Carbon accounting services vs software: what consultants do well, and what should be a system

Carbon accounting services means hiring people, a Big-4 sustainability practice, a boutique climate consultancy, or a fractional expert, to build your greenhouse gas inventory and disclosures for you. Software means licensing a system your own team (or your consultant) runs. The honest answer to "which one?" is that they solve different halves of the problem, most first-year buyers get the split wrong in a predictable direction, and the vendors on both sides have an incentive not to explain it. Full disclosure: we build carbon accounting software, it is in early access, and this post still says consultants are the right call for several things software cannot do.

What a services engagement actually delivers

A typical first-inventory engagement runs eight to twenty weeks: interviews with finance and operations, data requests, a boundary decision, a spreadsheet model mapping your spend and meters to factors, a totals deck, and, in better engagements, a methodology memo. The deliverables are real and the judgment is real. What the engagement is structurally unable to deliver is a system: the model lives in the consultant's spreadsheet dialect, keyed to this year's data layout, priced to be rebuilt next year.

Consultants are genuinely better than any software at:

• Boundary and materiality judgment. Which entities consolidate, whether category 11 applies to your product, how to treat a mid-year acquisition. These are judgment calls with disclosure consequences, and an experienced human is worth the fee.
• Regulatory interpretation. Whether your entity class is in CSRD scope this year, what your auditor's assurance provider will accept, how SB 253 phase-ins apply to your structure.
• Organizational persuasion. Getting procurement, facilities and HR to hand over data they have never been asked for. An external expert with board sponsorship moves organizations that an internal analyst with a spreadsheet cannot.
• One-off deep dives. A product life-cycle assessment, a supplier engagement pilot, a science-based target submission. Projects, not processes.

Where the engagement model breaks

• The deliverable goes stale immediately. The PDF describes last year. New invoices arrive the day after handover, and nothing posts them anywhere. By Q3 the "current footprint" is a forecast from a dead model.
• The lineage leaves with the consultant. Eighteen months later an auditor samples a reported tonne back to source. The person who knew which tab, which factor and which exclusion applied is on another client. This is the single most common way first inventories fail assurance readiness.
• The classification hours are the invoice. Most of the fee buys the mechanical middle: classifying thousands of AP lines to scopes and categories and matching factors. It is exactly the work AI now drafts well, with confidence flags for the tail that needs human judgment, which is what our demo shows on real lines.
• It reprices every year. The engagement is sold as a project, so year two costs like year one, minus a discount, forever. A system starts year two from year one's ledger.

What software delivers, and where it is oversold

Good software gives you the system of record: continuous ingestion of spend and meter data, classification with stated methods and confidence, factor and method versioning, evidence links from every figure to its source, and report outputs generated from the ledger rather than assembled by hand. That is the half of the problem that should never have been hours.

Software is oversold whenever a vendor implies it removes the need for judgment. It does not decide your consolidation boundary, negotiate with your auditor, or convince your biggest supplier to share their footprint. And enterprise platforms in this market carry their own services problem: quote-based pricing and implementation projects that look suspiciously like consulting engagements, a pattern we map honestly on carbon accounting platforms and the Persefoni comparison.

The division of labor that actually works

For a mid-market company in its first or second mandatory year:

• Software owns the record. The ledger, the classification pass, the factor library, the evidence trail, the report packs. Continuous, versioned, yours.
• A consultant owns the judgment calls, on top of the ledger. Boundary memo, materiality assessment, assurance preparation, tricky category decisions. Days of expert time instead of weeks of data wrangling, because the wrangling is already done when they arrive.
• Your team owns the review. Someone in finance or ESG reviews the low-confidence tail monthly and approves method changes. Hours per month, not a hire, once the drafting is automated.

Consultancies increasingly run this model themselves: the sharper boutiques license software, automate the classification, and sell pure judgment at better margin. That is why our launch plans include a consultant-shaped tier, described on the pricing page; the practices that adopt this split are not our competitors, they are our channel.

Cost, honestly

We will not print a "consultants cost $X" figure: engagements are quote-based and vary by an order of magnitude with scope and brand. The structural point survives without numbers: services price scales with hours, and the hours recur annually; software prices scale with the system, and the system compounds. Our planned tiers are public (Ledger $390/mo, Compliance $1,290/mo, Assurance $3,900/mo, yearly about 20% off), which at least gives your consultant comparison a fixed side to compare against. Nothing is charged during early access.

A realistic first-year timeline, both routes

Abstract comparisons hide the thing a CFO actually needs to plan: what happens in which month. Here is a typical first mandatory year on each route, for a mid-market company with a few thousand suppliers.

The services route:

• Weeks 1 to 4: scoping, engagement letter, kickoff interviews with finance, operations and facilities. Nothing is measured yet; the consultants are learning your org chart.
• Weeks 4 to 10: data requests go out. This phase slips more than any other, because the people holding utility bills, fleet records and AP exports have day jobs. Every week of slippage moves the whole timeline.
• Weeks 10 to 16: the model is built. Spend is mapped to categories, factors are matched, the boundary memo is drafted. Your team sees interim totals and answers clarification emails.
• Weeks 16 to 20: review cycles, the final deck, the methodology memo, handover. You now have a complete inventory describing a year that ended five months ago.
• Months 6 to 12: nothing. New transactions accumulate unclassified. The next engagement is scoped the following spring, and much of the first ten weeks repeats.

