The Connector that Holds vs. the Connector that Flakes

Picture this: It is 3:12 PM on a Tuesday. Diane, Controller at a $90M specialty distribution business, is on video with an FP&A vendor she has been evaluating for two weeks. The demo has been running 38 minutes. The AE is showing her a live ERP pull. Actuals streaming into a dashboard. Variance columns lighting up. The salesperson uses the word "live" four times in three minutes.

Diane runs on Microsoft Dynamics GP. 15 entities, 8 cost centers per entity, 3 fiscal-year variants (calendar, broker-year, distributor-year). Her monthly close eats days 1 through 8 every month, most of it exporting trial balances out of the GP Windows client, pasting them into a 47-tab master workbook, cleaning up formula breaks, and reconciling intercompany eliminations by hand. She is watching this demo because the CFO gave her a mandate: get us off the workbook by fiscal year-end.

She asks the AE a question. "The GL account structure in this demo. Is that our chart, or a sample one?" The AE pauses. Sample. "Can we see it against our actual GP data before we sign?" The AE says the "real hookup" happens during implementation, after contract signature. The dashboard she has been watching for the last 38 minutes was pre-loaded from a CSV the sales engineer generated the day before. There is no live GP pull. There will not be one until she buys.

Every week there is a version of Diane on a call like this, watching a "live integration" demo that is not. If you are the finance leader on the buying side of an FP&A evaluation, and your business runs on Dynamics GP, Sage 100, MIP, Blackbaud Financial Edge NXT, SYSPRO, Multiview, Elite, PeopleSoft, or any of the on-prem and legacy GLs the cloud-native FP&A crowd politely does not mention, the demo you are being shown is often a story about the connector, not the connector itself.

This piece is what is actually under the hood. The four honest tiers of FP&A-to-GL integration, which GLs live in which tier, why the difference matters when a real month-end hits, and the 12 questions you can walk any vendor through to know what you are actually buying before you sign. The last section is a downloadable Scorecard that turns the questions into a rubric you can score two or three vendors on side by side.

‍

Most FP&A "integrations" are a demo trick

The claim above sounds cynical. It is not. It is a description of what happens in about 60% of the mid-market FP&A demos the finance buyers we talk to sit through. Three specific tricks show up, in decreasing order of how often we see them.

Trick one: the pre-loaded CSV. The sales engineer generates a CSV export from a sample chart of accounts, uploads it into the vendor's sandbox tenant the day before the demo, and shows you actuals lighting up in a dashboard. What you are watching is a static file import. There is no ERP connection. There is no scheduled pull. When you ask "is that our chart?" the answer is either "sample" or "your chart, remapped by hand."

Trick two: the wrong-ERP sandbox. Most cloud-native FP&A vendors built against QuickBooks Online, NetSuite, and maybe Sage Intacct. Those are the demo sandboxes. If you are on Dynamics GP or Sage 100 or MIP, the demo you are watching is often a QuickBooks Online tenant with your industry's account labels swapped in. The vendor will tell you "the shape is the same across GLs." The shape is not the same. GP's segmented chart, MIP's fund-restriction dimensions, and QBO's flat account list are three different data models. The integration mechanics are three different mechanics.

‍Trick three: the "certified partner." Some vendors have "certified integrations" that turn out to be a partner shop or in-house services team that runs a CIF or CSV file for you every month. That is a real integration. It is a file-based one. If the vendor describes it in the same paragraph as their real-time API path, they are conflating two different products. One is automation. The other is outsourced data entry.

These tricks land because cloud-native FP&A tools were built against cloud GLs. That was the API surface those vendors had. The mid-market on-prem installed base got skipped. When a buyer on GP or Sage 100 or MIP shows up on a demo, the vendor shows what they have built, and the word "integration" ends up covering four very different mechanisms.

‍

The four honest connector tiers

Every FP&A-to-GL integration in the market today lands in one of four tiers. The vocabulary is ours; the mechanics are described in the same terms by every vendor engineer we have ever compared notes with. This taxonomy is what the Integration Scorecard scores against. Read it once and demo scripts get easier to see through.

