How to Reduce Reliance on Big 4 for Tax Compliance

How do you reduce reliance on the Big 4 for tax compliance? Take back the data, the workings and the know-how that left the building years ago, put AI to work on the labour, and decide what is still worth buying.

Ask a head of tax why the group's corporate tax returns are prepared by a Big 4 firm and the answer is rarely "because they are better at it than we would be." It is usually some version of "because they always have been." The engagement letter renews, the information request arrives, the trial balances go out, and some months later the returns come back.

I have spent a good part of my career on the other side of that arrangement, at PwC and Deloitte, and I want to be clear about what this article is not. It is not an argument that the Big 4 are poor value or that their people are anything other than very good. It is an argument that most businesses are buying the wrong thing from them. They are paying advisory rates for production work, and in the process handing over the one asset a tax function cannot afford to lose: an understanding of its own numbers.

The problem is dependency, not use. The two are worth separating. So the question I hear most often from heads of tax and CFOs, "how do I reduce reliance on the Big 4 for tax compliance?", deserves a practical answer rather than a slogan. This is my attempt at one.

There is also a reason to have this conversation now rather than in three years. The economics of outsourced compliance have always rested on labour: large teams of capable juniors extracting, mapping, reconciling and tying out. That is precisely the work artificial intelligence is getting good at. The firms know it and are investing heavily. The open question for every tax and finance leader is whether that gain is captured inside their own function or rented back to them at a margin.

Why Businesses Rely on the Big 4 for Tax Compliance

The reasons businesses lean on the Big 4 for compliance are rational, which is why the habit is hard to break.

• Headcount is harder to get than budget. A CFO will approve an adviser fee long before approving two permanent hires. Outsourcing is how tax functions grow without growing.
• It feels like risk transfer. A return with a Big 4 name behind it is easier to defend to an audit committee than one prepared by a team of three.
• The law moves faster than a small team can track. Pillar Two, UAE Corporate Tax, e-invoicing mandates and expanding disclosure rules have all arrived within a few years of each other.
• The process was built around the adviser. Once a firm has prepared the return for five years, the mappings, the adjustments, the prior-year logic and the working papers all live in its systems, not yours.

The last point is the one that turns a sensible choice into a dependency. Each year the firm knows a little more about how your ledger becomes your tax return, and you know a little less. Switching, or bringing the work in, starts to look like a rebuild from nothing. So nobody does it.

What It Actually Costs

The fee is the visible cost and usually the smallest one.

You pay to be re-learned every year. Compliance teams rotate. The senior who understood your intercompany structure last year has moved on, and part of this year's fee funds a new person asking the same questions.

The numbers arrive too late to be useful. When computation sits outside the business, tax becomes something reported after the fact. The CFO asks what the effective tax rate will be if a restructuring goes ahead, and the honest answer is "we will ask our advisers."

Nothing is reused. The provision, the return, the transfer pricing file and the Pillar Two calculation draw on largely the same data. Outsourced in separate engagements, sometimes to separate firms, that data is collected, cleansed and mapped several times over.

The risk never left. The return is signed by the company. Penalties, interest and reputational damage land on the taxpayer. What outsourcing transfers is effort and a measure of comfort, not accountability.

Pillar Two has made all of this harder to ignore. A calculation that needs a few hundred data points per constituent entity, many of them not in the ledger at all, cannot be run as an annual send-and-receive exercise. Groups that tried discovered that the bottleneck was never the adviser's technical knowledge. It was their own data.

Decide What Is Still Worth Buying

Reducing dependency does not mean ending the relationship. It means being deliberate about which side of a line each piece of work falls on. The line I use is judgement versus production.

Keep buying judgement:

• Novel or uncertain positions where a second opinion has real value
• Transactions, restructurings and disputes
• New legislation in its first year, before practice has settled
• Jurisdictions where you have one small entity and no local language capability
• Independent review of work your own team has prepared

Bring production in-house:

• Data collection, cleansing and mapping from ledger to tax
• Recurring computations in your core jurisdictions
• Tax provision and effective tax rate reporting
• Pillar Two safe harbour testing and the underlying data model
• Reconciliations, roll-forwards and return-to-provision true-ups

Notice what the first list has in common. It is work where the firm's breadth, seeing fifty groups face the same question, gives it something you cannot build internally. The second list is work where the knowledge that matters most is knowledge of your own business.

Strategy One: Own the Data Before You Buy Anything

The most common mistake is to start with a platform. Start instead with the thing the adviser has actually been doing for you each year, which is turning accounting data into tax-ready data.