The software route (described for the category; our own platform is in early access and the workflow pieces below are the planned shape, labeled as such on the carbon accounting software homepage):

• Weeks 1 to 2: export AP data, utility records and fleet data; import and map entities. The bottleneck is your own data hygiene, not the vendor.
• Weeks 2 to 5: the classification pass runs, and your reviewer works the low-confidence queue. This is where the real hours go on the software route, and it is measured in days, not weeks.
• Weeks 4 to 8: a consultant is brought in for days, not months: boundary memo, materiality screening across the Scope 3 categories, method sign-off. They start from a classified ledger instead of a blank data request.
• Month 3 onward: monthly cadence. New transactions are classified as they arrive, the reviewer approves the tail, and the reported number is never more than a month stale.
• Pre-filing: report packs are generated from the ledger, and pre-assurance review samples against live lineage rather than a memory of how the spreadsheet worked.

The elapsed calendar time to a first number is not radically different. The difference is what exists on day 140: a document on one route, an operating system of record on the other.

The hidden costs on each side

Both routes carry costs that never appear in the proposal, and honesty requires listing ours too.

Hidden costs of services:

• The annual rebuild. Year two is priced as a fresh project because it substantially is one: new data layout, new analyst, re-learned context. The discount rarely reflects how much should have been reusable.
• Key-person risk. The engagement's real deliverable is in one senior consultant's head. When they change firms, your methodology's institutional memory goes with them, and your next engagement pays to reconstruct it.
• Your own team's hours. Data requests do not answer themselves. Finance teams routinely spend more internal time servicing an engagement than they would running a monthly review queue, and none of it appears on the invoice.
• Change orders. An acquisition mid-year, a new entity, a regulator's clarification: each is out of scope and re-priced, because the model was built for the scope as quoted.

Hidden costs of software:

• Review time is real time. Someone competent has to own the low-confidence queue and method approvals, monthly, indefinitely. If no one in finance or ESG can commit those hours, the ledger rots exactly like the consultant's PDF did.
• Change management. Getting AP exports flowing on a schedule means asking IT and procurement for something new. Small ask, but organizations are organizations.
• Judgment is not included. The boundary memo and the materiality call still cost consultant days. A software subscription that looks cheap becomes less so once you add the judgment days it genuinely requires; we say this plainly because pretending otherwise is how software gets oversold.
• Vendor risk. The category is young and consolidating. Ask any vendor, including us, how you export the full classified ledger with evidence links if you leave. If the answer is vague, the asset is not really yours.

What auditors actually sample

First-year buyers consistently optimize for the wrong artefact: a polished report. Assurance providers do not audit the report's typography; under limited assurance they test the process, and under reasonable assurance they test it hard. Three things dominate the sampling in practice.

• Lineage. The auditor picks a reported figure and asks for the path back to source: the invoices or meter readings behind it, the classification applied, the calculation performed. A services deliverable can pass this in year one while the consultant is still reachable; the test is whether it passes in month 18 without them.
• Factor citations. Which factor database, which edition, which specific factor, and why that one. "Standard emission factors" is not an answer. A ledger that versions its factor library answers this mechanically; a spreadsheet answers it only if someone documented every cell.
• Method consistency. If category 1 was spend-based last year and hybrid this year, the change must be deliberate, documented and, where material, restated. Silent method drift between annual engagements, each built by a different analyst, is the most common finding, and it is invisible until the auditor lines up two years. The primer on how Scope 3 is defined and measured shows why these method choices carry most of the number.

The uncomfortable symmetry: services fail these tests through discontinuity, software fails them through unreviewed automation. The passing configuration on either route is the same, a continuous record with a named human approving changes.

When services alone is genuinely right

Since we sell software, the credible version of this post has to name the cases where we would tell you not to buy any, including ours.

• Very small companies with a one-off ask. Twenty people, one office, a single customer questionnaire demanding a footprint: a competent consultant, or honestly a diligent afternoon with the basics of carbon accounting and public factors, beats a subscription you will not use monthly.
• One-off deep analyzes. A product life-cycle assessment for a design decision, a science-based target submission, a supplier engagement pilot. These are projects with an end date, and engagement pricing fits projects.
• M&A due diligence. Estimating a target's footprint and transition exposure from outside, on a deal timeline, under NDA, is expert work on incomplete data. No ledger exists to keep.
• No internal owner, truly none. If nobody in the organization can own even a monthly review, buy the finished inventory and be honest that you are renting a capability. Revisit when a mandatory regime, CSRD, SB 253 or a customer's supply-chain program, makes the recurring cost indefensible.

The common property of these cases: the output is legitimately a document, not a system. The moment the ask becomes recurring, the economics flip, which is the whole argument of this post.

Six questions to ask either vendor

• Ask a consultancy: what specifically do we own and can we rerun after you leave?
• Ask a consultancy: how many of the quoted hours are line classification, and why are those not automated?
• Ask a consultancy: will your model survive our auditor sampling a figure to source in 18 months, without you on retainer?
• Ask a software vendor: can we see it work on our data before a sales call? (Ours: yes, here.)
• Ask a software vendor: what happens to last year's number when your factor library updates?
• Ask a software vendor: which parts of your pitch are shipped versus planned? (Our answer is labeled on every page, because we are in early access and say so.)

The pattern behind all six: whoever you pay, the inventory should end up as an asset you keep, a classified, evidence-linked ledger, not a document you commission again every spring. Judge services and software alike by whether they leave that asset behind.