Power Platforms for Business

Watch it in action

A short, plain-English explainer of what an ERP system is and why a growing business uses one.

"What is ERP? — A Simple Explanation" — by netX consult · watch on YouTube

What is ERP, in plain English?

ERP stands for Enterprise Resource Planning — but forget the jargon. It's simply one connected system for running the business: accounting, inventory, sales, purchasing, HR, and projects all in one place, instead of a dozen disconnected apps and spreadsheets that never quite talk to each other. Think of it as one connected nervous system for the business: when a sale happens, the same event updates your stock levels, your revenue figures, and your customer record automatically — no re-typing the same information into three different tools and hoping the numbers match up later.

Most small businesses grow into a familiar mess. Sales live in one spreadsheet, invoices in the accounting app, stock counts on a whiteboard in the back room, and customer notes in someone's inbox. Every one of those hand-offs is a place where a number gets copied wrong, a step gets forgotten, or two people end up looking at two different versions of the truth. ERP fixes that by making every part of the business read from and write to the same shared set of records. Ask "how much did we actually make last month, after refunds and unpaid invoices?" and there's a single, trustworthy answer instead of an afternoon of reconciling files.

"Open-source" means the software is free to use, you can self-host it (run it on your own account or server), and you can customize it — with Claude's help — instead of paying rising per-seat SaaS subscription fees for features you can't change. As proprietary license costs climb, open-source ERP adoption is growing fast: you get enterprise-grade capability while keeping control of your data and your budget.

Why does that ownership matter? With a typical subscription tool, your data lives on someone else's servers, the price per user tends to creep up every renewal, and the features you get are the features they decide to build. When you self-host an open-source ERP, the data is yours to export, back up, or move at any time, and the cost is mostly the modest hosting bill rather than a per-seat fee that multiplies as you hire. That is a meaningful difference for a growing team: adding your tenth or twentieth employee doesn't quietly add another line to a monthly invoice.

The three open-source ERPs worth knowing

There are dozens of ERP products out there, but for a small or medium business three open-source options stand out. They differ mostly in how much power they give you versus how simple they are to run. The table below is the quick version; pick the one whose "best for" row sounds most like you today, knowing you can always grow into a bigger system later.

ERPNext

Odoo

Frappe

Dolibarr

How to choose, in one breath: if you want the best all-round fit and room to grow, pick ERPNext. If you're a bigger or fast-scaling operation that already has, or can hire, a little technical help and wants the widest set of ready-made modules, look at Odoo. If you run a small, straightforward operation and just want the essentials up and running with the least fuss, Dolibarr is the gentle on-ramp. None of these choices is a trap — because your data is yours and exportable, you can start with the simplest option and move up later if you outgrow it.

What does that feel like day to day? A wholesaler stops guessing which products are running low because stock updates itself with every sale and purchase. A manufacturer sees, in one screen, which orders are waiting on which materials. An owner pulls a real profit-and-loss for last quarter in seconds instead of stitching together exports from three tools. The magic isn't any single feature — it's that the numbers are finally connected, current, and in one place, so the questions that used to take a day to answer now take a moment.

What switching actually looks like

Moving to an ERP sounds like a huge project, and for a giant corporation it can be. For a small business it's far more manageable — and you don't have to do it all at once. The sensible approach is to start with the one area that hurts most (often inventory or accounting), get that running cleanly, and then bring in sales, purchasing, and the rest module by module as you get comfortable. Your existing data usually comes across from spreadsheets or your old tools as simple CSV files, and Claude can help you clean up and map those columns so they land in the right places.

It's worth being honest about the trade-offs at a high level. The upside is real: one source of truth, no per-seat fees, and software you can bend to your process. The cost is some up-front setup time and a short learning curve for your team, since a connected system asks people to record things consistently rather than scribbling them wherever is handy. The good news is that the payoff compounds — every month the data stays clean, your reports get more trustworthy and your decisions get easier.