‍Tier 1: installed agent inside your environment. The vendor ships a Windows service (in Centage's case it is called CDI, Centage Data Integrator) that installs on a server your business already runs, ideally the one hosting your GL's SQL database. The agent authenticates with a dedicated SQL user, reads the GL's tables directly, extracts everything it needs (chart of accounts, base structures, transaction detail, account summary, account transactions, and typically GL budget versions), writes a flat file in the vendor's canonical format, and uploads it to the vendor's cloud on a schedule. Setup is a one-time IT touch: open a port, create a SQL user with read permissions, install the service, register a Windows Task Scheduler job. After that, the pull runs at 2 AM (or whenever you set it), and your finance team never thinks about it again. This is the tier that reaches on-prem and legacy GLs the cloud crowd cannot: Microsoft Dynamics GP, Dynamics NAV, Dynamics SL, Dynamics AX 2009 and 2012, Sage 100, Sage 300 (ACCPAC), Sage 500, SYSPRO, and QuickBooks Desktop (with some scheduling quirks that usually get run manually).

Tier 2: cloud API. The vendor talks to the GL's public API over HTTPS. Setup is a permissions grant on the GL side and a token exchange on the vendor side. Refresh cadence depends on the GL. Sage Intacct supports a nightly rolling pull. NetSuite exposes SuiteQL and REST endpoints. Microsoft Dynamics 365 Business Central speaks OAuth (the friction here is that customer Azure tenants often lock down BC permissions in ways the AE cannot tell you about on the demo). Blackbaud Financial Edge NXT has a documented marketplace integration path. QuickBooks Online is well-worn. Acumatica works but often requires objects to be imported into the client instance first. This tier is what most cloud-native FP&A tools built to. It is real, it works, and it does not reach the systems in Tier 1.

Tier 3: partner-facilitated. A partner shop or a scheduled adapter runs the extract on your behalf. Centage's MIP integration goes through Abila (MIP's parent). Emphasys is another example. From the buyer's perspective, this reads like an automated pull: files show up, the model refreshes, no one manually touches it. Under the hood, a scheduled adapter runs at a partner. If the partner adapter breaks (they occasionally do), the recovery path goes through the partner rather than your FP&A vendor's support team. That is a fair trade for reaching a GL that has no first-party API. Just know it is the tier.

Tier 4: file-based (CIF, CSV, SIF, IIF). You (or your IT admin) produce a flat file every month or every close cycle and upload it. Some vendors call this an integration; it is a supported data-loading path. Which GLs land here today, honestly? Epicor. Deltek. Multiview. Anything else not on the Tier 1 or Tier 2 lists. This tier is not automatically wrong. If the file is well-shaped, the schema is stable, and the upload is one screen with clear validation, a Tier-4 path can be a good production integration. What it is not is a scheduled automatic pull. Vendors who describe a Tier-4 path in the same sentence as their Tier-1 or Tier-2 path are conflating.

Every FP&A platform in the market ships some mix of these four tiers. What varies is which GLs land in which tier, and how honestly the vendor draws the map. The rest of this piece is 12 questions that force the map into daylight.

‍

The 12 questions to ask any FP&A vendor about your GL

Read these on a demo call. Ask them in order. Do not accept a "we will cover that in scoping" answer to any of them. Every one of these is answerable in a first demo. The Scorecard at the end of this piece lets you score up to three vendors on each question in a shared rubric.

1. Which of your existing customers is running the exact GL I have, and what are their industry and headcount? The right answer is a name (or an anonymized company profile), the industry, and a rough employee count. "We work with lots of GP shops" is not an answer. "One of our customers is a 700-person healthcare nonprofit running GP with 6 divisions; another is a $150M specialty manufacturer on Sage 100" is an answer. If the vendor cannot produce a specific customer for your GL, your GL is not in their real-world install base yet, whatever the marketing site says.

2. Show me exactly where the extraction happens. Is the code inside our environment, or outside it? Tier 1 code runs inside your firewall on a machine you own. Tier 2 code runs on the vendor's cloud, talking to your GL's API. Tier 3 code runs on a partner's infrastructure. Tier 4 code runs on your accountant's laptop, exporting a CSV. Ask the vendor to draw the arrow on screen. If they cannot, or if the answer changes across your GL slice, that is the answer.