Three assets need to live inside the business, documented and owned by a named person:

1. A tax-sensitised chart of accounts. If disallowable expenses, exempt income and provisions are buried in general accounts, someone has to dig them out every year. Fix it at source and the adjustment becomes a report rather than an analysis.
2. A clean entity master. Ownership percentages, tax residence, tax identification numbers, fiscal year ends, accounting standards and elections, held in one place and kept current. Pillar Two alone makes this non-negotiable.
3. The mapping from ledger to return. The logic that says which account feeds which line, with which adjustment and under which rule. This is the intellectual property you have been renting.

A practical first move, and one that costs nothing in software: at the next renewal, write into the engagement letter that the working papers, mappings and adjustment logic are deliverables, provided in an editable format. Many businesses are surprised to find they have never held their own tax workings.

Strategy Two: Choose Platforms You Can Run Yourselves

Once the data is under control, technology is what lets a small team do the production work. The market is crowded, and every vendor demonstration looks impressive. These are the questions I would put to any platform before signing.

• Does it sit on what we already own? If the group consolidates in Oracle EPM, OneStream or SAP, a tax solution on the same platform inherits the entity hierarchy, the data and the controls. A standalone tool needs integrations that someone has to maintain.
• Can every figure be traced? For any number on a return, you should be able to click back to the ledger balance, the adjustment and the rule that produced it. If the platform cannot show its working, your team cannot defend it under enquiry.
• Who maintains the rules when the law changes? Either the vendor ships content updates on a committed timeline, or your team maintains the logic. Both can work. Not knowing which one you have bought does not.
• Can our own people administer it? If every new entity, account or rate change needs a consultant, you have replaced one dependency with another.
• Can we get our data and logic out? Exportable data, documented calculations and no proprietary black box. The exit should be as clear as the entry.
• What does it cost over five years, not one? Licences, implementation, internal administration, annual updates and the inevitable second phase.

A word of disclosure belongs here. I work for a consulting firm that implements these platforms, so I have an interest in the answer. That is exactly why I would press the fourth question hardest. An implementation that leaves the client unable to operate its own system has failed, however well it was delivered. Reducing reliance on the Big 4 only to become reliant on a software vendor or a systems integrator is not progress.

Strategy Three: Put AI to Work on the Labour, Not the Numbers

For most of the last two decades, the honest objection to in-house compliance was arithmetic. A team of four cannot do what an adviser does with twenty. AI is what changes that arithmetic, and it is why this is a different conversation from the one tax leaders were having five years ago.

It helps to be specific about the jobs, because "use AI" is not a strategy.

• Reading documents. Subsidiary financial statements, invoices, contracts, assessments and notices from tax authorities. Extracting the figures and terms that matter from unstructured paper is the single largest block of junior time in any compliance cycle, and models now do it well.
• Proposing mappings. A new entity arrives with a local chart of accounts in another language. A model can propose how each account maps to the group's tax-sensitised structure, with its reasoning, for a person to accept or correct.
• Explaining movements. Why did the effective tax rate in one jurisdiction move by three points? A model with access to the underlying data can draft the variance explanation before the reviewer asks for it.
• Watching the law. Monitoring authority publications, summarising what changed and flagging which entities or positions are affected. This is the part of the "we cannot keep up" objection that has weakened the most.
• Drafting. First versions of technical memos, disclosure notes, transfer pricing narratives and responses to authority queries, written from your own facts rather than a template.
• Remembering. An assistant grounded in your own policies, prior-year workings and past advice can answer "how did we treat this last year, and why?" in seconds. That directly attacks the cost of being re-learned every year. The institutional memory stops walking out of the door.

Then there is the line that should not be crossed. A language model should not be the thing that produces a number on a return. It is a probabilistic system, and a tax computation has to be reproduced exactly, line by line, in front of an authority several years later. The design that works is a division of labour: AI prepares, proposes and drafts; a deterministic engine with explicit rules computes; a qualified person reviews and signs. If a vendor cannot tell you which of those three is responsible for a given figure, keep asking.

A few conditions make this safe to run inside a tax function:

• Use enterprise tools, not public ones. Tax data belongs in a tenant with contractual confidentiality and no training on your inputs.
• Demand sources. Every extracted figure should link back to the page it came from. Every summary of the law should cite the provision.
• Keep a human in the loop and a log. Who accepted which mapping, when, and on what basis. This is ordinary control design applied to a new kind of preparer.
• Fix the data first. AI makes good data faster to work with. It does not make bad data good, and it will explain a wrong number as fluently as a right one.

It is worth being realistic about scale. AI will not let three people do the work of thirty. In my view it lets three people do the work of six or seven, and do the more interesting half of it. For most groups, that is exactly the gap between needing to outsource production and being able to keep it.