Customizing ERPNext with Claude

Here's the part that changes the game for a non-technical owner. Traditionally, changing how ERP works — adding a field, tweaking a workflow, building a new report — meant filing a request with a consultant, waiting days or weeks, and paying by the hour. Because ERPNext is built on the low-code Frappe framework, you can instead make many of those changes yourself simply by describing what you want to Claude in ordinary language. You explain the outcome; Claude handles the technical steps and tells you what it did.

Once ERPNext is installed, you shape it to fit how you actually work. Just tell Claude what you want:

I've installed ERPNext. Help me add a custom field to the Sales Order for 'delivery route', and build a report that groups orders by route so I can plan our deliveries.

The same plain-English approach works for the everyday tweaks that make a system truly yours — renaming a status, adding an approval step for large purchases, or wiring up an automatic email. A few more examples you can adapt:

Set up a rule in ERPNext so any Purchase Order over $5,000 needs manager approval before it can be submitted, and email me a summary of what's waiting for approval each morning.

What is CRM, in plain English?

CRM stands for Customer Relationship Management. In plain terms, it's your single source of truth for leads, contacts, deals, and follow-ups — so nothing slips through the cracks and everyone on the team sees the same up-to-date picture of each customer. Like ERP, there are excellent open-source options you can self-host and customize.

If ERP is the nervous system for your operations, CRM is the memory for your relationships. Every business already has a CRM of sorts — it's just usually scattered across sticky notes, a personal phone, a shared inbox, and whatever the last salesperson happened to remember. The trouble with that scattered version is that it walks out the door when an employee leaves, and it goes quiet exactly when it matters most: the promising lead who asked you to "check back next month" and then never heard from you again. A real CRM turns those loose threads into a visible, shared list of who to talk to next and why.

Why it matters for a growing business: the cost of a lost lead is invisible, which is what makes it so dangerous — you never see the sale that didn't happen. A CRM makes the invisible visible. It shows you how many deals are in flight, which ones are stalling, and which have gone cold, so your follow-up is driven by the pipeline rather than by whoever happens to shout loudest. It also means the picture of a customer survives staff changes: when someone new picks up an account, the full history of past conversations, quotes, and orders is right there.

Picture a small services firm — a design studio, an accountancy, a contractor. Enquiries arrive by email, phone, and referral all week. In a CRM, each one becomes a card that moves along a pipeline from first contact to signed contract, with reminders so nobody has to hold it all in their head. The owner can glance at the board on Monday morning and instantly see the three deals that need a nudge this week — the single most valuable habit a small sales operation can build.

The open-source CRMs worth knowing

As with ERP, there's no single "best" CRM — there's the one that fits how you sell today. The four below range from a clean, modern fresh start to a mature system built for complex sales processes, plus the option of using the CRM already baked into ERPNext if you run it.

Twenty

EspoCRM

SuiteCRM

ERPNext CRM

Set it up by talking to Claude

The heart of any CRM is the pipeline — the set of stages a deal moves through from first contact to closed. Defining those stages is really just writing down how you already sell, in order. Getting your contacts in and a follow-up process running is a plain-English conversation:

Help me import my contacts from this CSV into Twenty, map the columns correctly, and set up a follow-up pipeline with stages: New Lead → Contacted → Proposal Sent → Won/Lost. Add a reminder to follow up if a lead sits in one stage for more than 7 days.

Once the basics are in place, the same conversational approach keeps the CRM working for you instead of becoming another chore. You can ask for the weekly summaries and nudges that turn a static contact list into an active sales habit:

Every Monday, give me a list of open deals sorted by value, flag any that haven't been touched in two weeks, and draft a short, friendly check-in email I can send to each of those stalled leads.

The one thing that makes or breaks a CRM isn't the software — it's the habit of keeping it current. A tool full of stale, half-entered records quickly becomes something the team stops trusting and then stops using. The fix is to make updating it almost effortless: keep the fields minimal, capture new leads the moment they arrive, and lean on the automatic reminders so the system nudges people rather than relying on willpower. Done that way, five minutes of upkeep a day buys you a picture of your sales that's worth far more than the hour of guesswork it replaces.

Watch it in action

A plain-English look at how Google's TimesFM forecasts time-series data like sales and demand.