3. What is the pull cadence, and can we change it? Never-real-time. Usually nightly. Sometimes on-demand. Any vendor claiming "real-time" against an on-prem GL is either exaggerating or referring to a very small sync window that is not real-time in the way the word is normally used. Ask specifically: what time does the pull run today for a customer of your size, how do we change it if we want to, and what happens if we need actuals right after a hot close? The vendor's answer here separates real support from a script.

‍4. Do you pull all of my account dimensions, not just the account code? A GP chart has segments. A MIP chart has fund and restriction dimensions. A Blackbaud FE NXT chart has projects and locations. A Sage 100 chart has divisions and departments. If the FP&A tool only pulls the primary account code, everything else has to be reconstructed inside the model, which usually means an Excel side-file, which is the whole problem you are trying to escape. Ask the vendor to list the dimensions they will pull from your GL by name. If they cannot, they have not built for your GL yet.

5. Do you pull budget versions from the GL, or only actuals? Most on-prem GLs hold a "posted budget" version alongside actuals. If the FP&A tool pulls only actuals, your forecast comparison base has to be reconstructed inside the model each cycle. Not always a dealbreaker (most FP&A budgets get built inside the FP&A tool anyway), but a real answer to know before contract.

‍6. Who installs the connector, how long does it take, and what does our IT admin have to do? For Tier 1, the honest answer is: "your IT admin spends about an hour once, setting up a SQL user with read permissions and a scheduled task; after that, no IT involvement." For Tier 2, the honest answer is: "your IT admin grants API permissions inside your GL and hands us the token; we handle the rest." For Tier 3 and Tier 4, the honest answer is: "our services team or partner does it for you; here is the timeline." If the vendor says "zero IT involvement" for a Tier 1 setup, they are skipping the SQL user and Task Scheduler steps, which are real steps that a real human does once.

7. What happens when we change our chart of accounts mid-year? Adding a new revenue account, splitting a cost center, retiring a fund, restructuring divisions. These are normal finance events. A connector that holds re-syncs the chart automatically. A connector that flakes silently drops the new accounts and posts them to a suspense line, or breaks the model until someone opens a ticket. Ask specifically: what is the customer's experience the first time they add a new account after go-live? The vendor's answer will tell you whether the connector was built once and shipped, or built and maintained.

8. What happens when we add a new company, entity, fund, or restriction? Same question, larger blast radius. A nonprofit adds a new federally-funded program. A distributor adds a new legal entity for a bolt-on acquisition. A healthcare group opens a new location with its own P&L. Does the FP&A tool absorb the new entity from the GL automatically, or does someone (yours or theirs) rebuild part of the model?

‍9. If we ever migrate GLs, what survives? This is the continuity question, and most FP&A vendors treat it as out of scope. It is not. If you are a GP shop looking at Business Central, a Sage 100 shop looking at Intacct, or a QuickBooks Desktop shop looking at anything, the FP&A model you build now needs to still work when the GL underneath it changes. Ask specifically: what does the vendor do when a customer moves GLs? Which pieces of the model (chart mapping, dimensions, historicals, formulas, budgets, forecasts) transfer, and which get rebuilt? A vendor who has done this migration for one or two customers can answer specifically. A vendor who has not will hedge.

10. What breaks the connector, in real customer history? Every connector has a small list of edge cases where it breaks. Real ones we have seen: daylight-savings changes flipping the scheduled pull time by an hour, older adapter versions with a hardcoded year that stop posting on January 1, Business Central permission changes that revoke the token silently, QuickBooks Desktop scheduling quirks that make the pull unreliable at exactly 2 AM. A vendor who says "nothing breaks" is either brand-new or not telling you the truth. A vendor who names three specific past breakages and describes how customers recovered is the vendor with a real support function behind the connector.

11. Who owns the connector once it is live, and what is the response time when it breaks? Named CSM or a ticket queue. Onshore or offshore. Same-day, next-day, or best-effort. If the vendor's connector runs at 2 AM and breaks at 2:15 AM, when is someone looking at it? For a mid-market finance team running a monthly close, this is the difference between the close eating one extra day and eating three.

12. What is the fallback path when the automated path cannot cover something? Every real integration has a manual escape hatch. A Tier 1 connector still needs a manual CIF import path when a customer's SQL server goes down for maintenance during month-end. A Tier 2 connector needs a way to hand-load a historical period the API does not expose. A vendor with a documented, supported manual fallback is a vendor whose automation is honest about its edges. A vendor whose demo pretends there is no fallback needed is a vendor whose customers will find one anyway, without help.