One last point. Your advisers are using the same technology. If their preparation effort falls and your fee does not, you are funding their transformation instead of your own. That is a reasonable thing to raise at the next renewal.

Strategy Four: Build the Team That Can Hold It

Technology without capability simply moves the dependency. The in-house model needs three kinds of skill, which in a small function may sit in two people.

• Technical tax depth in the jurisdictions that matter most to the group. Not everywhere. The three or four that drive most of the tax charge.
• Tax data and systems fluency. Someone who understands how a number travels from a subledger to a disclosure note, and who can configure the platform rather than raise a ticket about it.
• Process ownership. Calendars, controls, reviews and evidence. Unglamorous, and the reason in-house compliance survives a staff change.

How to build it without a hiring spree:

• Make knowledge transfer a contractual deliverable. Your team sits alongside the adviser's during preparation, not just at sign-off.
• Hire one hybrid. A single person who is both tax-qualified and comfortable with systems changes what the rest of the team believes is possible.
• Invest in structured learning. Tax technology qualifications now exist alongside the traditional technical ones, and platform vendors run certification paths. Fund them, and give people real work to apply them to.
• Make AI fluency part of the job, not a side project. Everyone on the team should know how to brief a model, check its output against source and recognise when it is confidently wrong. That is a reviewing skill, and tax people already have it.
• Let the team build something small. A single automated reconciliation, a document-extraction workflow or a self-service ETR report teaches more than a week of training.
• Create a career path. People leave in-house tax for the Big 4 because that is where the progression is. If the function becomes more interesting, it also becomes easier to staff.

Flip the Model, Then Shrink It

The transition I would recommend is gradual and reversible.

1. Pick one scope. One process or one jurisdiction where the data is cleanest and the team is strongest. The home-country return or the group tax provision is the usual candidate.
2. Reverse the roles. For a cycle, your team prepares and the firm reviews. The fee falls, the assurance stays, and your people learn under supervision.
3. Narrow the review. In the second cycle, limit external review to the areas of real judgement and let the routine pass through internal controls.
4. Extend. Add the next jurisdiction or process, reusing the data model and the platform already in place.
5. Renegotiate the relationship. What remains is an advisory retainer and a short list of jurisdictions that stay outsourced on purpose.

Expect this to take two to three compliance cycles, not one. And expect the first cycle to feel slower and more expensive than what it replaced. That is the cost of learning what you had been paying someone else to know.

What This Looks Like in Practice

To make this concrete, here is an illustrative example. It is a composite of the kind of group I work with, not a single client, and the figures are indicative.

Picture a Gulf-headquartered group with around forty entities in nine countries and a tax team of four. For years, every return and the year-end provision were prepared by a Big 4 firm. When UAE Corporate Tax and then Pillar Two arrived, the group's first instinct was to add two more engagements with the same firm. The quotes came back, and the head of tax noticed that all three pieces of work began with the same request: send us your trial balances and your entity list.

Instead, the group spent the first six months on the data. It rebuilt the chart of accounts so that tax-sensitive items were visible at source, created a single entity master, and asked the firm for its working papers and mappings as a deliverable. It then put the provision and the UAE computations onto the consolidation platform finance already ran, so no new system had to be bought.

In the first cycle, the in-house team prepared the UAE returns and the group provision, and the firm reviewed them. AI was used in two narrow places: extracting figures from the financial statements of the smaller overseas subsidiaries, and proposing account mappings for entities joining the model. Every extraction was checked against source by a person. No figure on a return came from a model.

By the second cycle, external review had narrowed to three judgement areas, and four of the nine jurisdictions had moved in-house. Five stayed outsourced on purpose, each a single small entity in a country where the group had no local expertise. Compliance fees fell by roughly a third. Advisory spend with the same firm went up, because the team now had the time and the understanding to ask better questions. And the provision closed in days rather than weeks.

The fee saving was not what the CFO noticed. What the CFO noticed was that a question about the tax effect of a planned acquisition got an answer the same afternoon.

How to Tell It Is Working

A few measures are worth tracking from the start:

• The share of returns and computations prepared in-house
• The split of external spend between advisory and compliance
• Days from period close to a tax number the CFO can use
• The number of manual adjustments in the provision, year on year
• Whether the team can answer a what-if question without picking up the phone

The last one matters most. The point of all this is not to save fees, although it usually does. It is to have a tax function that understands its own position well enough to advise the business, and that calls the Big 4 for the questions that are worth their rates.

That is a better relationship for both sides. The best advisers I worked with never wanted to be a compliance factory. They wanted the interesting call. Give them that, and keep the rest.