"TimesFM: Forecasting Made Easy" — by Hands-on AI · watch on YouTube

What is TimesFM, in plain English?

TimesFM is Google Research's pretrained foundation model for time-series forecasting — a decoder-only model that predicts what a number will do next based on its history. The latest version is TimesFM 2.5, released March 31, 2026. In plain terms: feed it the history of any number that changes over time, and it tells you where that number is likely headed. Perfect for the questions SMBs ask constantly — next month's sales, how much of each product to stock, your cash-flow outlook, website traffic, and staffing needs.

The best part: it's zero-shot, meaning it forecasts without you training a custom model first. Under the hood it has about 200 million parameters, supports up to 16,000 historical time-points of context, and can forecast up to a 1,000-step horizon. It was pre-trained on roughly 400 billion time-points drawn from Google Trends, Wikipedia page views, financial markets, retail sales, and IoT sensors. On retail datasets it cut error (MAE) by roughly 15–25% versus classic ARIMA forecasting.

A quick word on what "foundation model" and "zero-shot" mean here, because they're the reason this is now within reach for a small business. A foundation model is a large model pre-trained once, at great expense, on an enormous amount of data — so that everyone else can reuse it without repeating that work. Zero-shot means you point it at your own numbers and get a forecast immediately, with no training step of your own. In practical terms, the hard and expensive part has already been done for you; you just bring your history and ask the question.

When is forecasting actually worth it? Not every number needs a model, and it's fair to ask. Forecasting earns its keep when you have to commit resources ahead of time against an uncertain future — ordering stock weeks before it sells, scheduling staff before the rush, or setting a budget before the revenue lands. It works best when you have a decent run of history (many months of regular data), when there's some pattern to lean on such as seasonality or a weekly rhythm, and when being roughly right in advance beats being exactly right too late. If a number is pure coin-flip random with no history, no model can save you — but most business numbers aren't like that.

What can you forecast?

Two features that make forecasts genuinely useful

Quantile forecasts (a range you can plan around). Instead of a single number, TimesFM can give you a probabilistic range — a best-case and worst-case band around the prediction. So rather than "we'll sell 500 units," you get "likely between 430 and 580." That range is what lets you plan safely: stock for the high end, budget for the low end.

Covariates (tell it about your special events). Covariates are simply external factors you tell the model about — like holidays and promotions. If you have a Black Friday sale coming, you can tell TimesFM, and it will account for that spike instead of being caught off guard. The model can't know about a promotion you haven't run before, but you do — so you hand it that context, and the forecast stops treating your planned spike as a surprise.

How to read a forecast (without being a statistician)

A forecast has two parts worth separating in your head. The point forecast is the single most-likely number — the model's best single guess, useful for a headline like "we expect about 500 units next month." The quantile range is the band of realistic outcomes around it — say 430 on the low end and 580 on the high end. The gap between those two edges is really a measure of confidence: a narrow band means the model sees a steady, predictable pattern, while a wide band is the model honestly telling you "this one is hard to call, so leave yourself room."

The practical trick is to match the edge of the range to the risk you're managing. When running out is the expensive mistake — a missed sale, an idle crew — plan toward the high end so you're covered. When overcommitting is the expensive mistake — perishable stock, cash tied up in inventory — plan toward the low end so you don't get stuck. The point forecast tells you what to expect; the range tells you how much cushion to build in.

A few concrete pictures make it click. A wholesaler forecasts demand for each SKU separately so the fast movers get reordered early and the slow ones don't pile up in the warehouse. A retailer planning a holiday promotion feeds last year's sales plus the promo dates as a covariate, then stocks to the upper end of the range so a sold-out shelf doesn't turn eager customers away. A café or clinic forecasts footfall by day of week to schedule just enough staff — enough to serve the rush without paying people to stand around during the lull.

Two more terms help you read the settings without a glossary. The context is simply how much history you feed in — the longer and more regular your record, the more pattern the model has to work with. The horizon is how far ahead you're asking it to look; a near-term horizon (the next few weeks) is naturally more reliable than one stretching far into the future, in exactly the way a three-day weather forecast beats a three-week one. A sensible habit is to forecast only as far out as you actually need to make the decision in front of you.