‍

What a connector that holds looks like in practice

The reason the 12 questions matter is that the difference between a Tier-1 or Tier-2 connector that holds and one that does not shows up on day 45 of production use, not on day 1 of the demo. In a good production integration, the finance team's experience of the connector is that it is not there. The actuals appear when they should. The variance report opens. The close moves.

Jenny Barker, Senior Budget Manager, HASCO. HASCO is a government and nonprofit housing organization managing 6,000-plus budget line items across its properties, a level of detail Excel could not hold without falling over. HASCO runs on Elite, an on-prem GL with a first-party Centage connector.

Her verbatim on what the actuals pull looks like in a live production month:

‍"...I hit a couple of buttons, wait a few minutes and all my actuals data comes right into Centage."

The read: the operative phrase is "a couple of buttons, a few minutes." The connector runs on a schedule; when Jenny needs the actuals refreshed on-demand for an urgent variance question, it is two clicks and a short wait, then the numbers are in. Nobody is exporting a trial balance. Nobody is pasting it into a master workbook. Nobody is checking whether last month's formula chain caught the new fund code. The GL still lives on-prem. The FP&A layer lives in Centage. The connector is what makes the two sides act like one system.

What that looks like across the customer roster:

• A specialty adhesive manufacturer running SYSPRO (an on-prem manufacturing ERP that has no financial-statement generator natively) with a Controller who runs a five-minute Excel-ready close because the CDI agent pulls the trial balance every night.  
• A regional healthcare nonprofit running Dynamics GP with 6 divisions, refreshing actuals automatically after every monthly close post, and using drill-through to answer the "why is this variance?" question in the same session it comes up.  
• A community food bank running Dynamics NAV natively, moving from days of formula cleanup after every close to daily budget-vs-actuals variance because the connector runs unattended.

None of these customers are on Tier 4. All of them evaluated at least one FP&A vendor whose "integration" turned out to be a monthly CSV upload their internal IT admin was going to be responsible for. The Scorecard exists to make that comparison legible before the contract is signed.

‍

Interrogate, then decide

Nothing in this piece disqualifies a Tier-4 vendor. If your GL is Epicor or Deltek or Multiview, the honest answer today is that most FP&A tools reach you through a supported file-based path. Some of those paths are well-shaped, and a well-run monthly file import is not a bad way to run production. What matters is knowing what you are buying. A Tier-4 connector priced and supported as Tier 4 is a fair deal. A Tier-4 connector sold as Tier 1 is a demo trick.

Two working rules for a finance leader in the middle of an evaluation. First, ask the 12 questions on a first demo, not during scoping. Every one of them is answerable up front by any vendor who has built the integration for real. Second, get the answers in writing before contract. "Real-time," "seamless," and "fully integrated" get diluted the closer you get to signature. Specific claims about SQL users, refresh cadences, dimensions pulled, and named customer references do not.

The other reason the 12 questions land in a checklist is that they are the load-bearing questions for the buyers we watch actually win their evaluations. If you run a mid-market finance function on GP, Sage 100, MIP, Blackbaud FE NXT, SYSPRO, Multiview, Elite, or PeopleSoft, the connector is not a technical detail. The connector is the reason the FP&A layer works or does not.

The connector is the moat. Everything else follows from it. If the connector does not hold, the monthly close still eats week one, the workbook chain still breaks in November, and the FP&A tool becomes another export destination for a trial balance you are already tired of running.

‍

Download the Legacy-GL FP&A Integration Scorecard

Download the Legacy-GL FP&A Integration Scorecard. The 12 questions from this piece as a scoring rubric, plus a monthly-close hours calculator (loaded cost per hour × dump / consolidate / variance / board-pack hours) and a GL-migration continuity planner (what survives if you ever change GLs). Score up to three vendors side by side in the same file.

‍

If this shape of problem is what your last three demos have looked like, we should talk. Centage runs live customers on Microsoft Dynamics GP, Dynamics NAV, Sage 100, Blackbaud Financial Edge NXT, SYSPRO, Multiview, Elite, and PeopleSoft, with a 4-6 week finance-owned implementation and onshore FP&A-literate support. A 30-minute call starts with your actual GL and your actual chart.