research, document summarization, and report drafting. Generating investment ideas, market insights, and personalized client reports often took hours or days, limiting time for client interactions and strategic advising. This inefficiency was exacerbated post-ChatGPT, as the bank recognized the need for secure, internal AI to handle vast proprietary data without risking compliance or security breaches. The Private Bank advisors specifically struggled with preparing for client meetings, sifting through research reports, and creating tailored recommendations amid regulatory scrutiny and data silos, hindering productivity and client responsiveness in a competitive landscape.

Lösung

JPMorgan addressed these challenges by developing the LLM Suite, an internal suite of seven fine-tuned large language models (LLMs) powered by generative AI, integrated with secure data infrastructure. This platform enables advisors to draft reports, generate investment ideas, and summarize documents rapidly using proprietary data. A specialized tool, Connect Coach, was created for Private Bank advisors to assist in client preparation, idea generation, and research synthesis. The implementation emphasized governance, risk management, and employee training through AI competitions and 'learn-by-doing' approaches, ensuring safe scaling across the firm. LLM Suite rolled out progressively, starting with proofs-of-concept and expanding firm-wide.

Ergebnisse

  • Users reached: 140,000 employees
  • Use cases developed: 450+ proofs-of-concept
  • Financial upside: Up to $2 billion in AI value
  • Deployment speed: From pilot to 60K users in months
  • Advisor tools: Connect Coach for Private Bank
  • Firm-wide PoCs: Rigorous ROI measurement across 450 initiatives
Read case study →

Unilever

Human Resources

Unilever, a consumer goods giant handling 1.8 million job applications annually, struggled with a manual recruitment process that was extremely time-consuming and inefficient . Traditional methods took up to four months to fill positions, overburdening recruiters and delaying talent acquisition across its global operations . The process also risked unconscious biases in CV screening and interviews, limiting workforce diversity and potentially overlooking qualified candidates from underrepresented groups . High volumes made it impossible to assess every applicant thoroughly, leading to high costs estimated at millions annually and inconsistent hiring quality . Unilever needed a scalable, fair system to streamline early-stage screening while maintaining psychometric rigor.

Lösung

Unilever adopted an AI-powered recruitment funnel partnering with Pymetrics for neuroscience-based gamified assessments that measure cognitive, emotional, and behavioral traits via ML algorithms trained on diverse global data . This was followed by AI-analyzed video interviews using computer vision and NLP to evaluate body language, facial expressions, tone of voice, and word choice objectively . Applications were anonymized to minimize bias, with AI shortlisting top 10-20% of candidates for human review, integrating psychometric ML models for personality profiling . The system was piloted in high-volume entry-level roles before global rollout .

Ergebnisse

  • Time-to-hire: 90% reduction (4 months to 4 weeks)
  • Recruiter time saved: 50,000 hours
  • Annual cost savings: £1 million
  • Diversity hires increase: 16% (incl. neuro-atypical candidates)
  • Candidates shortlisted for humans: 90% reduction
  • Applications processed: 1.8 million/year
Read case study →

Best Practices

Successful implementations follow proven patterns. Have a look at our tactical advice to get started.

Centralise Cash-Relevant Data in BigQuery

Before you involve Gemini, consolidate your cash-relevant data into a single, queryable source—typically BigQuery. Ingest at least: bank transaction feeds, accounts receivable and payable schedules, payroll runs, tax payments, and key pipeline or order data. Use simple, scheduled ETL jobs (from your ERP, CRM, and banking APIs) to refresh these tables daily or multiple times per day.

Design a compact schema: a unified cash_events table with columns like date, amount, currency, counterparty, category (e.g. customer_receipt, supplier_payment, salary), and source_system. This simplifies queries that Gemini will consume and keeps your logic transparent for finance and IT.

Use Gemini with BigQuery to Generate Daily Rolling Forecasts

Once your data is centralised, you can use Gemini + BigQuery to generate short-term forecasts. A practical pattern is: BigQuery runs a SQL query to aggregate historical behavior by customer, vendor, weekday, and payment terms; Gemini then takes the aggregated result plus upcoming invoices and predicts expected cash movements for the next 7–30 days.