The easiest ways to use it

You don't need to stand up any machine-learning infrastructure, and you certainly don't need to understand the model's internals. Depending on where your data already lives, one of these paths will fit. The simplest paths:

• BigQuery: the AI.FORECAST and AI.EVALUATE SQL functions are now generally available — no ML setup needed. AlloyDB supports it too.
• HuggingFace: use the model directly, and fine-tune it with LoRA/PEFT if you have a special case.
• Open-source on GitHub: the full model is open at google-research/timesfm.

A BigQuery example (SQL)

SELECT * FROM AI.FORECAST(
  TABLE sales.daily_revenue,
  data_col => 'revenue',
  timestamp_col => 'date',
  horizon => 30
);

...or just describe it to Claude

Using our 2 years of daily sales in BigQuery, forecast the next 30 days with TimesFM, include an upper/lower range, and account for our Black Friday promo.

Whichever path you choose, the workflow is the same in spirit: gather a clean run of history, ask for a forecast over a sensible horizon, include a range and any covariates you know about, and then read the result as a decision aid rather than a crystal ball. A forecast is a well-informed expectation, not a promise — treat it as the starting point for a plan, revisit it as new numbers come in, and it will steadily earn your trust.

Learn more: TimesFM on GitHub · TimesFM in BigQuery & AlloyDB

Each piece on this page is powerful on its own — but the real payoff comes when they work together. Your ERP/CRM holds the business data. You export or connect that data to TimesFM, which produces forecasts. You display the current numbers and the forecast together in a dashboard, hosted (see the Build & Ship page) on services like Railway and Neon. And finally, your team acts on what they see.

The reason this loop is worth building is that each piece solves a problem the others create. An ERP or CRM is fantastic at recording what has happened, but it mostly looks backward — it tells you where you've been, not where you're going. A forecasting model is brilliant at looking forward, but it's useless without clean history to learn from. And a forecast buried in a spreadsheet nobody opens changes no decisions at all. Chain them together and each one covers the last one's blind spot: the ERP/CRM supplies trustworthy history, TimesFM turns that history into a view of the future, and the dashboard puts both in front of the people who make the calls.

Notice that your data feeds the dashboard twice: once as the current, real numbers, and once through TimesFM as the forecast. Side by side, your team sees where things stand and where they're headed — the whole point of the exercise.

Here's the loop in one everyday example. A small wholesaler's ERPNext holds every sale and every stock level. Once a week those figures flow to TimesFM, which returns a 30-day demand forecast per product, complete with a high/low range. The dashboard shows today's stock next to next month's expected demand, quietly highlighting the handful of items likely to run short. The buyer opens it Monday morning, sees the flagged products, and places a reorder before anything sells out — a decision made calmly in advance instead of scrambled together after an angry customer calls. That's the difference the loop makes: it moves you from reacting to what already went wrong toward planning for what's about to happen.

You don't have to build the whole loop on day one, and you shouldn't try. The wise path is to add one link at a time: get your ERP or CRM holding clean data first, then wire up a single forecast for the one number that keeps you up at night, then put that number on a simple dashboard your team checks on a regular rhythm — weekly is plenty for most decisions. Each small, working piece is easy to trust, easy to fix, and immediately useful, so the project pays for itself long before it's "finished."

Best of all, you build this the same way you built everything else on this page — by describing what you want to Claude, one step at a time. You don't need to be an engineer to connect a data export to a forecast and a simple chart; you need to know what question you're trying to answer, and to keep the pieces small enough that each one is easy to check.

Here's every platform mentioned on this page in one place, so you can jump straight to the official site to sign up, download, or read the docs. A good next step is to pick just one — most likely ERPNext for operations or Twenty for a standalone CRM — and spin up a trial you can poke around in while Claude walks you through the setup.

Open-Source ERP

Open-Source CRM

AI Forecasting

Keep Building

1. Course: the vibe-coding foundations (Home)
2. Build & Ship: databases, hosting, and deployment (Build & Ship)
3. Power Platforms: ERP/CRM + AI forecasting (this page)