For example, you can orchestrate a daily Cloud Function (or Cloud Run job) that:

  • Runs a BigQuery query to output all upcoming inflows/outflows and their historical payment behavior.
  • Passes the result as structured JSON into Gemini via the API.
  • Writes Gemini’s forecasted daily cash positions into a BigQuery table that feeds Sheets or a dashboard.
Example Gemini system prompt for forecasting:
You are a cash forecasting model for the finance department.
Using the provided historical payment patterns and upcoming invoices,
produce a day-by-day cash inflow and outflow forecast for the next 21 days.

Requirements:
- Output daily totals (inflow, outflow, net, running balance)
- Explicitly model weekends and holidays based on past patterns
- Highlight assumptions (e.g., average collection days per customer segment)
- Return JSON with a 'daily_forecast' array and an 'assumptions' section.

This setup means your forecasts update automatically, without manual spreadsheet work, and finance can still inspect the underlying SQL and JSON.

Embed Forecasts and Explanations Directly in Google Sheets

Most finance teams live in spreadsheets, so make Google Sheets the front-end for your Gemini-powered forecasts. Use the BigQuery connector to pull in the daily_forecast table, and then use Apps Script or Workspace add-ons to call Gemini for narrative explanations and scenario commentary.

For example, you can have a “Forecast Explanation” sheet where a script sends the latest forecast plus some metadata to Gemini and writes back a human-readable summary for management.

Example prompt for narrative explanation:
You are supporting the CFO with short-term cash insights.
Given this 21-day cash forecast (JSON) and the corresponding
prior 21-day forecast, explain:
- Key changes vs. last forecast
- Main drivers (customers, vendors, timing shifts)
- Risks: days with low buffer vs. credit facilities
- Recommended actions in clear, non-technical language.

Write 3–5 short paragraphs, suitable for inclusion
in a CFO weekly cash report.

This turns the model into a reporting assistant that saves hours of manual commentary writing while increasing consistency and clarity.

Configure Scenario Simulation Workflows

Unreliable forecasts often ignore “what if” questions: what if a major customer pays 10 days late, or you bring a large CAPEX forward by a month? You can use Gemini to generate scenario-based cash forecasts by passing in the base forecast and a list of hypothetical changes.

For example, build a small web form or Sheet where finance can specify scenario levers. A backend script takes the base data, applies the levers, and sends both base and scenario to Gemini for recomputation and explanation.

Example scenario prompt:
You are a scenario simulation engine for short-term cash.
Base forecast: <JSON_BASE>
Scenario adjustments: <JSON_SCENARIO_LEVERS>

Tasks:
1) Apply the scenario levers (e.g., delay customer X by 10 days,
   shift vendor group Y payments earlier by 5 days).
2) Recalculate the 21-day daily cash position.
3) Return:
   - 'scenario_forecast' (daily view)
   - 'delta_vs_base' (per day differences)
   - A concise explanation of the impact on liquidity risk.

This gives treasury a practical way to stress-test cash positions without building separate complex models for each scenario.

Set Up Alerting on Projected Shortfalls and Anomalies

Forecasts only matter if someone acts in time. Use Gemini’s pattern recognition to generate alerts when projected cash balances approach defined thresholds or when the pattern of inflows/outflows looks unusual compared to history.

Implement a daily job that:

  • Reads the latest 14–30 day forecast from BigQuery.
  • Identifies days where the projected balance is below a configured buffer.
  • Calls Gemini to classify the severity and suggest next steps (e.g., accelerate collections, delay non-critical payments, draw down lines).
  • Sends an email or chat message to treasury/finance channels with a concise summary.
Example alerting prompt:
You are monitoring short-term cash risk.
Given this 14-day forecast and company liquidity policy
(minimum cash buffer, available credit lines), identify:
- Dates where projected cash < minimum buffer
- Likely drivers (based on inflow/outflow composition)
- 3 concrete mitigation options for treasury.

Answer in bullet points, suitable for an email alert.

With this in place, your team doesn’t have to continuously scan dashboards to catch emerging issues; Gemini surfaces the important ones proactively.

Track Accuracy and Continuously Refine the Model

To build trust, treat the Gemini-based forecast like any other model: track its forecast accuracy and bias over time. Store both predicted and actual daily cash positions in BigQuery. Regularly compute metrics such as mean absolute percentage error by horizon (e.g., 3, 7, 14 days out) and identify where errors cluster (specific entities, currencies, customer segments).

Use these metrics as input to fine-tune prompts, adjust pre-processing logic (for example, better modeling of payment terms), or segment the model (e.g. separate logic for enterprise vs. SMB customers). Over a few months, you should see error bands tighten and fewer manual adjustments needed. Make this improvement visible to stakeholders—this is how you move from “experimental AI” to a trusted part of your finance toolkit.

Expected outcome: organisations that implement these practices typically see material improvements in short-term cash visibility—for example, reducing manual forecasting effort by 30–50%, detecting cash shortfalls 5–10 days earlier than before, and cutting unplanned use of expensive credit lines. The exact numbers will depend on your data quality and complexity, but the direction of travel is clear once a robust Gemini-powered workflow is in place.

Need implementation expertise now?

Let's talk about your ideas!

Frequently Asked Questions

Gemini improves short-term cash forecasting by detecting patterns your spreadsheets typically miss: day-of-week effects, customer- or vendor-specific payment behavior, seasonality, and the interaction between upcoming invoices and historical delays. Instead of relying on simple averages, Gemini can combine transaction histories, open AR/AP, and pipeline data to predict when cash will actually move.

In practice, you still define the data inputs and business rules (e.g. minimum buffers, treatment of one-off items), but Gemini automates the heavy pattern recognition and provides daily rolling forecasts and explanations. This reduces manual work and helps you anticipate shortfalls earlier than with static Excel models.

At minimum, you need three capabilities: data engineering to feed your ERP, banking, and CRM data into BigQuery; someone with finance or treasury expertise to define what should be in scope and how to interpret results; and basic familiarity with Google Cloud and Workspace to connect Gemini, BigQuery, and Sheets.

You do not need an in-house AI research team. A small, cross-functional squad (finance lead + IT/data engineer) can implement a first pilot, especially if they work with a partner like Reruption that brings ready-made patterns for pipelines, prompt design, and governance. From there, you can train more people internally to maintain and extend the solution.

If your transaction and invoice data are already accessible, a focused pilot for short-term cash forecasting with Gemini usually takes weeks, not months. A typical timeline is 2–4 weeks to connect BigQuery, define the data model, and build a first forecasting workflow, and another 2–4 weeks to run it in parallel, tune prompts, and validate accuracy against your existing process.

You’ll see early insights (e.g. improved visibility on the next 7–14 days, better explanation of deviations) within the first iteration, even if the forecast is not perfect yet. The key is to start with a narrow scope—e.g. one legal entity or one bank account—and expand as trust and data quality improve.

The direct costs of Gemini + BigQuery for finance are typically modest relative to the impact. You pay for data storage and queries in BigQuery, and for API or Workspace usage of Gemini, which is usually a small line item compared to the potential reduction in unplanned overdraft interest or suboptimal cash deployment.

On the ROI side, finance teams often see value through fewer last-minute funding gaps, lower reliance on expensive credit lines, better use of surplus cash, and reduced manual effort in preparing and explaining forecasts. To make the business case concrete, track metrics such as manual hours saved per cycle, change in average utilization of credit facilities, and how many days in advance you can now detect a potential shortfall.

Reruption combines AI engineering and a Co-Preneur mindset to build working solutions inside your finance organisation, not just slideware. With our AI PoC offering (9,900€), we can quickly validate whether Gemini-based cash forecasting works with your real data: define the use case, design the architecture around BigQuery and Sheets, build a functional prototype, and measure performance and robustness.

We then help you turn that PoC into a production-ready workflow: hardening integrations, setting up governance and monitoring, and embedding the tool into your existing treasury and FP&A routines. Because we work “in your P&L”, we focus on outcomes—more reliable short-term forecasts, earlier visibility into cash risks, and a practical system your finance team actually uses and trusts.

Contact Us!

0/10 min.

Contact Directly

Your Contact

Philipp M. W. Hoffmann

Founder & Partner

Address

Reruption GmbH

Falkertstraße 2

70176 Stuttgart

Social Media